Baidu unveiled Apollo World at its AI Day·Apollo event in Shenzhen on Tuesday, Nov. 3, describing it as a world model for training and validating Apollo Go robotaxis, with a daily simulation capacity of up to 100,000 kilometers (62,000 miles). Baidu says Apollo Go is already using it in testing and validation work around the world. In the simulated settings, vehicles and pedestrians act as AI agents that make decisions and respond to traffic changes, and Baidu says Apollo World can generate camera and lidar observations in real time and represent a continuous trip from pickup through drop-off. Real-road data helps identify situations for virtual testing, and results are fed back into the driving system. Baidu likened a day of simulated mileage to the road testing a single autonomous vehicle with a safety driver might complete in about three years. At the event, Baidu outlined a “2+3” framework for world models, with causality and interaction as two baseline qualities and physical accuracy, persistence and real-time performance as three additional measures. As of the event, Apollo Go had reached 28 cities and served more than 23 million rides, and had accumulated more than 350 million kilometers of autonomous driving, including more than 240 million kilometers with no driver in the vehicle.