Baidu Launches Apollo World AI Model to Train Robotaxis in Simulated Road Environments

The Chinese search and AI company says the system can generate up to 100,000 kilometres of simulated driving a day. Baidu puts that at roughly three years of real-world testing by an autonomous vehicle with a safety driver on board. The comparison is Baidu's own and has not been independently verified.

Apollo World is a world model, meaning it reproduces a driving scene and also predicts how that scene will change. It generates real-time sensor data, including camera and lidar observations, which self-driving software can use to perceive and navigate the simulated surroundings. According to Baidu, the system is already training the fleet of Apollo Go, its robotaxi service, which it intends to run in markets around the world.

Why Simulation Matters

The launch goes to the central economic problem in autonomous driving. Chinese robotaxi companies have mostly chosen asset-light models, supplying technology and partnering with carmakers and fleet operators rather than owning large fleets themselves. That keeps vehicle spending off their balance sheets, but developing the driving software is still expensive. "They still have the capex of training models," Frank McCleary of the consultancy Arthur D. Little told Reuters, adding that every autonomous vehicle company faces this challenge.

Simulation is one way to bring that cost down. It reduces the real-world mileage needed and lets developers train systems on rare or dangerous situations that are hard to stage on public roads. One question raised in coverage of the launch is how closely simulated roads will match the different traffic and road conditions in each city where Apollo Go plans to operate.

Competition in China is growing. Apollo Go is the country's largest robotaxi operator, and its rivals include WeRide, Pony.ai and Momenta Global. XPeng, Volkswagen's Chinese partner, plans to launch robotaxi operations in 2027. Reuters cited a Morgan Stanley estimate that China's robotaxi fleet could reach 1 million vehicles by 2035.

The Gulf Connection

Much of Apollo Go's international expansion has run through the UAE. In November 2025, Abu Dhabi's Integrated Transport Centre gave it the emirate's first fully driverless commercial permits. In January 2026 it started a public fully driverless service on Yas Island with local partner AutoGo. In Dubai, it began fully driverless commercial rides with Dubai Taxi Company at the end of March 2026. In August, Uber began matching riders in parts of Jumeirah and Umm Suqeim with its vehicles.

Reuters did not say how Apollo World is being used in individual markets, including the UAE. Still, Baidu's statement that the system is training the Apollo Go fleet puts it at the centre of a service that Gulf transport authorities are relying on to expand driverless mobility.

The launch comes as robotaxi developers in China, the US and the Gulf race to scale up fleets while keeping costs under control. If Baidu's simulation claims hold up, they could shorten the testing needed to enter new cities, which matters to the transport regulators and fleet partners now deciding how quickly to approve and expand driverless services.