Dr. Zhan Xianyuan from Tsinghua AIR: Data intelligence is the core of autonomous driving iteration

Recently, Dr. Zhan Xianyuan, Assistant Researcher/Assistant Professor of Tsinghua AIR, explained the characteristics and current progress of end-to-end autonomous driving AI algorithms from the perspective of decision optimization in a high-quality open course on autonomous driving jointly organized by Mimo Zhixing and Tsinghua University's Intelligent Industry Research Institute (AIR)

Recently, Dr. Zhan Xianyuan, Assistant Researcher/Assistant Professor of Tsinghua AIR, explained the characteristics and current progress of end-to-end autonomous driving AI algorithms from the perspective of decision optimization in a high-quality open course on autonomous driving jointly organized by Mimo Zhixing and Tsinghua University's Intelligent Industry Research Institute (AIR). He emphasized that data intelligence is the core of the entire autonomous driving iteration.

Dr. Zhan Xianyuan first explained the meaning of end-to-end autonomous driving AI algorithm and pointed out that this algorithm integrates all architectures and different modules into a complete whole, training directly from input to output, and learning signals are transmitted forward from decision-making. The end-to-end model can combine all modules together to form a super large model, with all objectives optimized and learned around the ultimate goal of decision-making.

It is understood that Dr. Zhan Xianyuan also detailed that the end-to-end driving model can be seen as a large-scale decision-making model that requires training using decision optimization algorithms. This involves two methods: imitation learning and reinforcement learning. Imitation learning is the method of directly training mappings from data through supervised learning; Reinforcement learning is not simply imitating data, it seeks to find a better decision-making mode through continuous learning and optimization than existing data.

At the end of the speech, Dr. Zhan Xianyuan emphasized that data intelligence plays a crucial role in autonomous driving iterations. By using end-to-end algorithms and decision optimization algorithms, we can better utilize massive amounts of data and improve the safety and generalization of autonomous driving. (Jiang Miao)