Autonomous driving technology is undergoing a gradual evolution from assisted driving to fully autonomous driving. According to SAE's classification standards, autonomous driving is divided into six levels, L0 to L5. Currently, mass-produced vehicles are mainly in the L2 (partially autonomous driving) to L2+stage, and L4 level autonomous driving has begun commercial operation in limited areas.
L2 level assisted driving is the most widely used autonomous driving technology, including adaptive cruise control (ACC), lane keeping (LKA), and automatic emergency braking (AEB) functions. These functions perceive the surrounding environment through cameras, millimeter wave radar, and ultrasonic sensors, providing assistance to drivers on highways and urban expressways. The system can significantly reduce driving fatigue, but the driver still needs to be monitored throughout the entire process.
L4 level autonomous driving achieves fully autonomous driving in limited areas without the need for human intervention. Autonomous taxis are the most typical L4 application scenario. Baidu Apollo, Xiaoma Zhixing and other companies have launched Robotaxi commercial operations in multiple cities in China, with a cumulative service mileage of tens of millions of kilometers. AI deep learning models recognize vehicles, pedestrians, traffic signs, and road boundaries in real-time from multi-sensor data, and the end-to-end architecture maps sensor inputs directly to driving decisions.
Data loop is the key to the continuous evolution of autonomous driving. The massive data collected during fleet driving is filtered, annotated, and trained to continuously optimize model performance. With the introduction of the large model, the auto drive system has shown stronger scene understanding ability and long tail scene processing ability.