Baidu unveils Apollo World model to simulate robotaxi operations
Baidu has launched Apollo World, an AI model designed to simulate complex driving environments and reduce the need for physical road testing of autonomous vehicles.
Baidu’s Apollo unit has introduced Apollo World, a multimodal world model intended for physical AI development. The system creates a closed-loop virtual environment that allows autonomous vehicles to interact with pedestrians and other vehicles, providing feedback based on both camera and lidar data.
By focusing on the simulation of complete trips rather than isolated maneuvers, the company aims to accelerate the safety validation process for its Apollo Go service. Historically, Baidu has required over 700,000 kilometers of physical road testing for each model iteration. The new system is trained on over one billion data frames to catch potential issues before deployment in real-world scenarios.
The initiative comes as Baidu scales its driverless operations beyond China. With over 240 million autonomous kilometers logged, the firm is currently testing or operating in markets including Dubai, London, and South Korea.
Baidu is operating in an increasingly crowded robotaxi market, facing global competition from firms such as Waymo, WeRide, and Pony.ai. The race to adapt driverless technology to diverse international regulatory environments remains a primary driver for investment in simulation capabilities.