Recently, James Peng, the founder of Pony.ai, expressed the view that the L3 level in the classification of autonomous driving may remain an unreachable goal. He explained that the classification system, ranging from L1 to L3, was introduced by the United States over twenty years ago. The primary concern at that time was determining liability in the event of safety incidents. From a legal perspective, clearly assigning responsibility for L3 autonomous driving is a complex issue, rendering all L3 systems, in practical terms, equivalent to L2.
Regarding technical strategies, Peng differentiated between the 'world model' and 'end-to-end' approaches. The world model, used as a training method, encounters difficulties in achieving high simulation fidelity and accuracy—a challenge that Pony.ai has been tackling for the past five to six years. On the other hand, the end-to-end approach involves on-vehicle decision-making, which necessitates the verification of intermediate results to prevent the occurrence of black box effects.