Baidu has launched Apollo World, an AI system that builds simulated versions of the real world to train robotaxis. The company says the system is already being used to train its Apollo Go fleet for global operation, turning the race to deploy driverless cars into a race to build better virtual worlds.
What Apollo World Does
Launched on Tuesday, November 3, 2026, Apollo World simulates real-world environments for training robotaxis and predicts how those environments evolve. It produces real-time, multimodal data streams, such as camera and lidar observations, so self-driving systems can directly sense and navigate the simulated world rather than simply watching a recording of it.
Why Simulation Matters
Baidu says Apollo World can generate up to 100,000 km (62,000 miles) of simulated driving every day, which it says equals about three years of real-world testing by a safety-driver-equipped autonomous vehicle. That scale matters because real road testing is slow, costly, and limited by weather, traffic, and geography. In simulation, engineers can replay rare and dangerous situations again and again, test how a vehicle reacts to a new city layout, and gather training data without putting a test car on a public road for every mile.
The Global Robotaxi Race
Apollo Go is already China's largest robotaxi service operator, competing with rivals including WeRide, Pony.ai and Momenta Global. The contest is also widening beyond China: XPeng, Volkswagen's Chinese partner, plans robotaxi operations in 2027. Morgan Stanley has estimated that China's robotaxi fleet could grow to 1 million vehicles by 2035, a scale that would make fast, reliable training a decisive advantage.
Why It Matters
Developing this technology remains expensive, and launch speed comes down to how fast the technology can adapt to and learn a new driving environment, according to an expert from Arthur D. Little. Simulation directly targets that bottleneck: a fleet can learn a new city's roads, rules, and edge cases in a virtual world before it ever carries a passenger there. The wider shift is visible across AI, where large AI models are increasingly trained on vast amounts of data, and where other teams are teaching robots to read and understand their surroundings.
The Take
Apollo World does not replace real-world testing, but it changes what real-world testing is for. If Baidu's claims hold up, the hardest part of going global may no longer be collecting miles, but building a simulation faithful enough to trust them. That is why the robotaxi race is now, in large part, a simulation race.
Comments
Post a Comment