Autonomous Driving Technology

map-first · vision-only · fusion · suppliers

Self-driving systems differ on two choices: whether they need a prior HD map, and whether they carry lidar. The industry is converging on end-to-end neural networks, and the remaining split is over sensors and maps.

Driving systems placed by HD-map dependency (horizontal) and sensor set (vertical). Map-dependent with lidar, radar and cameras: Waymo, Baidu Apollo Go and Mercedes-Benz DRIVE PILOT. Little or no map dependency with lidar, radar and cameras: Huawei ADS, Rivian Gen 3, Li Auto and NIO. Cameras and radar with light map use: Momenta, and Mobileye SuperVision with its crowdsourced REM map. Cameras and radar with no HD map: XPeng. Cameras only with no HD map: Tesla FSD and Wayve. Chinese systems are shown in red, Western in blue. DRIVING SYSTEMS BY MAP DEPENDENCY AND SENSOR SET No HD map HD map required map dependency Lidar, radar, cameras Cam, radar Cameras Waymo Apollo Go DRIVE PILOT Huawei ADS Rivian Gen 3 Li Auto, NIO Momenta Mobileye SuperVision XPeng Tesla FSD Wayve
Driverless leaders
Waymo, Apollo Go
Vision-only
Tesla, Wayve
Largest supplier
Huawei (China)
Hub
Autonomy

Finding

Every major driving stack is moving to end-to-end neural networks; what still separates them is maps and lidar

Waymo and Baidu Apollo Go run the largest driverless fleets on a map-first design with lidar. Tesla runs camera-only end-to-end networks with no HD map and has driverless service only in Austin. Huawei, Rivian, Li Auto and NIO sit between: lidar-rich sensing feeding a learned model without a prior map.

The map-first design gives the clearest safety case because the operational design domain (ODD) is drawn on a map, but every new city has to be mapped and kept current. Vision-only systems avoid that cost and train on the largest fleets (Tesla has several million cars collecting data), but a single network is hard to verify formally. Chinese makers have pushed further into vision-language-action models that add a language-model reasoning layer, and they iterate fastest because their domestic fleets are large and data stays in China.

Suppliers matter as much as automakers. Huawei's Qiankun ADS ships on more than a dozen brands without Huawei building a car; Momenta, Horizon Robotics and DeepRoute.ai supply most other Chinese brands; Nvidia, Mobileye and Qualcomm supply most Western ones. US rules adopted in January 2025 bar Chinese and Russian automated-driving software from US vehicles starting with model year 2027, which splits the supplier market in two. Texas robotaxi deployments are tracked on AustinIO.

Approaches

The five architectures

Autonomous driving architectures
ApproachPractitionersHow it worksStrengthConstraint
HD map-firstWaymo, Baidu Apollo Go, Pony.ai, WeRide, Mercedes DRIVE PILOTLidar, cameras, radar plus a centimeter-level prior mapClearest safety case inside a mapped ODDMap building and upkeep per city; Waymo runs in 15 US cities
Vision-only end-to-endTesla FSD, Wayve, comma.aiCameras; one neural network from pixels to controlsLowest hardware cost; largest training fleetHardest to verify; Tesla driverless only in mapped Austin zones so far
Fusion with learned modelRivian Gen 3, Huawei ADS, Li Auto, NIO, XiaomiLidar, cameras, radar feeding an end-to-end model, no HD mapRedundant sensing grounds training dataHigher per-vehicle cost
Chinese vision-language-actionXPeng VLA, Li Auto VLA, Momenta, DeepRoute.aiVision-first end-to-end plus a language-model reasoning layerFast iteration on large domestic fleetsData-export limits outside China; US connected-vehicle rule
Full-stack supplierHuawei Qiankun ADS, Nvidia DRIVE, Mobileye, Momenta, Horizon RoboticsChip, software and tools sold to many automakersScale across brands; shared fleet dataExport controls in both directions

Players

Who builds what

Autonomous driving developers and suppliers
CompanyBaseApproachStatus, October 2026
WaymoUSMap-first, lidarDriverless in 15 US cities; 6th-generation driver on Zeekr-built Ojai and Hyundai IONIQ 5
Baidu Apollo GoChinaMap-first, lidarDriverless in 27 cities including Wuhan, Beijing, Dubai and Abu Dhabi
Pony.ai, WeRideChinaMap-first, lidarDriverless robotaxis in Chinese cities, the Gulf and Europe
Tesla FSDUSVision-only end-to-endFSD (Supervised) v14 on owner cars; driverless robotaxi in Austin since January 2026
Huawei Qiankun ADSChinaFusion, full-stack supplierADS 4 generation on HIMA brands (AITO, Luxeed, Stelato, Maextro) and others
XPengChinaVision-first, VLATuring chip in-house; supplies Volkswagen China models; robotaxi trials in 2026
MomentaChinaSupplier, end-to-endSupplies SAIC-GM, Mercedes-Benz China, Toyota and others; Uber robotaxi partner in Europe
RivianUSFusion with lidarGen 3 hardware with RAP1 chip and lidar on R2 from late 2026
Mercedes-Benz DRIVE PILOTGermanyMap-first, lidarLevel 3 on mapped highways in Germany, California and Nevada
MobileyeIsrael, USCamera-first with REM crowd mapSuperVision on Zeekr and others; Mobileye Drive in Volkswagen ID. Buzz robotaxis
WayveUKVision-only end-to-endNissan ProPILOT integration from 2027; Uber trials in London
Nvidia DRIVEUSCompute and software supplierThor platform and Alpamayo models; Uber partnership for robotaxis from 2027

Sources

Sources and scope

Scope
Autonomous Driving Technology
Sources
Waymo; Baidu; Tesla; Huawei; XPeng; Momenta; Rivian Autonomy Day (December 2025); Mobileye; Wayve; Nvidia; US Commerce Department connected-vehicle rule (January 2025)
Updated
October 2026