WAYMO-自动驾驶基础模型-2025-09-自动驾驶.docxVIP

FoundationmodelsForautonomousdriving

VincentVanhoucke

DistinguishedEngineerWaymo

p.2

OURMISSION—

Betheworld’s

mosttrusteddriver.

p.3

UpnextAtlantaMiamiD

Upnext

Atlanta

Miami

D.C.

p.4

Seattle,WA

Truckee,CA

SanFranciscoBayArea,CA

DeathValley,NV

LasVegas,NV

LosAngeles,CA

Phoenix,AZSanDiego,CA

Phoenix,AZ

2025Roadtripcities

ComingsoonComingsoonComingsoonComingsoonComingsoon

Comingsoon

Comingsoon

Comingsoon

Comingsoon

Comingsoon

Comingsoon

WaymooperatingcitiesOperationcomingsoonWaymotestedcities

UpperPeninsula,MI

UpstateNYBuffalo,NY

●

MetropolitanDetroitArea,MI

MetropolitanDetroitArea,MI

WashingtonDC

Nashville,TN

Atlanta,GATokyo,Japan

Austin,TXNewOrleans,LA

Miami,FL

p.8

STEP2What’saround

STEP2

What’s

aroundme?

STEP1

WhereamI?

STEP3

Whatwill

happennext?

What

shouldIdo?

p.9

p.9

Confidentialandproprietaryp.10

ForeignObjectsonthe

ForeignObjectsontheRoad

Unusua

UnusualBehaviors

ToddlersontheLooseFallingSkateboardersStray

Toddlers

ontheLoose

Falling

Skateboarders

Stray

Cyclists

ABBQGrillFallingOff

Truckdrivingscenariospresents

Truck

significantchallenges

Unique

UniqueInteractions

Ext

ExtremeWeather

RoadLastMinuteLaneChangesFallingTrees

Road

LastMinuteLaneChanges

Blockages

DrivingasaConversation

Trajectoriesassentencesinanewlanguage

Vocabularyconsistsofstate/motionwords(vectors)

Likelanguage,trajectorieshavelocalcontinuityandglobalcontextModelarchitectureissimilartothatofaLLM

Motiontokensequence:

t=1t=2t=3t=4

MotionLM:Multi-AgentMotionForecastingasLanguageModeling

AriSeff,BrianCera,DianChen,MasonNg,AurickZhou,NigamaaNayakanti,KhaledS.Refaat,RamiAl-Rfou,BenjaminSappICCV2023

Examples

MARGINAL

MARGINALPREDICTION

p