{"slug": "baidu-tests-a-new-way-to-train-robotaxis", "title": "Baidu Tests a New Way to Train Robotaxis", "summary": "Baidu's Apollo unit unveiled Apollo World, an agent-driven multimodal world model for physical AI, at its AI Day event in Shenzhen on Nov. 3, and the company said the model is already being used to test and validate the Apollo Go driverless ride-hailing service as it expands into more markets. Apollo World creates a closed-loop virtual environment in which AI-driven cars and pedestrians react to the vehicle while simulated camera and lidar data feed back into the AI driver, aiming to cut road testing that currently averages more than 700,000 kilometers per model iteration; the model was trained on more than 1 billion high-value frames. \"A world model is not a painter that generates videos, but an examiner of autonomous-driving safety,\" said Liang Wang, chief architect of Baidu's Intelligent Driving Group. Apollo Go reports more than 23 million rides and 240 million fully driverless kilometers, with fully driverless commercial rides launched in Dubai and testing or expansion underway in London, Switzerland, Hong Kong, Abu Dhabi and South Korea.", "body_md": "**SHENZHEN, China — Nov. 3:** Baidu’s Apollo unit unveiled Apollo World, an agent-driven multimodal world model for physical AI, at its AI Day event in Shenzhen.\n\n- The model is already being used to test and validate Apollo Go as the driverless ride-hailing service expands into more markets, according to the company.\n\n**Why it matters:** Apollo defines a world model by two requirements: causality and interactivity — the future follows the past, and the world responds to the vehicle’s actions.\n\n- “A world model is not a painter that generates videos, but an examiner of autonomous-driving safety,” said Liang Wang, chief architect of Baidu’s Intelligent Driving Group.\n\n**What happened:** Apollo World creates a closed-loop virtual environment where AI-driven cars and pedestrians react to the vehicle, while simulated camera and lidar data feed back into the AI driver.\n\n- It can test complete trips, not just isolated maneuvers.\n  - “Apollo World focuses more on closed-loop simulation of complete tasks,” said Ze Yang, Baidu IDG distinguished architect and world-model lead.\n\n- The system aims to cut costly road testing.\n  - Each model iteration requires, on average, more than 700,000 kilometers of road testing. Apollo World was trained on more than 1 billion high-value frames.\n- The closed loop uses both camera and lidar data.\n  - In response to a *TechNode* question, Wang said the model fuses the two, while Yang said combining them gives the system more information than vision alone.\n- In response to a \n- Safety is a key goal. Wang said virtual testing helps catch problems before less-predictable end-to-end models are deployed at scale.\n\n**The big picture:** Robotaxi competition is increasingly playing out across markets, with companies racing to adapt driverless systems to different cities, regulations and driving environments.\n\n- Apollo Go built its base in China, where its mainland operations are now 100% fully driverless.\n- It has logged more than 23 million rides and 240 million fully driverless kilometers across its operations.\n- Overseas, it has launched fully driverless commercial rides in Dubai, with testing or expansion underway in London, Switzerland, Hong Kong, Abu Dhabi and South Korea.\n- The push comes as Waymo, WeRide and Pony.ai also expand abroad, intensifying competition to establish robotaxi services in new markets.", "url": "https://wpnews.pro/news/baidu-tests-a-new-way-to-train-robotaxis", "canonical_source": "https://technode.com/2026/11/05/baidu-tests-a-new-way-to-train-robotaxis/", "published_at": "2026-11-05 06:54:18+00:00", "updated_at": "2026-11-05 07:20:36.262248+00:00", "lang": "en", "topics": ["autonomous-vehicles", "artificial-intelligence", "ai-agents", "robotics"], "entities": ["Baidu", "Apollo", "Apollo Go", "Apollo World", "Liang Wang", "Ze Yang", "Waymo", "Pony.ai"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/baidu-tests-a-new-way-to-train-robotaxis", "markdown": "https://wpnews.pro/news/baidu-tests-a-new-way-to-train-robotaxis.md", "text": "https://wpnews.pro/news/baidu-tests-a-new-way-to-train-robotaxis.txt", "jsonld": "https://wpnews.pro/news/baidu-tests-a-new-way-to-train-robotaxis.jsonld"}}