Ps Adaptiv Moving Average Signalet

Tradisjonell trefasade for ditt hjem vil alltid være det foretrukne valget i en hvilken som helst sidesporingsinstallasjon. Men du trenger den rette siding entreprenøren for å hjelpe deg å velge den beste tre siding med minst mulig vedlikehold av sidespor. Det er mange forskjellige typer trefasade skaper laget av sub-par tre materialer. Don8217t la din nexthellip Les mer Profesjonell taktekking og takreparasjon til rimelige priser Hvis du har taklekkproblemer, må du ikke kontakte en taktekkingskontraktør. Langvarig taklekkasje virker ufarlig i begynnelsen, men kan forvandle seg til alvorlige problemer raskt. Å ha vedvarende taklekkasje kan forårsake muggvekst, skadet tak, møbler. Ring oss i dag hvis you8217re trenger shellip Les mer Når du prøver å redusere home8217s varmekjølingsregning, er det ikke noe bedre å spare penger enn å ha nye vinyl - eller treutskiftningsvinduer installert. Fremfor alt, med den nye føderale skatterabatten på opptil 1500 der8217er ingen grunn til ikke. Å ha en vakker dekk vil alltid ha en positiv avkastning på eiendommens verdi. Å legge et dekk er en billig måte å utvide ditt boareal på. Den gjennomsnittlige kostnaden for å bygge en dekk er omtrent 7000 og gir en avkastning på ca 15.000 ved salg av hjemmet ditt 8211 Ikke til dårlig, rett Så vær så snill å ha ahellip Les mer Velge riktig sidespredningsentreprenør er avgjørende for enhver sidesporingsinstallasjon. Enten det er å installere vinylsider over sidespor eller helt fjerne din eksisterende sidespor for ny sidespor. Å ha en profesjonell siding entreprenør som kan gi deg den beste løsningen for den jevneste siding installasjon vil spare deg for mye hodepine, tid andhellip Les mer Hva våre kunder sier Veldig fornøyd Jeg ville bare uttrykke hvor fornøyd var med vårt nye tak og sømløse takrenner. Mike og hans arbeidere er veldig hyggelige og godt mannered å være rundt. Jeg kunne ikke tro hvor fort de fullførte taktekking vårt hjem og garasje. De forlot stedet renere enn da de startet og beskyttet våre busker, planter som lovet. Var så fornøyd med takinstallasjonen, at vi vil ha dem tilbake til vinduesskift. Takk Mike Se deres hjem: taktekking Melrose MA mdash Robert Patricia Quinn, Melrose, MA MBM Construction er vurdert 5 5 basert på 3 vurderinger. Å finne den rette entreprenøren bør ikke være smertefullt Å gjøre det rette hjemmeforbedrings - eller ombyggingsprosjektet kan legge til ekte verdi for alle typer hjem, hvis det gjøres riktig og effektivt av en lisensiert og forsikret profesjonell. Bruke materialer av høy kvalitet som er energieffektive, tiltalende og viktigst, pålitelige som lavt vedlikeholdsvindu. vindu utskiftninger. Shingle tak og tilpassede dekk vil legge til ekte verdi. I de fleste tilfeller kan du forvente og umiddelbar avkastning på investeringen etter at du har gjort disse forbedringer i hjemmet. Velg en entreprenør som vil få jobben, gå gjennom hvert trinn i prosjektet fra start til slutt uten noen skjulte tillegg. Våre hjem forbedring tjenester har gitt oss en fordel over andre hjem remodeling selskaper. Å være en av områdene toppleverandører i sidespor, taktekking, utskiftingsvinduer og romtilsetninger. gir oss stor kjøpekraft gjennom våre leverandører og i retur kan passere langs besparelsene til deg Så hvorfor velge oss som din boligbygging entreprenør Vi lytter til dine behov. Vi bruker ikke høytrykkssalg, eller prøver å selge deg noe du ikke trenger eller vil ha. Kommunikasjon er nøkkelen til alle typer remodeling-prosjekter, og vi vil at prosjektet ditt skal få best mulig utfall. Du får et detaljert prosjekt beregnet sluttidspunkt - uten hodepine Du vil også få en av de beste garantiene i bedriftsforbedringsbransjen, bør du noen gang trenge å bruke den. Hva slags hjem forbedringstjenester du ser etter Velg en taktekking entreprenør som vil jobbe i din beste interesse og ikke i hvor mye fortjeneste han kan gjøre ved å kutte hjørner. Som takarbeidere tror vi på å bruke de beste takbjelkene og underlaget for å gi våre kunder ro i sinnet. For mer info om taktekkingstjenester, besøk: Commercial Flat Roofing eller Residential taktekking Ikke alle vinyl og tre sidespor er de samme. Velg en profesjonell siding entreprenør som vil hjelpe deg å forstå hva slags vinyl sidespor for å unngå, og hva vil gi deg det beste bang for pengene dine på lang sikt. Fra en standard trykkbehandlet dekk, mahagony dekk, eller en kompositt dekk - vi har fått deg dekket. Les mer om Decks and Porches Deck buildersSteering Behavior For Autonome Karakterer Craig W. Reynolds Sony Computer Entertainment Amerika 919 East Hillsdale Boulevard Foster City, California 94404 Nøkkelord. animasjonsteknikker, virtuelle interaktive miljøer, spill, simulering, adferds animasjon, autonom agent, plassert, belagt, reaktiv, kjøretøy, styring, sti planlegging, sti følge, jakten, unnvikelse, hindring unngåelse, kollisjon unngås, flocking, gruppe oppførsel, navigasjon, kunstig liv, improvisasjon. Dette papiret presenterer løsninger for ett krav til autonome tegn i animasjon og spill: evnen til å navigere rundt i verden på en livlig og improvisasjonsmessig måte. Disse 147steering-atferdene148 er i stor grad uavhengige av detaljene til karakteren av karosserien. Kombinasjoner av styringsadferd kan brukes til å oppnå høyere nivåmål (For eksempel: Hent fra her til der, samtidig som du unngår hindringer, følg denne korridoren, delta i denne gruppen av tegn.) Dette papiret deler bevegelsesadferd i tre nivåer. Det vil fokusere på midtnivået på styringsadferd, kort beskriv det lavere nivået av lokomotiv, og berør lett på det høyere nivået av målinnstilling og strategi. Innledning Autonome tegn er en type autonom agent beregnet for bruk i datamaskinanimasjon og interaktive medier som spill og virtuell virkelighet. Disse agenter representerer et tegn i en historie eller et spill og har noen evne til å improvisere sine handlinger. Dette står i kontrast både til et tegn i en animert film, hvis handlinger er skrevet på forhånd, og til en 147avatar148 i et spill eller en virtuell virkelighet, hvis handlinger er rettet i sanntid av en menneskelig spiller eller deltaker. I spill blir autonome tegn noen ganger kalt non-player tegn. En autonom karakter må kombinere aspekter av en autonom robot med noen ferdigheter av en menneskelig skuespiller i improvisasjonskunst. Disse tegnene er vanligvis ikke ekte roboter, og er absolutt ikke menneskelige skuespillere, men deler noen egenskaper av hver. Begrepet 147autonom agent148 brukes i mange sammenhenger, så det følgende er et forsøk på å finne terminologien til dette papiret i forhold til andre fagområder. En selvstendig agent kan eksistere isolert, eller den kan ligge i en verden som deles av andre enheter. En 147data mining148 agent er et eksempel på den tidligere, og en kontroller for et strømnettet er et eksempel på sistnevnte. Et lokalisert agent kan være reaktivt (instinktivt, drevet av stimulans) eller det kan være deliberativt (147intellectual148 i den klassiske AI-sans). En autonom agent kan kun forholde seg til abstrakt informasjon (147softbot148, 147knowbot148 eller 147information agent148), eller den kan bli gjengitt i en fysisk manifestasjon (en typisk industrirobot eller et autonomt kjøretøy). Kombinasjoner av liggende, reaktive og legemliggjorte definerer flere forskjellige klasser av autonome midler. Kategorien av belagte, formidlede agenter antyder vanligvis autonome roboter: mekaniske enheter som eksisterer i den virkelige verden. Noen ganger blir roboter studert via beregningsmessig simulering. Men den praktikken er sett med mistanke av purister i robotteknologifeltet fordi simuleringen kan avvike fra virkeligheten på uforutsigbare måter. Det er en annen klasse av lokalisert, legemliggjort agent basert på en beregningsmodell. Dette papiret vil benytte begrepet virtuelt (som i den virtuelle virkeligheten) for å betegne disse agenter som, i stedet for å være simuleringer av en mekanisk enhet i den virkelige verden, er i stedet virkelige agenter i en virtuell verden. (Analog til en fysisk basert modell i datamaskinanimasjon.) Derfor er de autonome tegnene til denne papirtypen tittelen: Liggende, legemliggjorte, reaktive, virtuelle agenter. Betegnelsen oppførsel har mange betydninger. Det kan bety kompleks handling av et menneske eller annet dyr basert på volisjon eller instinkt. Det kan bety de i stor grad forutsigbare handlinger av et enkelt mekanisk system, eller den komplekse virkningen av et kaotisk system. I virtuelle virkelighets - og multimedieapplikasjoner brukes det noen ganger som et synonym for 147animasjon.148 I dette papiret benyttes begrepet atferd for å referere til improvisasjons - og livsaktige handlinger av en autonom karakter. Oppførelsen av en autonom karakter kan forstås bedre ved å dele den i flere lag. Disse lagene er kun ment for klarhet og spesifisitet i diskusjonen som vil følge. Figur 1 viser en oppdeling av bevegelsesadferd for autonome tegn i et hierarki av tre lag: handlingsvalg. styring. og lokomotiv. Sikkert andre disseksjoner er mulige. Et lignende trelags hierarki er beskrevet av Blumberg og Galyean Blumberg 95, de kaller lagene: motivasjon. oppgave. og motor. Vær oppmerksom på at mens det atferdshierarkiet som presenteres her, er ment å være meget aktuelt for bevegelsesadferd, er det ikke godt egnet for andre typer autonome handlinger, for eksempel er konversjonsadferdene til en 147chatterbot148 en betydelig forskjellig struktur. Figur 1: Et hierarki av bevegelsesoppførsel For eksempel tenker noen cowboys på en storfe storfe på rekkevidde. En ku vandrer bort fra flokken. Stien sjefen forteller en cowboy å hente den svarte. Cowboyen sier 147giddy-up148 til hesten sin og leder den til kua, muligens unngår hindringer underveis. I dette eksemplet representerer sti-sjefen handlingsvalg: merker at verdens tilstand har forandret seg (en ku forlot flokken) og satt et mål (hent strøket). Styringsnivået er representert av cowboyen, som dekomponerer målet i en serie enkle delmål (nærme kua, unngå hindringer, hente kua). En subgoal tilsvarer en styreadferd for cowboy-og-hesteteamet. Ved hjelp av forskjellige kontrollsignaler (vokalkommandoer, sporer, tøyler) styrer cowboy hesten mot målet. Generelt sett uttrykker disse signalene konsepter som: gå raskere, gå sakte, ta til høyre, sving til venstre, og så videre. Hesten implementerer lokomotivnivået. Ved å bruke cowboy146s kontrollsignaler som inngang, beveger hesten seg i den angitte retningen. Denne bevegelsen er resultatet av et komplekst samspill av hestens visuelle oppfatning, dens sans for balanse og dets muskler som bruker dreiemomenter til leddene i skjelettet. Fra et teknisk synspunkt, legged lokomotion er et svært vanskelig problem Raibert 91, Hodgins 95, men verken cowboy eller hesten gir det en annen tanke. Dette papiret vil fokusere på styring, midtlag av adferdshierarkiet. Det vil kort beskrive en enkel modell av lokomotivet, men bare i nok detalj for å gi et konkret grunnlag for diskusjonen av ulike styringsadferd. Det vil bli noen kort diskusjon om handlingsvalg, men først og fremst i sammenheng med kombinering og blanding av grunnleggende styringsoppførsel. Banevurdering er et emne relatert til, men skilt fra, emnet i dette papiret. Banevurderingsalgoritmer som A og Dijkstras opererer på nettverk (ofte representerer grids) og i hovedsak løser labyrinter. En slik løsning kan tjene som en spesifikasjon for styringsteknikkene som er beskrevet i dette papiret. En analogi kan være å sammenligne de skriftlige kjøreinstruksene for å komme fra ett sted til et annet med handlingen med å kjøre bilen langs den ruten. For et utmerket overblikk over baneoppdagelse, se Reese 99. For å forstå drivkraften til dette arbeidet, bør det bemerkes at styringsatferdene som diskuteres her, vedrører 147fast148 bevegelse: kjører mot kryptering. Dette er en uformell oppfatning, men er ment å antyde at den typiske hastigheten til et tegn er stor i forhold til sin maksimale svingakselerasjon. Som et resultat må styringsadferdene forutse fremtiden, og ta hensyn til eventuelle konsekvenser av dagens tiltak. Beslektet arbeidsstyreadferd for autonome tegn trekker på en lang historie med relatert forskning på andre felt. Autonome maskiner, servomekanismer og kontrollteori har sine røtter på 1940-tallet, som beskrevet i Norbert Wiener146s 1948, bok Cybernetics, eller Control and Communication i Animal and the Machine Wiener 48. Begrepet cybernetikk kom fra et gresk ord som betyr styrmann. Under den sena 40-tallet utviklet nevrofysiologen Gray Walter autonome robotskildpadder Walter 50, som inneholdt flere av styringsadferdene beskrevet her og var blant de første maskinene som viste fremvoksende livslignende oppførsel. I begynnelsen av 1980-tallet ekstrautstyrte Valentino Braitenberg Walter146s prototyper til tankeeksperimenter om en rekke fantasifulle oppgaver148 med gradvis mer komplisert oppførsel Braitenberg 84. David Zeltzer begynte å anvende teknikker og modeller fra kunstig intelligens til animasjonsapplikasjoner Zeltzer 83. Og i 1987 skapte jeg en animert adferdsmodell av fugleflokker ved hjelp av teknikker som er nært relatert til de som presenteres i denne artikkelen Reynolds 87. Listen nedenfor med tilhørende forskning er delt inn i tre generelle kategorier: robotteknologi, kunstig intelligens og kunstig liv, selv om skillet i noen tilfeller er noe vilkårlig. Vanligvis er disse verkene orientert mot animasjon i en viss grad: de ligger i overlapp mellom animasjon (eller spill, VR og multimedia) og disse tre andre feltene. Arbeid knyttet til robotteknologi. Rodney Brooks populariserte den så radikale oppfatningen av å bygge reaktive kontroller for robotic systemer Brooks 85. Mens opprinnelig inspirert av etologisk (dyremessig oppførsel) forskning, har Ron Arkin Arkin 87, 89, 92 arbeidet fokusert på anvendelse av styringsadferd i mobile robotter . Arkin146s forskning har parallelt mye av arbeidet som presenteres i dette papiret, men hans skjema (perception155action mappings) er uttrykt i form av potensielle feltmodeller i motsetning til den prosedyriske tilnærmingen beskrevet her. I noen tilfeller er dette en forskjell uten forskjell, men i andre tilfeller (som hindring av hindring) fører det til betydelig forskjellig agenteradferd. Marc Raibert og Jessica Hodgins begynte begge i legged robotikkforskning, og begge jobber nå med animasjonsapplikasjoner av fysisk realistiske legged-systemer. I begge tilfeller har deres arbeid berørt styrings - og baneplaneringsaspekter av disse systemene Raibert 91, 91b, Hodgins 95. Arbeid av Zapata et al. på styrekontrollere for raske mobile roboter fokusert på strategier som måtte håndtere momentum og andre aspekter ved rask mekanisk bevegelse Zapata 1992. Maja Mataric har jobbet mye i den kollektive roboten Mataric 93, og et sentralt tema i dette arbeidet er styring. Arbeid knyttet til kunstig intelligens. Ken Kahn opprettet et tidlig system som genererte animasjon av tegnbevegelse fra historiebeskrivelser Kahn 79. David Zeltzer Zeltzer 83, 90 pionerer AI-basert animasjon, populariserer ideen om abstrakt 147task level148 spesifikasjon av bevegelse. Gary Ridsdale Ridsdale 87 skapte tegn som kan improvisere komplisert bevegelse, komme fra A til B mens du unngår statiske hindringer og andre skuespillere. Steve Strassmann146s Desktop Theatre-arbeid Strassmann 1991 utvidet disse forestillingene til å omfatte håndtering av rekvisitter og følelsesmessig skildring. Mocircnica Costa146s agentbasert atferdsarbeideringsarbeid Costa 90 tillater et tegn å navigere rundt i et hus, samtidig som det hindrer hindringer. Forskning på improvisasjonsrike, dramatiske tegn, som berører styringsadferd, foregår på Project Oz (og nå Zoesis) av Joseph Bates et al. Bates 92 og på The Virtual Theatre Project av Barbara Hayes-Roth et al. Hayes-Roth 96. Arbeid knyttet til kunstig liv (og andre felt). 1987-boidemodellen av flokker, besetninger, skoler og tilhørende gruppebevegelse Reynolds 87, dekomponert denne komplekse gruppen atferd til tre enkle styreadferd på individnivå. Følgende års relaterte styringsoppførsel for hindring av hindringer Reynolds 88 ble presentert. På 1987 Kunstig Livsverksted Mitchel Resnick presenterte arbeidet med autonome kjøretøy implementert i LEGO LOGO Resnick 89 og Michael Travers demonstrerte sitt AGAR Animal Construction Kit Travers 89. (Se også nyere arbeid av disse forfatterne Resnick 93 og Travers 94.) Styringsadferdene var et nøkkelelement i The Virtual Fishtank, en multiuser VR-installasjon på The Computer Museum opprettet av lag fra MIT146s Media Lab og NearLife Resnick 98. Armin Bruderlin prosessoralt generert målrettet animasjon av menneskelig gang Bruderlin 1989. Randall Beer146s avhandling på en kunstig kakerlakk Øl 90 er bemerkelsesverdig for dybden og kompleksiteten til sin nevroetologiske modell. Sentralt i denne modellen er nevrale gjennomføring av flere troper (som kjemotaksier og thigmotaxier) som er direkte analoger av styringsadferd som beskrevet nedenfor. I Wilhelms 90 undersøker Jane Wilhelms og Robert Skinner arkitekturer for kjøretøylignende tegn. Thalmann et al. skapte atferds animasjon tegn som navigerte ned korridorer og rundt hindringer bruker visjon som simulert med 3D-rendering Thalmann 90. Michiel van de Panne opprettet kontrollører for oppgaver som parallell parkering av en bil ved hjelp av state-space søk van de Panne 90. G. Keith Still har modellerte store menneskelige folkemengder ved hjelp av en modell av styreadferdigheten til hver enkelt Still 94. Ved å bruke en modifisert genetisk algoritme utviklet Karl Sims samtidig hjerner og organer for kunstige skapninger for ulike former for lokomotiv og for mål som søker Sims 94. I arbeidet ble det først rapportert til SAB94 og oppdatert på SAB96 Cliff 96, Cliff og Miller coevolved jakten og unnvikende atferd for rovdyr og byttedyr. Xiaoyuan Tu et al. utviklet en forseggjort og slående realistisk modell av biomekanikk, lokomotiv, oppfatning og oppførsel av fisk i Tu 94, 96 som inkluderte fysisk basert lokomotiv, styringsadferd og et etologisk basert system for handlingsvalg. I Blumberg 94 beskriver Bruce Blumberg en detaljert mekanisme for komplekse handlingvalg og med Tinsley Galyean i Blumberg 95 diskutert designet for et VR-tegn som kan både autonom improvisasjon og respons på ekstern retning. En applikasjon av disse tegnene var i ALIVE-systemet Maes 95 av Patties Maes et al. Improv-systemet av Ken Perlin og Athomas Goldberg Perlin 96 dekker også spekteret fra lokomotiv til handlingseleksjon, men bruker en unik tilnærming basert på atferdsskripting og Perlin146s 1985 prosessorisk syntese av teksturer Perlin 85 anvendt til bevegelse. James Cremer og kolleger har skapt autonome drivere til å fungere som 147extras148 å skape omgivende trafikk i interaktive bilkjøringssimulatorer Cremer 96. Robin Green (fra BullfrogEA) har utviklet et modent system for autonome tegn brukt i Dungeon Keeper 2 som delvis ble inspirert av en tidlig inspirasjon utkast til dette papiret. Dave Pottinger har gitt en detaljert diskusjon om styring og koordinering for grupper av tegn i spill Pottinger 1999. Locomotion Locomotion er bunnen av det tre nivået atferdshierarkiet beskrevet ovenfor. Lokasjonslaget representerer en utførelsesform for karakter 146. Den konverterer styresignaler fra styrelaget til bevegelse av tegnet146s 147body.148 Denne bevegelsen er gjenstand for begrensninger som er påkrevet av kroppens fysisk baserte modell, som for eksempel vekselvirkningen av momentum og styrke (begrensning av krefter som kan påføres av kroppen ). Som beskrevet ovenfor kan en cowboy146 hest betraktes som et eksempel på lokomotiv laget. Rider146s styringsbeslutninger overføres via enkle kontrollsignaler til hesten som konverterer dem til bevegelse. Poenget med å lage det abstrakte skillet mellom styring og lokomotiv er å forutse 147plugging in148 en ny lokomotivmodul. Tenk deg å løfte rytteren av hesten og plassere ham på en langrennsmotor. Målvalg og styringsadferd er forblir de samme. Alt som har endret seg er mekanismen for kartlegging av kontrollsignalene (gå raskere, sving til høyre.) I bevegelse. Opprinnelig involvert legged lokomotion (balanse, bein, muskler) og nå innebærer det hjulbremsning (motor, hjul, bremser). Rytterens rolle er uendret. Dette antyder at med en hensiktsmessig konvensjon for å kommunisere kontrollsignaler, kan styringsadferdene være helt uavhengig av det spesifikke lokomotivsystemet. Selv om det i praksis er nødvendig å kompensere for 147agilitet148 og forskjellige 147håndteringsegenskaper148 av individuelle lokomotivsystemer. Dette kan gjøres ved å justere stemmeparametere for et gitt lokomotivsystem (som er tilnærmingen tatt i styringsadferdene beskrevet nedenfor) eller ved hjelp av en adaptiv, selvkalibrerende teknikk (måten en menneskelig sjåfør raskt tilpasser seg kjennskapene til en ukjent bil). I det første tilfellet kan styringsadferdene avgjøre ved å forhåndsinnstille at karakterhastigheten i en gitt situasjon bør være 23 km / t, i andre tilfelle kan det si at 147 sluk litt148 til det samme resultatet ble oppnådd. Bevegelsen av en autonom karakter kan være basert på, eller uavhengig av, den animerte skildringen. Et tegn kan representeres av en fysisk basert dynamisk balansert simulering av turgåing, som gir både realistisk animasjon og atferdsmessig bevegelse. Eller et tegn kan ha en veldig enkel lokomotivmodell (som beskrevet i neste avsnitt) som en statisk (si et romskip) eller pre-animert (som en menneskelig figur utfører en turcykel), er vedlagt. En hybrid tilnærming er å bruke en enkel lokomotivmodell og en adaptiv animasjonsmodell, som en inverskinematisk drevet turcykel, for å bygge bro mellom gapet mellom abstrakt lokomotiv og betong terreng. Endelig kan lokasjonen begrenses til bevegelsen som er innebygd i et fast sett pre-animerte segmenter (gå, kjøre, stoppe, sving til venstre.) Som enten velges diskret eller blandes sammen. En enkel kjøretøymodell Tilnærmingen tatt i dette papiret er å vurdere styringsadferd som vesentlig uavhengig av det underliggende lokomotivsystemet. En enkel lokomotivmodell vil bli presentert for å gjøre diskusjonen om styringsadferdene mer konkret. Denne fremdriftsmodellen vil være basert på et enkelt idealisert kjøretøy. Valget av begrepet 147vehicle148 er inspirert til en viss grad av Braitenberg 84. Det er ment å omfatte et bredt spekter av transportmidler, fra hjulutstyr til hester, fra fly til ubåter, og (mens det sikkert strekker terminologien) for å inkludere lokasjon av en character146s egne ben. Bilmodellen beskrevet her er så forenklet og generisk at det er en like god (eller like dårlig) tilnærming til alle disse. Denne kjøretøymodellen er basert på en punktmassens tilnærming. På den ene siden tillater det en veldig enkel og beregningsmessig billig fysisk basert modell (for eksempel har en punktmasse hastighet (lineært momentum), men ikke et moment av treghet (rotasjonsmoment)). På den annen side kan det ikke være en veldig overbevisende fysisk modell fordi punktmasser ikke eksisterer i den virkelige verden. Ethvert fysisk objekt med masse må ha en null-radius og dermed et moment av treghet. Denne bruken av en forenklet, ikke-fysisk kjøretøymodell er bare for enkelhets skyld og er beregnet på å være 147 uten tap av generalitet148 151 det bør alltid være mulig å erstatte en mer troverdig, mer realistisk fysisk basert kjøretøymodell. En punktmasse er definert av en posisjonsegenskap og en masseegenskap. I tillegg inkluderer den enkle kjøretøymodellen en hastighetsegenskap. Hastigheten er modifisert ved å påføre krefter. Fordi dette er et kjøretøy, er disse kreftene vanligvis selvpåførte, og dermed begrenset. For eksempel er en typisk kraft som justerer en hastighet for en kjøretøy 146 trykk. generert av kjøretøyets egen kraftverk, og dermed begrenset av kapasiteten til kraftverket. For den enkle kjøretøymodellen er denne oppfatningen oppsummert av en enkelt 147maximum force148 parameter (maxforce). De fleste kjøretøy er preget av topphastighet. Typisk er denne begrensningen på grunn av samspillet mellom akselerasjon på grunn av deres endelige trykk og decelerasjonen på grunn av viskøs drag, friksjon eller (i leggede systemer) fremdriften av frem - og tilbakegående deler. Som et alternativ til realistisk simulering av alle disse begrensningskrefter, inkluderer den enkle kjøretøymodellen en 147maximum speed148 parameter (maxhastighet). Denne hastighetsgrensen blir håndhevet ved en kinematisk trunking av vehikel 14s hastighetsvektor. Endelig inneholder den enkle kjøretøymodellen en orientering. som sammen med kjøretøyets 146 posisjon danner et hastighetsjustert lokalt koordinatrom som en geometrisk modell av kjøretøyet kan festes til. (Begrepet lokaliseres og globaliseres vil bli brukt i dette dokumentet for å angi transformere vektorer inn i og ut av dette lokale rommet.) For en 3D-kjøretøymodell har posisjons - og hastighetsvektorverdiene tre komponenter og orienteringsverdien er et sett av tre vektorer (eller en 3x3-matrise eller en kvaternion). For et 2D-kjøretøy har vektorene hver to komponenter, og orienteringsverdien er to 2D-basisvektorer eller kan representeres som en enkelt skalarhøyvinkel. Fysikken til den enkle kjøretøymodellen er basert på videre Euler-integrasjon. Ved hvert simuleringstrinn blir adfærdsbestemte styringskrefter (som begrenset av maksforce) påført på bilens 147 punktmasse. Dette gir en akselerasjon lik styrestyrken dividert med kjøretøyets masse. Denne akselerasjonen legges til den gamle hastigheten for å produsere en ny hastighet, som deretter avkortes av maxhastigheten. Endelig legges hastigheten til den gamle posisjonen: Den enkle kjøretøymodellen opprettholder sitt hastighetsjusterte lokale rom ved trinnvis justering fra forrige tidstrinn. Det lokale koordinatsystemet er definert i form av fire vektorer: en positionsvektor som spesifiserer lokal opprinnelse og tre retningsvektorer som tjener som basisvektorer i rommet. Basisvektorene angir retningen og lengden av koordinatenehetene i hver av tre innbyrdes vinkelrette retninger i forhold til kjøretøyet. Disse aksene vil bli referert til her som fremover. opp. og side. (Disse korresponderer selvsagt til X, Y og Z akser på R 3. Men noen mennesker tenker opp er selvsagt Y, mens noen synes det er åpenbart Z. De beskrivende termer vil bli brukt i stedet for de kartesiske navnene for klarhet.) For å forbli innrettet med hastighet ved hvert trinn, må basisvektorene roteres til en ny retning. (Hvis hastigheten er null beholdes den gamle retningen.) I stedet for å bruke eksplisitte rotasjoner, rekonstrueres det lokale rommet ved hjelp av en kombinasjon av substitusjon, tilnærming og reorthogonalisering. Vi starter med den nye hastigheten og en tilnærming til den nye opp retningen. For eksempel kan den gamle oppretningen brukes som en tilnærming til den nye oppingen. Vi bruker vektorkorsproduktoperasjonen for å konstruere de nye basisvektorer: Den grunnleggende ideen er at den omtrentlige oppstillingen er nesten vinkelrett på den nye fremadrettede retningen, fordi ramme-til-rammeendringer i orientering er vanligvis små. Den nye sidelinjen vil være vinkelrett på ny fremover. fra definisjonen av kryssprodukt. Den nye opp er kryssproduktet av vinkelrett fremover og side og så er vinkelrett på hver. Konseptet med 147velocity alignment148 spesifiserer ikke en orientering unikt. Graden av frihet som tilsvarer rotasjon rundt foraksen (også kjent som rull) forblir ubegrenset. Konstruksjon av det nye lokale rommet i forhold til det forrige (ved for eksempel å bruke den gamle oppretningen som den første tilnærmingen til den nye) vil sikre at rulleorienteringen i det minste forblir konsistent. Definere rolleværdien 147correct148 krever ytterligere heuristikk, basert på den tilsiktede bruken av kjøretøymodellen. For et 147flying148-kjøretøy (som fly, romskip og ubåter) er det nyttig å definere rulle i form av bank. Den grunnleggende ideen om banktjenester er å justere 147floor148 av kjøretøyet (oppakse) med den tydelige tyngdekraften på grunn av sentrifugalkraft under en sving. Omvendt vil vi opp retningen til å justere med den sentripetale kraften som produserte manøvreren. I nærvær av tyngdekraft skal nedretningen justere seg med summen av svingakselerasjon og tyngdekraftakselerasjon. Vi vil også legge til i den aktuelle retningen for å dempe brå endringer i rulle. Så for å implementere banktjenester i den enkle kjøretøymodellen, er den omtrentlige oppveksten en vektet sum av: styreakselerasjon, tyngdekraft akselerasjon og den gamle opp. For en 147surface hugging148 (hjul, skyve eller legged) kjøretøy, vil vi begge begrense bilens 1414 posisjon til overflaten og justere kjøretøyet 146s opp til aksen til overflaten normal. I tillegg bør hastigheten begrenses til å være rent tangentiell mot overflaten. Disse kravene kan lett møtes hvis overflate manifolden er representert på en slik måte at et vilkårlig punkt i rommet (tilsvarende den gamle kjøretøyposisjonen) kan kartlegges til: (1) nærmeste punkt på overflaten og (2) overflaten normal på det tidspunktet. Hastigheten kan gjøres tangent ved å trekke av delen som er normal til overflaten. Vehicle146s posisjon er satt til punktet på overflaten, og overflaten normal blir sin oppakse. I denne enkle kjøretøymodellen består kontrollsignalet fra styreadferdene til bevegelsesadferdigheten av nøyaktig en vektomengde: en ønsket styringskraft. Mer realistiske kjøretøymodeller ville ha svært forskjellige sett med styresignaler. For eksempel har en bil et ratt, akselerator og bremse som hver kan representeres som skalar mengder. Det er mulig å kartlegge en generell styrkraftsvektor i disse skalar-signalene: styringsvektorens sidekomponent kan tolkes som styresignalet, styringsvektorens forover-komponent kan kartlegges i akseleratorsignalet dersom det er positivt eller inn i bremsesignal hvis negativ. Disse mappings kan være asymmetriske, for eksempel kan en typisk bil reduseres på grunn av bremsing mye raskere enn den kan akselerere på grunn av motorstøt, som vist i figur 2. Figur 2: Asymmetriske styrkrefter På grunn av antagelsen om hastighetsjustering, er dette enkle kjøretøyet Modellen kan ikke simulere effekter som skygger, spinn eller lysbilder. Videre tillater denne modellen kjøretøyet å svinge når hastigheten er null. De fleste virkelige kjøretøy kan ikke gjøre dette (de er 147 ikke-holonomiske148) og i alle fall tillater det uønskede store endringer i orientering i løpet av et enkelt trinn. Dette problemet kan løses ved å legge ytterligere begrensninger på retningsendring, eller ved å begrense sidestyringskomponenten ved lave hastigheter, eller ved å simulere treghetsmoment. Styreadferd Denne diskusjonen om spesifikke styreadferdier antar at lokomotiv er implementert av den enkle kjøretøymodellen beskrevet ovenfor, og er parameterisert av en enkelt styrkraftsvektor. Derfor er styringsadferdene beskrevet i form av den geometriske beregningen av en vektor som representerer en ønsket styrekraft. Merk at vanligvis størrelsen på disse styringsvektorer er irrelevant, siden de typisk vil bli klippet til maksimal styrke ved kjøretøymodellen. Legg også merke til at mange av samtalene til lengden og normaliseringsfunksjonene i disse formuleringene kan erstattes av raske rutiner som bruker en tilnærming til lengden som i Ohashi 94. Begrepet 147we148 eller 147our148 vil noen ganger bli brukt til å indikere tegnets førsteperspektiv blir styrt av en bestemt oppførsel. Animerte diagrammer som illustrerer disse atferdene kan bli funnet på nettet på Redcwrsteer Seek (eller jakten på et statisk mål) virker for å styre tegnet mot en bestemt posisjon i det globale rommet. Denne virkemåten justerer karakteren slik at dens hastighet er radialt rettet mot målet. Merk at dette er forskjellig fra en attraktiv kraft (som tyngdekraften) som ville produsere en bane rundt måletpunktet. Den 147verdige hastigheten148 er en vektor i retning fra tegnet til målet. Lengden på 147verdig hastighet148 kan være maksimalhastighet. eller det kan være character146s nåværende hastighet, avhengig av bestemt applikasjon. Styringsvektoren er forskjellen mellom denne ønsket hastighet og den aktuelle hastigheten for character146s, se figur 3. Hvis et tegn fortsetter å søke. Det vil etter hvert passere gjennom målet, og deretter gå tilbake for å nærme seg igjen. Dette gir bevegelse litt som en møl summende rundt en lyspære. Kontrast dette med beskrivelsen av ankomst nedenfor. Flee er rett og slett den inverse av søke og handlinger for å styre karakteren slik at dens hastighet er radialt rettet bort fra målet. Den ønskede hastigheten peker i motsatt retning. Figur 4: Forfølgelse og unnfangelse Forfølgelse ligner på å søke bortsett fra at steinbruddet (målet) er et annet bevegelige tegn. Effektiv forfølgelse krever en prediksjon av målets fremtidige posisjon. Tilnærmingen tatt her er å bruke en enkel prediktor og å revurdere det hvert simuleringstrinn. For eksempel tilsvarer en lineær hastighetsbasert prediktor antakelsen om at steinbrudd ikke vil vende under forutsigelsesintervallet. Selv om denne antagelsen ofte er feil, vil den resulterende prediksjonen bare bli brukt i ca. 130 sekunder. Posisjonen til et tegn T-tidsenheter i fremtiden (forutsatt at den ikke manøvrerer) kan oppnås ved å skalere sin hastighet med T og legge til den forskyvningen til dens nåværende stilling. Styring for jakt er da rett og slett resultatet av å søke søkestyringens oppførsel til den forutsagte målplasseringen. Se figur 4. Nøkkelen til denne gjennomføringen av forfølgelsen er metoden som brukes til å estimere prediksjonsintervallet T. Ideelt sett vil T være tiden til avlyting, men den verdien er ukjennelig fordi steinbruddet kan gjøre vilkårlig og uforutsigbar manøvrering. T kunne antas å være en konstant, som mens naiv, ville produsere bedre forfølgelse enn enkelt søk (som tilsvarer T0). Men for rimelig ytelse bør T være større når etterfølgeren er langt fra steinbrudd, og liten når de er i nærheten. En enkel estimator av moderat kvalitet er TDc hvor D er avstanden mellom etterfølger og steinbrudd, og c er en svingparameter. En mer sofistikert estimator kan oppnås ved å ta hensyn til de relative overskriftene for etterfølger og steinbrudd, og om forfølgeren er generelt foran, bak eller til siden av steinbrudd. Disse to beregningene kan uttrykkes i form av enkle prikkprodukter (mellom enhetens fremadvektorer og mellom steinbruddene146s fremover og motsetningen til forfølgeren146s posisjon). Vær oppmerksom på at det må tas hensyn til å redusere T (for eksempel til null) når etterfølgeren finner seg i samsvar med og foran sin steinbrudd. Another approach to both seek and pursuit is based on the fact that when our character is on a collision course with a target, it will appear at a constant heading in our character146s local space. Conversely our character can steer toward interception by contriving to keep the target at a constant heading. Evasion is analogous to pursuit . except that flee is used to steer away from the predicted future position of the target character. Optimal techniques for pursuit and evasion exist in the field of control theory Isaacs 65. The versions given here are in tended to be lightweight and are nonoptimal. In natural systems, evasion is often 147intentionally148 nonoptimal in order to be unpredictable, allowing it to foil predictive pursuit strategies, see Cliff 96. Figure 5: offset pursuit Offset pursuit refers to steering a path which passes near, but not directly into a moving target. Examples would be a spacecraft doing a 147fly-by148 or an aircraft doing a 147strafing run148: flying near enough to be within sensor or weapon range without colliding with the target. The basic idea is to dynamically compute a target point which is offset by a given radius R from the predicted future position of the quarry, and to then use seek behavior to approach that offset point, see Figure 5. To construct the offset point: localize the predicted target location (into our character146s local coordinate space) project the local target onto the character146s side-up plane, normalize that lateral offset, scale it by - R, add it to the local target point, and globalize that value. Arrival behavior is identical to seek while the character is far from its target. But instead of moving through the target at full speed, this behavior causes the character to slow down as it approaches the target, eventually slowing to a stop coincident with the target, as shown in Figure 6. The distance at which slowing begins is a parameter of the behavior. This implementation is similar to seek . a desired velocity is determined pointing from the character towards the target. Outside the stopping radius this desired velocity is clipped to maxspeed . inside the stopping radius, desired velocity is ramped down (e. g. linearly) to zero. Real world examples of this behavior include a baseball player running to, and then stopping at a base or an automobile driving towards an intersection and coming to a stop at a traffic light. Figure 7: obstacle avoidance Obstacle avoidance behavior gives a character the ability to maneuver in a cluttered environment by dodging around obstacles. There is an important distinction between obstacle avoidance and flee behavior. Flee will always cause a character to steer away from a given location, whereas obstacle avoidance takes action only when a nearby obstacle lies directly in front of the character. For example, if a car was driving parallel to a wall, obstacle avoidance would take no corrective steering action, but flee would attempt to turn away from the wall, eventually driving perpendicular to it. The implementation of obstacle avoidance behavior described here will make a simplifying assumption that both the character and obstacle can be reasonably approximated as spheres, although the basic concept can be easily extend to more precise shape models. Keep in mind that this relates to obstacle avoidance not necessarily to collision detection . Imagine an airplane trying to avoid a mountain. Neither are spherical in shape, but it would suffice that the plane146s bounding sphere avoids the mountain146s bounding sphere. A decomposable hierarchy of bounding spheres can be used for efficient representation of shapes for collision detection Hubbard 96, and presumably for obstacle avoidance too. An unrelated obstacle avoidance technique is described in Egbert 96. The geometrical construction of obstacle avoidance behavior bares some similarity to the offset pursuit behavior described above. It is convenient to consider the geometrical situation from the character146s local coordinate system. The goal of the behavior is to keep an imaginary cylinder of free space in front of the character. The cylinder lies along the character146s forward axis, has a diameter equal to the character146s bounding sphere, and extends from the character146s center for a distance based on the character146s speed and agility. An obstacle further than this distance away is not an immediate threat. The obstacle avoidance behavior considers each obstacle in turn (perhaps using a spatial portioning scheme to cull out distance obstacles) and determines if they intersect with the cylinder. By localizing the center of each spherical obstacle, the test for non-intersection with the cylinder is very fast. The local obstacle center is projected onto the side-up plane (by setting its forward coordinate to zero) if the 2D distance from that point to the local origin is greater than the sum of the radii of the obstacle and the character, then there is no potential collision. Similarly obstacles which are fully behind the character, or fully ahead of the cylinder, can be quickly rejected. For any remaining obstacles a line-sphere intersection calculation is performed. The obstacle which intersects the forward axis nearest the character is selected as the 147most threatening.148 Steering to avoid this obstacle is computed by negating the (lateral) side-up projection of the obstacle146s center. In Figure 7 obstacle A does not intersect the cylinder, obstacles B and C do, B is selected for avoidance, and corrective steering is to the character146s left. The value returned from obstacle avoidance is either (a) the steering value to avoid the most threatening obstacle, or (b) if no collision is imminent, a special value (a null value, or the zero vector) to indicate that no corrective steering is required at this moment. A final note regarding interaction of obstacle avoidance and goal seeking. Generally we only care about obstacles which are between us and our goal. The mountain beyond the airport is ignored by the airplane, but the mountain between the plane and the airport is very important. Wander is a type of random steering. One easy implementation would be to generate a random steering force each frame, but this produces rather uninteresting motion. It is 147twitchy148 and produces no sustained turns. A more interesting approach is to retain steering direction state and make small random displacements to it each frame. Thus at one frame the character may be turning up and to the right, and on the next frame will still be turning in almost the same direction. The steering force takes a 147random walk148 from one direction to another. This idea can be implemented several ways, but one that has produced good results is to constrain the steering force to the surface of a sphere located slightly ahead of the character. To produce the steering force for the next frame: a random displacement is added to the previous value, and the sum is constrained again to the sphere146s surface. The sphere146s radius (the large circle in Figure 8) determines the maximum wandering 147strength148 and the magnitude of the random displacement (the small circle in Figure 8) determines the wander 147rate.148 Another way to implement wander would be to use coherent Perlin noise Perlin 85 to generate the steering direction. Related to wander is explore (where the goal is to exhaustively cover a region of space) and forage (combining wandering with resource seeking). See Beer 90 and Tu 96 for more details. Figure 9: path following Path following behavior enables a character to steer along a predetermined path, such as a roadway, corridor or tunnel. This is distinct from constraining a vehicle rigidly to a path like a train rolling along a track. Rather path following behavior is intended to produce motion such as people moving down a corridor: the individual paths remain near, and often parallel to, the centerline of the corridor, but are free to deviate from it. In the implementation described here, a path will be idealized as a spine and a radius . The spine might be represented as a spline curve or a 147poly-line148 (a series of connected line segments). The path is then a 147tube148 or 147generalized cylinder:148 a circle of the specified radius, swept along the specified spine. The goal of the path following steering behavior is to move a character along the path while staying within the specified radius of the spine. If the character is initially far away from the path, it must first approach, then follow the path. To compute steering for path following . a velocity-based prediction is made of the character146s future position, as discussed above in regard to obstacle avoidance behavior. The predicted future position is projected onto the nearest point on the path spine. See Figure 9. If this projection distance (from the predicted position to the nearest on-path point) is less than the path radius, then the character is deemed to be correctly following the path and no corrective steering is required. Otherwise the character is veering away from the path, or is too far away from the path. To steer back towards the path, the seek behavior is used to steer towards the on-path projection of the predicted future position. Like in obstacle avoidance . a null or zero value is returned is returned if no corrective steering is required. A path can be followed without regard to direction, or in a specified direction (from A to B or from B to A) by adjusting the target point along the path in the desired direction. Figure 10: wall following and containment Variations on path following include wall following and containment as shown in Figure 10. Wall following means to approach a 147wall148 (or other surface or path) and then to maintain a certain offset from it Beer 90. For a discussion of offset goals, see offset pursuit above. Containment refers to motion which is restricted to remain within a certain region. Path following is a type of containment where the allowable region is a cylinder around the path146s spine. Examples of containment include: fish swimming in an aquarium and hockey players skating within an ice rink. To implement: first predict our character146s future position, if it is inside the allowed region no corrective steering is necessary. Otherwise we steer towards the allowed region. This can be accomplished by using seek with an inside point (for example, we can project the future position to the obstacle surface, and then extend this offset to obtain a target point) or we can determine the intersection of our path with the boundary, find the surface normal at that point, and then use the component of the surface normal which is perpendicular to our forward direction as the corrective lateral steering. Figure 11: flow field following Flow field following steering behavior provides a useful tool for directing the motion of characters based on their position within an environment. It is particularly valuable in some production teams because it allows motion specification to be made without use of programming and so can used by the art staff directly. In the case of game production this person might be a 147level designer148 and in animation production they might be a 147scene planner148 or 147layout artist.148 In flow field following behavior the character steers to align its motion with the local tangent of a flow field (also known as a force field or a vector field ). The flow field defines a mapping from a location in space to a flow vector: imagine for example a floor with arrows painted on it. Such a map, typically representing the floor plan of an environment, can be easily created by an artist with a special purpose 147paint148 program which allows them to draw the desired traffic flow with a paint brush. The implementation of flow field following is very simple. The future position of a character is estimated and the flow field is sampled at that location. This flow direction (vector F in Figure 11) is the 147desired velocity148 and the steering direction (vector S) is simply the difference between the current velocity (vector V) and the desired velocity. Figure 12: unaligned collision avoidance Unaligned collision avoidance behavior tries to keep characters which are moving in arbitrary directions from running into each other. Consider your own experience of walking across a plaza or lobby full of other walking people: avoiding collisions involves predicting potential collisions and altering your direction and speed to prevent them. If all nearby characters are aligned . a less complicated strategy can be used, see separation below. To implement this as a steering behavior, our character considers each of the other characters and determines (based on current velocities) when and where the two will make their nearest approach . A potential for collision exists if the nearest approach is in the future, and if the distance between the characters at nearest approach is small enough (indicated by circles in Figure 12). The nearest of these potential collisions, if any, is determined. The character then steers to avoid the site of the predicted collision. It will steer laterally to turn away from the potential collision. It will also accelerate forward or decelerate backwards to get to the indicate site before or after the predicted collision. In Figure 12 the character approaching from the right decides to slow down and turn to the left, while the other character will speed up and turn to the left. Figure 13: neighborhood The next three steering behaviors: separation . cohesion . and alignment . relate to groups of characters. In each case, the steering behavior determines how a character reacts to other characters in its local neighborhood . Characters outside of the local neighborhood are ignored. As shown in Figure 13, the neighborhood is specified by a distance which defines when two characters are 147nearby148, and an angle which defines the character146s perceptual 147field of view.148 Figure 14: separation Separation steering behavior gives a character the ability to maintain a certain separation distance from others nearby. This can be used to prevent characters from crowding together. To compute steering for separation . first a search is made to find other characters within the specified neighborhood. This might be an exhaustive search of all characters in the simulated world, or might use some sort of spatial partitioning or caching scheme to limit the search to local characters. For each nearby character, a repulsive force is computed by subtracting the positions of our character and the nearby character, normalizing, and then applying a 1r weighting. (That is, the position offset vector is scaled by 1r 2 .) Note that 1r is just a setting that has worked well, not a fundamental value. These repulsive forces for each nearby character are summed together to produce the overall steering force. See Figure 14. Figure 15: cohesion Cohesion steering behavior gives an character the ability to cohere with (approach and form a group with) other nearby characters. See Figure 15. Steering for cohesion can be computed by finding all characters in the local neighborhood (as described above for separation ), computing the 147average position148 (or 147center of gravity148) of the nearby characters. The steering force can applied in the direction of that 147average position148 (subtracting our character position from the average position, as in the original boids model), or it can be used as the target for seek steering behavior. Figure 16: alignment Alignment steering behavior gives an character the ability to align itself with (that is, head in the same direction andor speed as) other nearby characters, as shown in Figure 16. Steering for alignment can be computed by finding all characters in the local neighborhood (as described above for separation ), averaging together the velocity (or alternately, the unit forward vector) of the nearby characters. This average is the 147desired velocity,148 and so the steering vector is the difference between the average and our character146s current velocity (or alternately, its unit forward vector). This steering will tend to turn our character so it is aligned with its neighbors. Flocking behavior: in addition to other applications, the separation . cohesion and alignment behaviors can be combined to produce the boids model of flocks, herds and schools Reynolds 87 (see also Tu 94, Tu 96 and Hodgins 94). In some applications it is sufficient to simply sum up the three steering force vectors to produce a single combined steering for flocking (see Combining Behaviors below). However for better control it is helpful to first normalize the three steering components, and then to scale them by three weighting factors before summing them. As a result, boid flocking behavior is specified by nine numerical parameters: a weight (for combining), a distance and an angle (to define the neighborhood, see Figure 13) for each of the three component behaviors. Figure 17: leader following Leader following behavior causes one or more character to follow another moving character designated as the leader . Generally the followers want to stay near the leader, without crowding the leader, and taking care to stay out of the leader146s way (in case they happen to find them selves in front of the leader). In addition, if there is more than one follower, they want to avoid bumping each other. The implementation of leader following relies on arrival behavior (see above) a desire to move towards a point, slowing as it draws near. The arrival target is a point offset slightly behind the leader. (The offset distance might optionally increases with speed.) If a follower finds itself in a rectangular region in front of the leader, it will steer laterally away from the leader146s path before resuming arrival behavior. In addition the followers use separation behavior to prevent crowding each other. See Figure 17. Finally, here are quick sketches of some other steering behaviors that fit into the same general category as those described in more detail above. Interpose steering behavior attempts to put its character between two other moving characters, for example a soccer player trying to block a pass between two members of the opposing team. The general approach is similar to pursuit described above: predict the future position of the two other characters, determine a target point by interpolating between the future positions, and use seek to steer toward the target point. Related to leader following and pursuit . we could shadow our quarry by approaching and then using alignment to match their speed and heading. The arrival behavior described above can be considered a constraint on position and speed. This can be extended to simultaneously constrain orientation (to produce docking ), or a non-zero velocity, andor to meet these constraints at a given time. Hide behavior involves identifying a target location which is on the opposite side of an obstacle from the opponent, and steering toward it using seek . Combining Behaviors The individual steering behaviors described above serve as building blocks for more complex patterns of behavior. They are components of a larger structure, like notes of a melody or words of a story. In order to make interesting and life-like behaviors we need to select among, and blend between, these individual components. Unless an autonomous character exists in an very simple world, it would seldom make sense for the character to continually execute a single steering behavior. Combining behaviors can happen in two ways. A character may sequentially switch between behavioral modes as circumstances change in its world. For example, imagine caribou grazing in a meadow when suddenly they sense wolves approaching. This event triggers a discrete behavioral switch. All thoughts of grazing are forgotten as the caribou herd turns to flee from the predators. There is no tendency to mix these behaviors: a caribou will not slow down while running from a wolf in order to grab another bite of food. These discrete changes of behavioral state take place at the action selection level, the top of the three level behavioral hierarchy discussed in the Introduction. There is a extensive discussion of action selection in Tu 94, Tu 96 and in Blumberg 94. On the other hand, some kinds of behaviors are commonly blended together, effectively acting in parallel. For example, as the caribou flee through the forest, they blend evasion and obstacle avoidance together to allow them to escape from the wolves while dodging trees. A caribou cannot afford to ignore either component behavior, it must always be moving in a direction that both takes it away from the wolf and avoids collisions with trees. This behavioral blending occurs at the middle steering level of the behavioral hierarchy. Blending of steering behaviors can be accomplished in several ways. The most straightforward is simply to compute each of the component steering behaviors and sum them together, possibly with a weighting factor for each of them. (Note that steering vectors are especially easy to blend, other kinds of behaviors, producing other kinds of values (e. g. conversational behaviors) could be much harder to combine.) This simple linear combination often works well, but has at least two shortcomings: it is not the most computationally efficient approach, and despite adjusting the weights, component behaviors may cancel each other out at inopportune times. The computation load can be decreased by observing that a character146s momentum serves to apply a low-pass filter to changes in steering force. So rather than compute several steering components each simulation step and average them together, we could instead select one steering component to compute and apply each frame, and depend on momentum (and perhaps some explicit damping of acceleration) to blend them together. The problem of components canceling each other out can be addressed by assigning a priority to components. (For example: first priority is obstacle avoidance . second is evasion . ) The steering controller first checks to see if obstacle avoidance returns a non-zero value (indicating a potential collision), if so it uses that. Otherwise, it moves on to the second priority behavior, and so on. A hybrid of these techniques that the author has found useful is 147prioritized dithering148: with a certain probability the first priority behavior is evaluated, and if it returns a non-zero (non-null) value that will be used. Otherwise (if the behavior returns zero, or it was skipped over due to the random selection) the second priority behavior is considered, and so on. In Reynolds 87 a blending scheme called 147prioritized acceleration allocation148 was used with the boids flocking model. The basic idea was that by adjusting their magnitude, higher priority behaviors could decide whether or not to leave any steering force for use by lower priority behaviors. In the course of several reimplementations of boids over the years, a simple linear combination of the component behaviors has proved sufficient. When combining flocking with other behaviors such as obstacle avoidance, both simple summing and prioritized dither have been used successfully. Conclusions This paper defined 147autonomous character148 in terms of autonomous agents and improvisational action. It presented a decomposition of the task of constructing motion behaviors for autonomous characters into a three level hierarchy of: action selection, steering, and locomotion. It has defined a minimal implementation of the locomotion level in terms of a 147simple vehicle model.148 It has then presented a collection of simple, common steering behaviors. (Including: seek . flee . pursuit . evasion . offset pursuit . arrival . obstacle avoidance . wander . path following . wall following . containment . flow field following, unaligned collision avoidance . separation . cohesion . alignment . flocking . and leader following .) Finally it has described some techniques for blending these simple steering behaviors together. Acknowledgments The techniques described in this paper have been developed over the last twelve years, for many different projects, at several companies. I wish to acknowledge the helpful cooperation of all the companies and coworkers involved. Specifically I wish to thank the following people for managerial support and technical collaboration. At Sony Computer Entertainment America: Phil Harrison, John Phua, Attila Vass, Gabor Nagy, Sky Chang, and Tom Harper. At DreamWorks Feature Animation: Dylan Kohler, Bart Gawboy, Matt Arrott, Lance Williams, Saty Raghavachary, and Mike Ullner. At SGI146s Silicon Studio: Bob Brown, Leo Blume, Roy Hashimoto, and especially my behavioral animation colleague Xiaoyuan Tu. At Electronic Arts: Luc Barthelet, Steve Crane, Kelly Pope, Steve Sims, and Frank Giraffe. At Symbolics Graphics Division: Tom McMahon, Andy Kopra, Larry Malone, and Michael Wahrman. Finally, loving thanks to my wife Lisa and our children Eric and Dana. Before they become fully autonomous, I hope I steer the kids in the right direction. Certainly they are already real characters . References Arkin, Ronald (1987) 147Motor Schema Based Navigation for a Mobile Robot: An Approach to Programming by Behavior148, Proceedings of IEEE Conference on Robotics and Automation, pages 264-271. Arkin, Ronald (1989) 147Motor Schema-Based Mobile Robot Navigation148, International Journal of Robotics Research, 8(4) pages 92-112. Arkin, Ronald (1992) 147Behavior-based Robot Navigation in Extended Domains148, Journal of Adaptive Behavior, 1(2) pages 201-225. Bates, Joseph Loyall, Bryan Reilly, Scott (1992) 147An Architecture for Action, Emotion, and Social Behavior148, Proceedings of the Fourth European Workshop on Modeling Autonomous Agents in a Multi-Agent World, S. Martino al Camino, Italy. cs. cmu. eduafscs. cmu. eduprojectozwebpapersCMU-CS-92-144.ps Beer, R. D. (1990). Intelligence as Adaptive Behavior: Experiments in Computational Neuroethology. New York: Academic Press. See also: yuggoth. ces. cwru. edujohngrobopaperrobopaper. html Blumberg, Bruce and Galyean, Tinsley (1995) Multi-Level Direction of Autonomous Creature for Real-Time Virtual Environments, Proceedings of SIGGRAPH 95, in Computer Graphics Proceedings, Annual Conference Series, ACM SIGGRAPH, pages 47-54. bruce. media. mit. edupeoplebruceSiggraph95.final. ps Blumberg, Bruce (1994) 147Action-Selection in Hamsterdam: Lessons from Ethology148 in the Proceedings of the Third International Conference on the Simulation of Adaptive Behavior (SAB94), D. Cliff, P. Husbands, J-A Meyer, and S. Wilson, Editors, MIT Press, Cambridge, Massachusetts, pages 108-117 Braitenberg, Valentino (1984) Vehicles: Experiments in Synthetic Psychology, The MIT Press, Cambridge, MA. Brooks, Rodney A. (1985) quotA Robust Layered Control System for a Mobile Robot, quot IEEE Journal of Robotics and Automation 2(1), March 1986, pp. 14--23 also MIT AI Memo 864, September 1985. See: ai. mit. edupeoplebrookspapersAIM-864.pdf Bruderlin, Armin and Calvert, Tom (1989) quotGoal-Directed, Dynamic Animation of Human Walkingquot, ACM SIGGRAPH89, Proceedings, vol. 23, pp 233-242. Cliff, Dave and Miller, Geoffrey (1996) 147Co-Evolution of Pursuit and Evasion II: Simulation Methods and Results148, From Animals to Animats 4: Proceedings of the Fourth International Conference on Simulation of Adaptive Behavior (SAB96), Maes, Mataric, Meyer, Pollack, and Wilson editors, ISBN 0-262-63178-4, MIT Press. cogs. susx. ac. ukusersdavecpe. html Costa, Mocircnica Feijoacute, Bruno and Schwabe, Daniel (1990) Reactive Agents in Behavioral Animation, Anais do SIBGRAPI 95 . Lotufo and Mascarenhas editors, Satildeo Carlos, SP, pages 159-165. icad. puc-rio. br Cremer, James Kearney, J. Willemsen, P. (1996) 147A Directable Vehicle Behavior Model for Virtual Driving Environments148, Proceedings of 1996 Conference on AI, Simulation, and Planning in High Autonomy Systems, La Jolla, CA. cs. uiowa. edu Egbert, Parris and Winkler, Scott (1996) 147Collision-Free Object Movement Using Vector Fields148, IEEE Computer Graphics and Applications, 16(4), pages 18-24. wwwputer. orgcgacg1996g4toc. htm Johnson, Michael (1994) WavesWorld: A Testbed for Three Dimensional Semi-Autonomous Animated Characters, PhD Thesis, MIT Media Lab. media. mit. edu Hayes-Roth, Barbara. and van Gent, R. (1996) 147Improvisational Puppets, Actors, and Avatars148, Proceedings of the 1996 Computer Game Developers146 Conference. Stanford Knowledge Systems Laboratory Report KSL-96-09. file:www-ksl. stanford. edupubKSLReportsKSL-96-09.ps Hodgins, Jessica and Brogan, David (1994) 147Robot Herds: Group Behaviors for Systems with Significant Dynamics148, Proceedings of the 4th International Workshop on the Synthesis and Simulation of Living Systems (Artificial Life IV), Brooks and Maes editors, MIT Press, Cambridge, MA, pages 319-324. cc. gatech. edugvuanimationpapersalife. ps. gz Hodgins, Jessica Wooten, W Brogan, D and O146Brien, J (1995) 147Animating Human Athletics148, Proceedings of SIGGRAPH 95, in Computer Graphics Proceedings, Annual Conference Series, Robert Cook editor, ACM SIGGRAPH, pages 71-78. cc. gatech. edugvuanimationpaperssig95.ps. gz Hubbard, Philip (1996) 147Approximating Polyhedra with Spheres for Time-Critical Collision Detection148, ACM Transactions on Graphics, 15(03), pages 179-210. acm. orgpubstoghubbard96 Isaacs, Rufus (1965) Differential Games: A Mathematical Theory with Application to Warfare and Pursuit, Control and Optimization, John Wiley and Sons, New York. Kahn, Kenneth (1979) 147Creation of Computer Animation from Story Descriptions148, Technical Report 540, MIT AI Lab, Cambridge, Mass. Maes, Pattie Darrell, T. Blumberg, B. (1995) 147The Alive System: Full Body Interaction with Autonomous Agents148, Computer Animation 14695 Conference, IEEE Press, pages 11-18. Mataric, Maja (1993) 147Designing and Understanding Adaptive Group Behavior148, Adaptive Behavior 4(1), pages 51-80. www-robotics. usc. edu Ohashi, Yoshikazu (1994) 147Fast Linear Approximations of Euclidean Distance in Higher Dimensions148, in Graphics Gems IV, Paul Heckbert editor, Academic Press. See ftp:princeton. edupubGraphicsGraphicsGemsGemsIV Perlin, Ken (1985) 147An Image Synthesizer148, in SIGGRAPH 85 Proceedings, Computer Graphics 19(3), pages 287-296. See mrl. nyu. eduperlindocoscar. html Perlin, Ken (1995) 147Real Time Responsive Animation With Personality148, IEEE Transactions on Visualization and Computer Graphics, 1(1) pages 5-15, ISSN 1077-2626. Perlin, Ken and Goldberg, Athomas (1996) 147Improv: A System for Scripting Interactive Actors in Virtual Worlds148, Proceedings of SIGGRAPH 96, in Computer Graphics Proceedings, Annual Conference Series, ACM SIGGRAPH, pages 205-216. mrl. nyu. eduimprov mrl. nyu. eduimprovsig-paper96 Raibert, Marc and Hodgins, Jessica (1991) 147Animation of Dynamic Legged Locomotion148, Computer Graphics 25(4), Proceeding of SIGGRAPH 91, Thomas Sederberg editor, ISSN 0097-8930, pages 349-358. Raibert, M. Hodgins, J. Ringrose, R. Playter, R. Borvansky, L. Campbell, L. Evans, D. Crane, C. Lamb, M. (1991) 147On The Run148, SIGGRAPH 91 Electronic Theater, SIGGRAPH Video Review, Issue 72, ai. mit. eduprojectsleglabsimulationsotrotr. html Reese, Bjoslashrn and Bryan, Stout (1999) 147Finding a Pathfinder148 to appear: AAAI 99 Spring Symposium on Artificial Intelligence and Computer Games. See also: redbreesenavigation. html Resnick, Mitchel (1989) 147LEGO, Logo, and Life148, Proceedings of the Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (ALife 14687), SFI Studies in the Sciences of Complexity, Volume 6, Christopher Langton editor, Addison-Wesley, Redwood City, CA, USA, pages 397-406. Reynolds, C. W. (1987) Flocks, Herds, and Schools: A Distributed Behavioral Model, in Computer Graphics, 21(4) (SIGGRAPH 14687 Conference Proceedings) pages 25-34. redcwrboids. html Reynolds, C. W. (1988) Not Bumping Into Things, in the notes for the SIGGRAPH 88 course Developments in Physically-Based Modeling, pages G1-G13, published by ACM SIGGRAPH. redcwrnobumpnobump. html Ridsdale, Gary (1987) 147The Director146s Apprentice: Animating Figures in a Constrained Environment148, Ph. D. thesis, School of Computing Science, Simon Fraser University. Sims, Karl (1994) 147Evolving Virtual Creatures148, Proceedings of SIGGRAPH 94, Computer Graphics Proceedings, Annual Conference Series, Andrew Glassner editor, ACM SIGGRAPH, ISBN 0-89791-667-0, pages 15-22. biota. orgksimsblockiesindex. html. ftp:ftp. de. uupubresearchciAlifekarl-sims Still, G Keith (1994) 147Simulating Egress using Virtual Reality - a Perspective View of Simulation and Design148, IMAS Fire Safety on Ships. Strassmann, Steve (1991) Desktop Theater: Automatic Generation of Expressive Animation, PhD thesis, MIT Media Lab. methodstrazstraz-phd. pdf Thalmann, Daniel Renault, Olivier and Magnenat-Thalmann Nadia (1990) 147A Vision-Based Approach to Behavioral Animation148, Journal of Visualization and Computer Animation, John Wiley amp Sons, 1(1) pages 18-21. Travers, Michael (1989) 147Animal Construction Kits148, Proceedings of the Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (ALife 14687), SFI Studies in the Sciences of Complexity, Volume 6, Christopher Langton editor, Addison-Wesley, Redwood City, CA, USA, pages 421-442. lcs. media. mit. edupeoplemtagaragar. html Travers, Michael (1994) 147LiveWorld: A Construction Kit for Animate Systems148, Proceedings of ACM CHI14694 Conference on Human Factors in Computing Systems, pages 37-38. mt. media. mit. edupeoplemtpaperschi94chi94.html Tu, Xiaoyuan and Terzopoulos, Demetri (1994) 147Artificial Fishes: Physics, Locomotion, Perception, Behavior148, Proceedings of SIGGRAPH 94, Computer Graphics Proceedings, Annual Conference Series, Andrew Glassner editor, ACM SIGGRAPH, ISBN 0-89791-667-0, pages 43-50. dgp. toronto. edupeopletupaperssig94.ps Tu, Xiaoyuan (1996) Artificial Animals for Computer Animation: Biomechanics, Locomotion, Perception, and Behavior, PhD dissertation, Department of Computer Science, University of Toronto. dgp. toronto. edupeopletuthesisthesis. html van de Panne, M. Fiume, E. and Vranesic, Z. G. (1990) quotReusable Motion Synthesis Using State-Space Controllersquot, Proceedings of SIGGRAPH 90, In Computer Graphics Proceedings, ACM SIGGRAPH, pages 225-234. See dgp. toronto. edupeoplevanpapers. html Grey-Walter, W. (1953) The Living Brain. Gerald Duckworth and Co. Ltd. Wiener, Norbert (1948) Cybernetics, or control and communication in the animal and the machine. Cambridge, Massachusetts: The Technology Press New York: John Wiley amp Sons, Inc. Wilhelms, Jane and Skinner, Robert (1990) A 147Notion148 for Interactive Behavioral Animation Control, IEEE Computer Graphics and Applications, 10(3), pages 14-22. Zapata, R. Lepinay, P. Novales, C. and Deplanques, P. (1992) 147Reactive Behaviors of Fast Mobile Robots in Unstructured Environments: Sensor-Based Control and Neural Networks148 in From Animals to Animats 2: Proceedings of the Second International Conference on Simulation of Adaptive Behavior (SAB92), Meyer, Roitblat and Wilson editors, MIT Press, Cambridge, Massachusetts, pages 108-115 Zeltzer, David (1983) Knowledge-Based Animation, Proceedings SIGGRAPHSIGART Workshop on Motion, pages 187-192. Zeltzer, David (1990) 147Task Level Graphical Simulation: Abstraction, Representation and Control,148 in Making Them Move: Mechanics, Control and Animation of Articulated Figures, N. Badler, B. Barsky and D. Zeltzer, editors. Morgan Kaufmann Publishers. January 14, 1997Integrated Amplifiers amp Receivers Audio Research VSi60: 5000 10025 In the 50Wpc VSi60, a passive line stage is combined with a JFET input stage driving one 6H30 driver tube per channel. Each channels output stage has a matched pair of Svetlana 6550C push-pull tubes with a combination of pentode operation and ARCs partially cathode-coupled topology. Convenient to use, the VSi60 provides output taps at 4 and 8 ohms, as well as four pairs of voltmeter test points. Though it lacked the ultimate control of more powerful amplifiers, the VSi60 combined a glorious midrange with clean, detailed high frequencies and outstanding low-level dynamic articulation. In the VSi60, Audio Research has produced an integrated amplifier of staggering quality, versatility, and value, said BJR. Add 300 for tube cage. (Vol.33 No.9 WWW) Ayre Acoustics AX-5: 9950 Ayres statement integrated amplifier measures 17.25 W by 5 H by 18.75 D, weighs 46 lbs, and has a brushed-aluminum chassis with four balanced and two single-ended inputs. It uses a variable-gain transconductance circuit trickled down from Ayres top-of-the-line KX-R preamp, a class-AB diamond-circuit output section, and delivers 125Wpc into 8 ohms. The AX-5 offered a combination of incisive detail, perfect musical timing and momentum, and naturally warm timbral colors, said AD. Its one of the three best, most musical, most human-sounding solid-state amps Ive ever heard, he concluded. Its wide bandwidth and low distortion are a testament to the intrinsic linearity of the diamond circuit, added JA. In his Follow-Up, Jim Austin expressed admiration for the Ayres minimalist aesthetic, but had no love for its user interface or loudspeaker connectors151and, as did AD in his full review, criticized the noise made by its volume control. Yet he thought the AX-5 sounded superb: immediate and full, with much (but not all) of the richness, texture, and corporeality I associate with a good tube amp. About to be replaced by 20th Anniversary edition. (Vol.36 No.8, Vol.37 No.10 WWW) Ayre Acoustics AX-7e: 3500 10025 The success of this 60Wpc, solid-state, two-channel, fully balanced, integrated amplifier depended on the associated sources. Used from balanced output to balanced input, It was brilliant. Fantastisk. Stirring, even, said AD. However, used as an unbalanced amp, The AX-7 still sounded good, but its musical performance lacked momentum and, ultimately, excitement. Overall, the Ayre was colorful, clear, well-textured, and spatially convincing. It seemed sensitive to the type and length of speaker cable AD used, and seemed more sensitive to AC power quality than average. I strongly recommend the Ayre AX-7 for use only in an all-balanced system. The 7es power supply now includes greater filtering of the AC mains, increased peak current delivery, and filtering of the rectifier switching noise. In addition, the AX-7es gain stages now use two-stage voltage regulators in place of the earlier versions single-stage regulators. The sound now combined classic Brit-style pacing and tunefulness with near-SET levels of presence and a fine sense of musical flow, a combination that allowed AD to become emotionally involved in the music. The AX-7e is the best integrated Ive ever heard, endorsed WP. One heck of an involving amplifier, he summed up. Compared to the Luxman MQ-88 power amplifier, the Ayre offered greater bass extension and soundstage control but lacked the Luxmans beguiling midrange, said JM. Original AX-7s can be fully upgraded for 250150350, depending on the age of the unit. (Vol.26 No.10 AX-7 Vol.29 No.1, Vol.31 No.3, AX-7e WWW see also The Fifth Element in Vol.34 No.2 and Vol.35 No.4 WWW) Bel Canto C7R: 2595 Made in the US, the versatile C7R is a 60Wpc solid-state FM receiver with four SPDIF digital inputs (two coaxial, two TosLink), two pairs of analog inputs (one phono, one line), one pair of line-level analog outputs, a front-panel headphone jack, and a built-in DAC with USB input capable of handling resolutions up to 24-bit96kHz. Though it lacked some midrange clarity and high-frequency smoothness, the C7R sounded clear, immediate, and powerful, with excellent dynamics and tight, tuneful bass, said EL. I loved the Bel Canto C7R. I recommend it to anyone who wants simplicity and great sound, he concluded. Overall, the C7R is a well-designed piece of hardware, said JA. (Vol.36 No.3 WWW) Jadis I-35: 7995 Made in France, the beautiful I-35 is a tubed, integrated amplifier with five line-level inputs. Though rated to deliver 35Wpc into 115016 ohms, the I-35 produced just 17W into 8 ohms at 1 THD. It uses five small-signal tubes (three 12AU7s and a pair of 12AX7s) and two pairs of KT120 output tubes run in autobias mode in an Ultralinear circuit, with plates and screen grids for each channel tied to the split-coil primaries of transformers expertly designed and made in-house. Build quality and cosmetics were outstanding, inside and out. Though it lacked the fullness and richness of Arts Shindo separates, the Jadis produced a natural and engaging overall sound, with an excellent sense of momentum and a very good sense of the spatial relationships between different sounds in a stereo recording, said AD. This is a damn good amp for getting to the essence of music, he concluded. (Vol.37 No.1 WWW) Kondo Overture: 33,900 Made in Japan, the 32Wpc Overture uses a class-A, Ultralinear output circuit with minimal (3dB) global feedback, executed with split-primary output transformers custom-wound by Tango. It uses two Electro-Harmonix EL34 output pentodes, one 6072, and one 12BH7 per channel. Build quality was exceptional and marked by silver wiring, handmade capacitors, bespoke resistors, a solid-copper ground plane, and a tuned chassis made from a combination of steel, brass, and aluminum. The Overture has no balance control, remote control, mono switch, headphone amplifier, or phono stage, but does provide four pairs of line-level inputs and a choice of 4- or 8-ohm output sockets. Though it was slightly lean and not quite as colorful as ADs Shindo separates, the Overture produced a natural, compelling overall sound with a well-extended treble and exceptional senses of drive and scale. (Vol.36 No.11 WWW) Leben CS300: 3395 10025 With its wood side panels, gold-toned faceplate, and large balance and bass-boost knobs, the line-only CS300 has a decidedly old-fashioned look and feel. It uses two pairs of EL84 power pentode tubes running in class-AB mode to deliver 12Wpc. A rear-panel output control allows the user to switch between transformer secondaries that are optimized for use with speakers with impedances of 4, 6, or 8 ohms. Construction quality was superb throughout. Though it lacked the color, presence, internote silence, and sense of flow of ADs Shindo separates, the Leben distinguished itself as a punchy and realistically textured amp with an especially deep, tight bottom end. It measured about as well as can be expected from its retro design, commented JA. (Vol.34 No.11 WWW) LFD NCSE: 6795 A bigger and more powerful version of LFDs excellent Mk. IV LE, the NCSE Mk. II measures 17.2 W by 3.25 H by 15.7 D, and is rated to deliver 70Wpc. The basic circuit remains the same, with a pair of MOSFET output transformers for each channel. The NCSE Mk. II retained the smaller models speed, transparency, harmonic accuracy, and illuminated-from-within quality, but added improved bass and dynamics, said ST. If you like the Mk. IV LE but feel youre running out of power, the NCSE Mk. II would be the better choice, particularly if youre driving bigger speakers in a bigger room, he advised. (Vol.36 No.11) LFD LE V: 4495 Dr. Richard Bews, proprietor of Englands LFD electronics, has once again upgraded his and Dr. Malcolm O. Hawksfords original integrated amplifier, the 65Wpc LE (n233e Mistral): a perennial ST favorite, owing to its complete lack of convenience or luxury features that degrade sound and add expense. To those ends, all iterations of LFDs solid-state amp have lacked balance and tone controls, headphone jacks, extra sets of speaker outputs, digital displays, and, especially, remote controls. The LE, which Dr. Bews builds in small batches, provides five line-level inputs (but no phono section), and is made with only the highest-quality parts, some vintage (NOS), some custom-made. According to ST, the new LE V151whose improvements all seem to arise from a combination of sturdier casework and more refined parts151is a showstopper. I mean that literally. I didnt want to write I wanted to listen. Compared to the LE IV, which ST owns and loves, the LE V represents a substantial improvement. Everything just fit into place: harmonics, timing, resolution. Quoth he: Buy your LFD LE V today. (Vol.37 No.7) Pass Labs INT-150: 7150 10025 The push-pull, class-AB INT-150 is rated to deliver 150Wpc (191Wpc at clipping), and has a high-quality volume control and five inputs: two XLRRCA and three RCA-only. It uses Passs Super-Symmetry Circuit, previously featured in all X-series models, to naturally eliminate distortions from the audio signal. Though it produced a slightly forward midrange and top end, the INT-150 combined wide dynamic range with great rhythmic drive, a broad soundstage, and tight, tuneful bass, said EL, who recommended high Class B. Measured performance was excellent in most respects, said JA, who feels low Class A is the appropriate rating. (Vol.34 No.1 WWW) PrimaLuna DiaLogue Premium HP: 4399 PrimaLunas DiaLogue Premium HP employs a total of eight power tubes151EL34s are standard, although the user can substitute a variety of other power pentodes151to produce 40Wpc in triode mode or 70Wpc in Ultralinear mode. (On-the-fly switching between modes can be performed with the DiaLogue Premium HPs remote handset.) Additional features include a newly designed, six-tube front end the highest-end implementation yet of PrimaLunas Adaptive AutoBias circuit and an all-tube headphone amp. RD, a fan of the companys earlier ProLogue Premium, found that the new amp provided higher power with no loss of finesse, and declared, PrimaLuna had taken a major step forward in amplifier performance. JAs measurements indicated that PrimaLunas output transformers are of excellent quality. Distortion wasnt the lowest JA has seen, but, fortunately, the distortion is heavily second-harmonic in nature in both modes, even at low frequencies. He concluded that, overall, the PrimaLuna DiaLogue Premium HP measures well for a design using push-pull pairs of EL34 tubes. RD summed up: 4199 for an integrated amp of the DiaLogue Premium HPs level of performance represents excellent value. (Vol.37 No.12 WWW) PrimaLuna DiaLogue Premium: 3399 PrimaLunas top-of-the-line integrated amplifier is rated to deliver 36Wpc with its stock EL34 output tubes or 43Wpc with optional KT120s. It uses PrimaLunas Adaptive AutoBias feature for easy swapping of output tubes, and has a bad-tube indicator, power-transformer protection, and output-transformer protection circuitry. The DiaLogue Premium was extremely quiet and sounded bigger than its power rating suggested, with a rich midrange and an excellent sense of timing, said ST. The DiaLogue Premium will be a dream come true for anyone who has a closetful of output tubes, he concluded. (Vol.36 No.6) PrimaLuna ProLogue Premium: 3399 Designed in Holland and made in China, the solidly built ProLogue Premium is rated to deliver 35Wpc with EL34 tubes or 40Wpc with KT88s. It has a heavy-gauge, ventilated case with a lustrous five-coat finish, features point-to-point wiring, and offers five pairs of RCA input jacks, a Home Theater bypass, and connections for speaker loads of 4 and 8 ohms. Though it lacked the three-dimensional imaging, detailed highs, and extended bass of more expensive amplifiers, the ProLogue Premium produced a natural, inviting midrange and performed well with a wide variety of speakers. The PrimaLuna ProLogue Premium offers outstanding sound quality at a very reasonable price, said RD. JA noted respectable measured performance for a classic tube design. (Vol.35 No.6 WWW) Simaudio Moon Evolution 700i: 14,000 10025 Robustly built of thick, ultrarigid aluminum, the 700i is a fully differential dual-mono design rated to deliver 175Wpc (190Wpc at actual clipping), running in class-A up to 5W and in class-AB thereafter. Its output stages are powered by six bipolar transistors per channel for a wide bandwidth and low noise floor, while its zero global feedback design works to boost the speed of the signal response and eliminate intermodulation distortion. Though it couldnt match the Krell FBIs transient speed or deep-bass extension, the 700i had a full-blooded, dynamic, seamless sound marked by vivid tonal colors, harmonic integrity, and a strong sense of rhythm. With the 700i, I found myself drawn deeper into the music, said FK. (Vol.34 No.3 WWW) TA Power Plant Balanced: 3300 10025 The 140Wpc Power Plant looks almost identical to TAs Music Player, and the two comprise a fully functional audio system in a single stack. Connection via a supplied RJ-12 cable coordinates the functions of the MP and PP, and allows the pair to be operated by a single remote. The PPs switch-mode output stages were developed in-house, and combine MOSFET transistors with high-energy driver modules. JI noted a dynamic-sounding amplifier section that exhibited ample and well-controlled bass along with a smooth, detailed top end. Surprisingly robust and detailed sounding amp for such a small cool-running package, he concludes. (Vol.32 No.8, Vol.35 No.9 WWW) Unison Research S6: 5395 Each channel of the Italian-made S6 uses three EL34 power pentode tubes in a class-A, single-ended, triple-parallel configuration to deliver 35Wpc into 8 ohms. Its distinctive chassis is longitudinally divided into three portions: a central raised center that houses the iron-core transformers is flanked by identical lower sections that hold the tube sockets, bias meters, and bias-adjustment controls. Compared with the Ars-Sonum Filarmona SE, the S6 was more forceful but gave up nothing in finesse when partnered with the Opera Callas loudspeakers, the S6 was fatigue-free, with surprising dynamics and bass, said JM. (Vol.36 No.8 WWW) Cambridge Audio Azur 851A: 1900 First cousin to the Cambridge Azur 851D DAC, the Azur 851A integrated amplifier provides 120Wpc into 8 ohms, two balanced inputs (XLR), eight unbalanced line-level inputs (RCA), and the rare luxury of bass and treble tone controls. HR wrote that he could best describe the sound of the Azur 851A as relaxed and enjoyably colorful, in a class-A triode sort of way. It sounded more naturally toned and weighty than my Creek 3330 or my Line Magnetic LM-518IA, and showed none of that off-putting grayness or brittleness often heard in low-priced, high-powered amps. In his measurements, JA discovered that the amps right-channel performance was not in keeping with that of the left channel151although he suggested that the right channels relative shortcomings were inaudible. His conclusion: Assuming that the less-good performance of its right channel was a sample-specific fault, Cambridge Audios Azur 851A is a well-built amplifier that offers a lot of power with very low distortion at an affordable price. HRs last word: a versatile and extraordinarily musical cornerstone on which to build a truly enjoyable high-end system that can play all types of music with righteous aplomb for little cost. (Vol.38 No.2 WWW) Cayin SP-10A: 2195 10025 With its brushed aluminum faceplate and cabinet of gorgeous cherrywood (walnut and piano lacquer are also available), the 38Wpc SP-10A has a decidedly old-fashioned appearance. It uses four 6L6GC output tubes, two 12AU7 driver tubes, and one 12AX7 input tube has four line-level inputs and offers both 4 and 8 ohm output transformer taps. Though it lacked top-end sparkle and ran out of gas at high volumes, the Cayin offered a lush, natural midrange with outstanding resolution of inner detail and ambience, said BJR. JAs measurements uncovered a curious instability and lack of distortion-free power from the amps 4 ohm tap. (Vol.35 No.4 WWW) Creek Evolution 50A: 1195 Creeks new entry-level model is a class-AB design claimed to deliver 55Wpc into 8 ohms. It offers four single-ended and one balanced input, one of which can accommodate one of three MM or MC plug-in phono boards. The clean front panel has input-selector and Volume knobs, Balance and Tone controls, and a large display. The Creek produced an uncolored midrange, impressive bass impact, and excellent resolution of high-frequency detail, delicacy, and air, said BJR. An involving, flexible, and good-sounding piece of electronics, he concluded, adding that it sounds more powerful than its rating would imply. Considering its affordable price, Creeks Evolution 50A measures fundamentally well, said JA, though he points out that its output stage is underbiased. (Vol.36 No.8 WWW) Croft Phono Integrated: 1895 Crofts 45Wpc Phono Integrated combines in a single package the companys Micro 25 preamplifier and Series 7 power amplifier to create a hybrid integrated in which line - and moving-magnet150compatible phono-stage gain is provided by ECC83 vacuum tubes, output power by transistors. In addition to its phono input, the Croft has three analog line inputs, but offers no remote control, digital inputs, headphone jack, or upgrade paths for USB connectivity. Apart from a small circuit board containing the bipolar timer and relays, the Phono Integrated is hand-wired, point to point, with neatly made solder joins and Bakelite terminal strips. AD liked the Crofts dual-mono volume controls, SM not so much. They agreed, however, that the Crofts sound was extraordinary: smooth, coherent, open, naturally detailed, forceful, physical, and dynamic, with a great sense of space and an expert ability to drive a beat forward. If I were a designer or builder, this is how I would do the thing. If I were buying in this price range, this is the one Id choose, raved AD. On JAs test bench, however, the Croft exhibited a nonflat RIAA response and high levels of harmonic and intermodulation distortion. According to ST, the sound of the Phono Integrated was musical in a way that very few hi-fi components are. Apart from noting this models 1970s-style cosmetics (A pox on cosmetics), minimalist conveniences (A pox on convenience), and slightly plump bass, ST declared the Croft one of the best integrated amplifiers I have ever heard. (Vol.36 No.10, Vol.37 No.5 WWW) HarmanKardon HK 990: 2599 The gorgeous, versatile, 150Wpc HK 990 is a two-channel integrated amplifier with analog and digital inputs, tone controls, bass management, and system equalization. Line-level analog inputs include six pairs single-ended, one pair balanced, a processor HT bypass, and two subwoofer digital inputs include two optical and two coaxial. In addition, there are moving-magnet and moving-coil phono inputs, two subwoofer outputs, two coaxial outputs, and a front-panel headphone jack. Its Analog Devices AD1955 DAC chip handles resolutions up to 24-bit96kHz. Though it was less transparent than the Parasound Halo JC 2 preamp, the HK 990 offered impressive performance with both analog and digital sources, exhibiting taut bass, clean mids, and outstanding power and control. The HK 990 should be on every audiophiles shopping list, said KR, adding that its a harbinger of the future of integrated amplifiers. JA noted that the HKs measured performance was not compromised by its wealth of versatile features. I am impressed, he said. (Vol.34 No.12 WWW) Manley Labs Stingray iTube: 6000 10025 Manley Labs Stingray II: 5650 10025 Direct descendent of the acclaimed Stingray (reviewed by CS in December 1999), the rugged-looking, gorgeously constructed Stingray iTube is built around eight EL84 output tubes, rated to deliver 32Wpc in ultralinear mode and 18Wpc in triode mode. It offers three single-ended inputs, a certified iPod dock, a subwoofer output, and a 14 headphone jack. An IRRF remote control provides full control of the amp, as well as the track functions of a docked iPod. After about 250 hours of break-in, the iTube exhibited a sound marked by elegance, subtlety, and charm. Though it lacked some body and color, and sometimes struggled at high volumes, the iTube had a grainless, extended treble and an overall sound that was engaging and relaxing, said EL. Though it lacked the balanced overall sound of the Simaudio Moon i3.3, the iTube had a midrange to die for, and it effortlessly hung images in the air, summed up EL. The Stingray II is identical to the Stingray iTube, but without the iPod dock. Though it lacked some bass control and low-level resolution, the Stingray II offered a surprisingly dynamic and expansive sound with a relaxed, romantic color, said ST. (Vol.22 No.12, original version, WWW Vol.33 Nos.3, 9, 11, Stingray II WWW) Naim NAIT 5si: 1895 Introduced 30 years after Naim Audios original Nait, the new Nait 5si is a 60Wpc (into 8 ohms) integrated amp with a built-in headphone amp, and with inputs for four line-level sources151but no phono preamp. Inputs are selected with soft-touch buttons, and all are addressed with RCA plugs, while two of those are also equipped with DIN inputs: Naims traditional preference. HR enjoyed the Nait 5si for tracking complex rhythms with perfection, and for keeping even the most microscopic pitch intervals in good focus. The Nait 5si lacked color and spaciousness compared to HRs more expensive tubed integrated, but made it easy to hear151 watch 151the musics rhythms. Forward momentum was spellbinding: The Nait directed my attention toward how the players attacked their instruments. Apart from inverted signal polarity on the headphone output, JA found nothing untoward in his measurements151noting, in fact, that the Nait developed more output power than specified. HR summed up: The Nait 5si is a world-class integrated amplifier that delivers more audio precision and musical enjoyment than any self-respecting anti-imperialist should ever need. (Vol.37 No.10 WWW) Octave Audio V 40 SE: 5300 The entry-level V40 SE Line, from Germanys Octave Audio, is an integrated amplifier with an active line-only preamp section and151despite its model designation151a push-pull output section, using two KT88 pentode tubes per side to produce 40Wpc. (A wide variety of similar pentodes can be used instead, but not all will produce the same amount of power.) The output tubes are operated as true pentodes, and the design entails 10dB of global feedback output-tube bias is user-adjustable via system of which AD remarked, I have never encountered a surer, safer, less ambiguous, or altogether better means of checking and adjusting tube bias. He was similarly impressed by the Octaves musical performance, describing its ability to portray musical timing and momentum as superb. AD also enjoyed the V40 SEs up-front sound and better-than-average sense of scale, also noting that while it didnt sound bright, it had sufficiently extended trebles that reasonable care should be taken in the setup and adjusting of partnering gear. An easy-to-install power-supply enhancement, the Octave Black Box (1200), made an audible improvement, but shouldnt be considered mandatory. JA noted that the V40 SE measured as I would expect from a traditional design that uses a pair of KT88 output tubes for each channel, and praised the amps impressively high standard of construction. (Vol.37 No.8 WWW) Peachtree Audio nova125SE: 1499 The handsome, remote-controlled nova125 combines a 125Wpc amplifier, preamplifier with tubed buffer stage, headphone amp, and asynchronous USB DAC capable of handling resolutions up to 24-bit192kHz. It offers three SPDIF inputs, one analog input, and a preamp output. Though it lacked some resolution, the Peachtree had a slightly warm and soft overall sound, with excellent tonality, well-defined bass, and smooth highs, said ST. Peachtree Audio has delivered a plum, he concluded. Add 100 for Cherry or Rosewood finish. (Vol.36 No.1 WWW) Rogers High Fidelity EHF-100: 7000 Made in the US by former NASA engineer Roger Gibboni, the EHF-100 is rated to deliver 65Wpc (JA measured 35Wpc) into 8 ohms offers four pairs of line-level inputs and uses two EF86 miniature pentode, two 12AX7 triode, and four KT88 power tubes. Fit and finish were excellent. Though not as nuanced, colorful, or dramatic as ADs reference Shindo separates, the EHF-100 distinguished itself as a tight, punchy-sounding amplifier with loads of natural detail, a very good sense of momentum, and an excellent sense of space. Despite differences in the noise floor between its two channels, the EHF-100 measured well for a classic design, said JA. (Vol.35 No.11 WWW) Rogue Sphinx: 1295 The Sphinx is that rarity in contemporary audio: a US-made integrated amplifier with a tubed (12AU7) line stage, a MM-appropriate phono stage, and a headphone jack, all for less than the price of a round-trip ticket to Paris151and Rogue Audio doesnt even make you buy their remote handset (100). Using Bruno Putzeyss Hypex class-D power modules in tandem with a nonswitching power supply, this hybrid integrated delivers 100Wpc into 8 ohms or 200Wpc into 4 ohms. HR enjoyed his time with the Sphinx, noting its greater scale and bass force than his Creek 4330 integrated, and praising its line stage as perhaps the best of its many good features: Everything I played was enjoyably detailed, transparent, and spacious. HRs verdict: Judging by my experiences with the Sphinx, Rogues owner and designer, Mark OBrien, has taken this stigmatized, lower-class mode of operation to a new, more refined level. According to JA, apart from a bit of ultrasonic noise in its output, the Sphinxs amplifier section avoided most of the usual class-D pitfalls, and he particularly praised the MM phono section. (Vol.37 No.8 WWW) Roksan Kandy K2 BT: 1900 This solid-state integrated amp, descended from the original Roksan Kandy K2, is notable for offering both an MM phono stage and a Bluetooth input alongside its three line-level inputs. The Kandy K2 BT, which also provides a pair of preamp-out jacks and a bypass input for AV fans, delivers 140Wpc into 8 ohms and 250Wpc into 4 ohms. In HRs system, The K2 BT showed consistently good tone and scale, but music often felt a tad soft and round151especially at the edges of string plucks and keystrokes. That said, HR declared that the Roksans phono stage was the best Ive heard in a moderately priced integrated151it played LPs in living color. As for Bluetooth, HR was underwhelmed, noting that the detail, life and naturalness exceeded my expectations but I was disappointed by the amount of low-level fuzz and blur I was noticing. He found that Bluetooth music sounded best when playback levels were kept in check. In his measurements, JA echoed that last sentiment151if you use the Roksans BT input, keep your sources volume control down151and commented that the K2 BTs frequency response depends to a larger extent than usual on its volume-control setting. (Vol.37 No.11 WWW) Swissonor V. S.O. P. 5475 The elegant, 8Wpc V. S.O. P. is a push-pull design with three line-level inputs. It uses a pair of ECC81 tubes for the preamp stage, and two pairs of 6V6 power tubes. The V. S.O. P. lacked some of the speed and articulation of the Unison Research Simply Italy, but nevertheless offered a smooth, coherent overall sound, said ST. The V. S.O. P. is an amplifier to hear if you get the chance, especially with high-sensitivity speakers in a small listening space, he concluded. Options: moving-magnet phono stage, add 500 high-output moving-coil phono stage, add 750. (Vol.36 No.2) Unison Research Simply Italy: 2695 10025 The solidly built Simply Italy uses an ECC82 driver tube and an EL34B output tube to deliver 12Wpc. It measures just 10 W by 7.5 H by 15.5 D and offers four line-level inputs, a tape loop, and a single set of outputs optimized for 41508 ohm speakers. Solid-wood inlays around the hefty, stainless-steel volume and selector knobs help damp vibrations. Fit and finish were outstanding. Though it lacked the dimensionality and expansiveness of larger Unison Research amplifiers, the Simply Italy had a confident, solid sound with surprisingly tight bass, said ST. My little bambino. He sums up. Clear, crisp sound, tight bass for all of its 12 watts. ST thinks there may be no better amp for small-group jazz. The Simply Italy was an especially good partner for DeVore Fidelitys Orangutan O93 loudspeakers, said ST. Replacing the amps stock EL34Bs with Genelex KT77 output tubes resulted in an airier sound with less robust bass and improved top-end extension. (Vol.35 No.8 Vol.37 No.1) Arcam FMJ A19: 999 Rated to deliver 50Wpc into 8 ohms, the FMJ A19 is Arcams most affordable integrated amplifier. It provides six line-level inputs, tape and preamplifier outputs, a moving-magnet phono stage, and two front-panel mini-jacks: one for driving headphones, the other for connecting an iPod. While the A19 gets its power from a hefty toroidal transformer, a second internal power supply can deliver a direct, isolated, and regulated 6V to two of Arcams r-series products, such as the rBlink Bluetooth DAC that ST enjoyed. Build quality was excellent, setup simple, operation flawless. Though it lacked some smoothness and drama, the A19 combined a sweet treble, a clean midrange, and well-defined bass for a sound that was fun, involving, and never fatiguing, said SM. JA noted excellent measured performance, but warned that the A19 shouldnt be used to drive at high levels loads much below 6 ohms. Borderline Class B. (Vol.37 No.1 WWW) AVM Inspiration C8: 4190 Made in Germany, the handsome AVM Inspiration C8 is a solid-state CD receiver with a 150Wpc class-D output stage, a phono stage, and digital inputs and outputs, including a USB input limited to 16-bit48kHz resolution. Partnered with the Canalis Anima speakers, the C8 produced a neutral, powerful overall sound, said JM. However, with the speakers placed too low, JM noted an overabundance of high-frequency detail. Expensive for the sound quality on offer. (Vol.36 No.2 WWW) Music Hall a15.3: 549 Rated at 50Wpc into 8 ohms or 75Wpc into 4 ohms, the Music Hall a15.3 is a full-size (16.9 W by 3.1 H by 13.2 D) integrated amplifier that includes a MM phono input, mini-jack and RCA line-level inputs, and a front-panel headphone jack. SM admired the a15.3s high level of fit and finish, adding that the amp performed without flaw in his system. Using the a15.3 with Music Halls companion c15.3 CD player150DAC and comparing it with NADs C 515BEE CD player and C 316BEE integrated amplifier, SM found that the a15.3 sounded more open and airy, and produced a wider, deeper soundstage with beautifully focused images, although it lacked the NADs intangible smoothness and musical flow. Among the inputs offered, none impressed SM more than the a15.3s phono stage, which sounded superb151quiet, dynamic, and emotionally compelling. (Vol.37 No.6 WWW) NAD D 3020: 499 Launched to celebrate NADs 40th anniversary, the 30Wpc D 3020 takes only its name from the companys iconic 3020 integrated amplifier every other aspect of the design has been thoroughly modernized. It uses a switch-mode power supply, lacks a phono stage, and has only a single analog input, but includes a front-panel headphone minijack an optional 6dB bass boost a subwoofer output coaxial, optical, and 24-bit96kHz-capable asynchronous USB inputs and uses an audio-optimized aptX codec for Bluetooth streaming. Weighing just 3 lbs and measuring an unusual 7 38 H by 2 516 W by 8 58 D, the D 3020 can be placed horizontally, like a traditional component, or stood upright, like a modem or hard drive. Uncommonly sensual for a hi-fi product, it has a large, textured volume knob soft, smooth side panels and a touchscreen that occupies its entire front panel and extends through one entire side panel. The sound from every input was warm, present, and naturally detailed even low-quality MP3s streamed wirelessly via Bluetooth were engaging, said SM. Right now, NADs D 3020 is the best bargain in all of hi-fi, added ST. Borderline Class B. (Vol.36 Nos. 11 12 WWW) NAD C 316BEE: 379 10025 Descendant of NADs famed 3020 integrated amplifier, the 40Wpc C 316BEE uses a new variant of the PowerDrive technology found in NADs Master Series components, said to maximize the short-term dynamic power sent to loudspeakers. It has five inputs, a single set of user-friendly binding posts for easy connections, defeatable tone controls, a headphone jack, and an iPod minijack. The NAD matched power with grace, providing a rich, forceful overall presentation and an impressive ability to follow complex musical passages and make clear, truthful distinctions among musical instruments. Compared to the JoLida FX 10, the NAD produced a far more compelling listening experience, with faster attacks, longer decays, and a wider soundstage, said SM. (Vol.34 No.7 WWW) Outlaw Audio RR2150: 699 10025 This 100Wpc, two-channel receiver showcases stylish, deco-like looks and a full range of features that include line, iPod, phono, and USB digital inputs, tape and processor loops, tone controls, headphone output, speaker equalization, bass management, and a mono line-level subwoofer output. JA was astonished to discover what the bargain-basement-priced RR2150 offered, both on the test bench and in the listening room. The RR2150s self-explanatory setup, versatility and convenience, and open, focused, and well-organized overall sound (though somewhat opaque and not fully fleshed out) make it a great intro to hi-fi for a younger generation, said MF. Problems with production led to delivery delays through July 2006, but the situation is now resolved. Current production samples (made in a different factory) offer the same excellent measured performance as the original, but the RR2150s USB digital input, marred by limited resolution and high jitter, should be regarded as being for convenience only, advised JA. (Vol.29 No.3, Vol.31 No.1 WWW) Rega Brio - R . 995 10025 Minimalist in design and appearance, the 50Wpc Brio-R measures just 8.5 W by 3.25 H by 13 D and, like the Rega DAC, is housed in an attractive aluminum-and-steel case with a reflective front panel. It offers five line-level RCA inputs and one phono input. Though it lacked the definition, detail, focus, and frequency extension of the much more expensive LFD LE IV, the Brio-R delivered a relaxed, nonfatiguing sound with tight, full bass, said ST. The Brio-R showed the excellence thats possible when a manufacturer aspires to deliver less, he said, adding that it includes a very good moving-magnet phono stage, enough to satisfy the less fremerous among us. Yes, its umble eye-fye, but very well done, ST sums up. (Vol.34 No.12) Lepai LP7498E: 99.90 Rated to deliver 100Wpc via its TriPath class-D output stage, the Lepai is a small (4.5 W by 1.25 H by 7 D) integrated amplifier with one pair of RCA inputs, two pairs of speaker binding posts, and a dedicated 36V DC power supply. It uses STMicroelectronics TDA7498 class-D module and supports Bluetooth streaming but not aptX. CDs played through the Lepais RCA input sounded big, bold, and emotionally compelling, with a natural midrange, sweet highs, good bass weight, and well-focused images, but digital files streamed via Bluetooth sounded gritty, compressed, and murky, said SM. Sold by Parts Express with a 45-day money-back guarantee and lifetime service warranty. (Vol.37 No.2 WWW) Hegel H160, PS Audio Sprout, Onkyo TX-8020. Allnic T-1500 300B, Audio Note Jinro, not auditioned in too long a time to be sure of rating Devialet D-Premier discontinued.