Baidu Apollo World Generates 100,000 Kilometres of Robotaxi Simulation Data Daily
Baidu has launched Apollo World, an artificial intelligence system designed to simulate real-world road environments and train autonomous vehicles. The platform can generate up to 100,000 kilometres of simulated driving each day, a volume the company says is equivalent to about three years of real-world testing by an autonomous vehicle equipped with a safety driver.
The launch gives Baidu another tool in its effort to expand robotaxi operations while reducing the time and expense involved in developing self-driving technology. Apollo World is already being used to train the company’s Apollo Go fleet, which Baidu says is intended for operation in international markets.
How Apollo World trains robotaxis
Apollo World creates virtual road environments and predicts how those environments change. Rather than relying only on recorded driving footage, the system can produce real-time, multimodal data streams, including camera and lidar observations. These inputs allow autonomous-driving systems to perceive and navigate simulated surroundings in a way that resembles the sensor data they encounter on actual roads.
Baidu says the platform can generate up to 100,000 kilometres of simulated driving per day. The comparison with three years of road testing illustrates the potential scale of the system, but it is a company-provided equivalence rather than evidence that simulated kilometres can replace real-world safety testing.
Why simulation matters to the robotaxi industry
Developing autonomous vehicles remains expensive, even as robotaxi companies increasingly pursue partnerships instead of owning and operating every vehicle themselves. Chinese firms have often supplied driving technology to automakers, ride-hailing platforms and local fleet operators in overseas markets. That approach can reduce fleet-related costs, but companies still need to invest heavily in training models and developing vehicles.
Frank McCleary, an autonomous mobility and urban transportation expert at consulting firm Arthur D. Little, told Reuters that the cost of training models and vehicles remains a challenge across the industry. He also said a system’s ability to adapt to and learn a new driving environment affects how quickly a service can launch.
Simulation could help companies expose their software to a wider range of road situations before deploying vehicles in a new city. However, the scale of simulated data alone does not establish how accurately those environments reflect real traffic or whether training on them improves safety.
Baidu faces a growing robotaxi market
Apollo Go is China’s largest robotaxi service operator, Reuters reports, and competes with companies including WeRide, Pony.ai and Momenta. Baidu says Apollo World is already supporting training for the fleet as it prepares for wider global operations.
The competition is also drawing in established automakers. Volkswagen’s Chinese partner XPeng plans to launch robotaxi operations in 2027. Morgan Stanley has estimated that China’s robotaxi fleet could reach one million vehicles by 2035, although that figure is a long-term forecast rather than a current market count.
Baidu has not provided independent performance results showing how Apollo World’s simulated training compares with conventional road testing. Its launch therefore highlights a potentially important development in autonomous-driving infrastructure, while leaving questions about validation and real-world safety outcomes open.
