Baidu unveils ApolloWorld driving world model: 100,000 km of virtual road tests per day
机器之心 · wechat · 2026-11-04
Baidu Apollo has disclosed ApolloWorld, an embodied world model that lets robotaxis train inside AI-generated virtual cities where vehicles and pedestrians are agents with their own goals. Analysts estimate it enables 100,000 km of virtual road testing per day — roughly three years of real-world safety-driver testing.
Technically: traffic participants are modeled as AI agents that react in real time to the ego car's actions; the causal autoregressive multimodal generator produces synchronized camera frames and LiDAR point clouds; roads can be re-skinned (rainforest, desert, snow, tunnel) on demand, full robotaxi trips can be simulated end-to-end, and virtual fleets can run in parallel. Baidu proposes a "2+3" standard (causality, interactivity as the baseline; physical accuracy, persistence, real-time performance as gains). Compared with Waymo's Genie 3-based world model, which excels at exotic scenarios, ApolloWorld emphasizes agent autonomy and complete task flows, grounded in Apollo Go's real data from 28 cities, 23M rides and 350M km — a data flywheel between real roads and virtual tests.
Related event: Baidu Unveils ApolloWorld Physical AI World Model for Autonomous Driving(4 posts)→
More from Embodied
- NYU team open-sources MABEL, an $8K mobile bimanual robot built in 2 weeks — RemiCadene · 2026-11-05
- Turning an AirPods case into a working mouse — TheMoonMidas · 2026-11-05
- Tsinghua, NYU and Northwestern unveil RoboDoctor, a training-free framework for fixing robot actions — jiqizhixin · 2026-11-05
- Robotics exec says deployments start in Singapore, aiming for everywhere from cafes to factories — micoolcho · 2026-11-05
- Gus Robotics watches its Microduck robot production line run for the first time in Shenzhen — mishig25 · 2026-11-05
- Smartglasses CV tells flower pickers what to harvest, cutting seasonal training to zero — nptacek · 2026-11-05