Baidu wants AI to teach its robotaxis to drive. The company launched Apollo World on Nov. 3, a world model that simulates road environments and predicts how they change. Reuters reports Baidu says it can produce «up to 100,000 km» of simulated driving every day, about three years of real-world testing by a car with a safety driver.
What it does
- Sensor data: Apollo World generates real-time multimodal data streams, including camera and lidar observations, so a self-driving system can sense and drive through the simulated world directly.
- Reactive traffic: IThome, reporting from the Apollo session of Baidu's AI Day in Shenzhen, says other road users are simulated as agents that respond to the robotaxi rather than following scripts, and that the system can run a whole trip from pickup to drop-off (translated from Chinese).
- Already in use: Baidu says Apollo World is being used to train Apollo Go's robotaxi fleet; IThome says it is used in Apollo Go's global testing and validation (translated).
Why it matters
Apollo Go is China's largest robotaxi operator. IThome reports it now covers 28 cities worldwide and has passed 350 million km of cumulative driving, more than 240 million km of it fully driverless (translated). Reuters notes that rivals such as WeRide, Pony.ai and Momenta are expanding too, that XPeng plans robotaxi operations in 2027, and that Morgan Stanley estimates China's robotaxi fleet could reach 1 million vehicles by 2035.
Simulation is one way to cut the cost of that race. Training models and building test fleets remain expensive, and every simulated kilometre is one that does not need a car, a driver and a road.
