Baidu Discloses Apollo World Model for the First Time, Virtual Road Testing Covers 100,000 Kilometers per Day
Baidu Apollo has disclosed for the first time Apollo World, an omni-modal world model for physical AI, and proposed a "2+3" evaluation criteria for world models. The model's core capabilities are agent-driven operation, omni-modal perception, and trip closed-loop, and it can generate camera images and LiDAR point clouds in real time for the driving system to perceive and make decisions and to update the traffic environment. Currently, Apollo World can complete about 100,000 kilometers of virtual road testing per day and has been used in the global testing and validation of Apollo Go.
Why read this
AI score 58
Baidu publicly discloses for the first time the capability composition and evaluation criteria of Apollo World, allowing readers to understand how the world model supports virtual road testing for autonomous vehicles.
Topics
Body · AI translation
On November 3, Baidu Apollo disclosed for the first time Apollo World, an omni-modal world model for physical AI, and proposed a "2+3" evaluation standard for world models. The model is currently being used for the global testing and validation of Apollo Go.
It is reported that the core capabilities of Apollo World include agent-driven operation, omni-modal perception, and a closed-loop trip. In the virtual city it constructs, traffic participants such as vehicles and pedestrians are modeled as AI agents with decision-making and interaction capabilities, and can adjust their behavior according to changes in the surrounding traffic and the actions of the robotaxi, such as deceleration and overtaking. The model can generate multi-modal observation data such as camera images and LiDAR point clouds in real time for the driving system to perceive and make decisions, and then update the traffic environment according to the vehicle's actions, forming a continuously running closed loop.
In terms of test scope, Apollo World supports robotaxis traveling between any origin and destination in the virtual city, continuously passing through different roads and traffic scenarios, and completing the full trip task from accepting an order and carrying passengers to drop-off. At the same time, the model can extract high-value scenarios from real operating data, turning complex road conditions and risks into repeatable virtual tests for driving strategy validation and system iteration, and the updated system then returns to real roads for verification.
The "2+3" evaluation standard proposed by Baidu this time lists causality and interactivity as basic thresholds, and lists physical accuracy, persistence, and real-time performance as advanced indicators, emphasizing that models should follow real physical laws, respond to external actions, and maintain long-term stable operation with low latency.
According to Baidu's disclosure, Apollo World can complete about 100,000 kilometers of virtual road testing per day, equivalent to about three years of continuous testing by a real vehicle equipped with a safety officer. Apollo Go has entered 28 cities worldwide, completed more than 23 million trips in total, and accumulated more than 350 million kilometers of autonomous driving mileage, of which more than 240 million kilometers were fully driverless.
Source: Baidu
Submissions/Tips: tougao@youxituoluo.com
Article/Business Cooperation: Lin Nan (WeChat 19250561593)
Join the industry exchange group: Lin Nan (WeChat 19250561593)
Digital technology industry information and service platform
Download the "Tuoluo Technology" APP to get cutting-edge, in-depth digital technology industry information