Anonymous Intelligence Signal

Auto-pilot of the L3 level technical route dispute: a progressive and cross-cutting strategic game

ai The Lab unverified 2026-03-26 18:48:20 Source: 自动驾驶行业

From a technical point of view, the evolutionary nature of L2 to L4 is a leap in perception, decision-making, implementation of the three core modules, and a continuous succession of bottom technologies, such as data closures, hardware redundancy. The need for L3 autopilots as a transition phase linking L2 and L4 are being challenged in two directions. At the technical level, L2 auxiliary driving is centred on "Auxiliary Implementation" with hardware using a single chip, a single perceptive link programme with a perception range of about 100 metres and a positioning accuracy of up to a metre level. L3 has landscape control capability, hardware is upgraded to double chip, double braking, two-way configuration, laser radars are introduced, knowledge is extended to 150-200 metres, and positioning accuracy is up to the degree of metre. L4 achieves full-scale autonomous decision-making, hardware has a fully redundant structure with a perceived distance of 250-300 centimetres and positioning precision up to cm. Mainstream car companies generally believe that L3 is a mandatory route from L2 to L4. L4 is completely redundant, full-scale scenario algorithms cannot be fully validated through the laboratory, technology must be disassemblyded and data must be accumulated through L3 mass production sites. Core technical challenges such as multisensor integration, optimization of decision algorithms, etc. require gradual validation and breakthroughs in real scenes. However, the technologists, represented by Tesla and Waymo, believe that L3 is a "premature compromise". To accommodate the human take-over needs, the L3 algorithm needs to set aside the trigger logic for taking over, and the redundancy system needs only to meet the basics. This design prevents L3 technology accumulation from being directly reused to L4 and requires substantial modifications and optimization. The Tsla FSD system already has autonomous decision-making capacity close to L4, which allows for a direct leap in L2+ to L4. At the policy level, China is moving forward with the L3 Autopilot Access Pilot. In late 2025, the Ministry of Trade and Communications officially approved the first L3 autopilot pilot. The mandatory standards of the Automated Driving Data Recording System were formally implemented on 1 January 2026, and key standards, such as the Safety Requirements for Automatic Driving Systems, are in the drafting phase. Industry observations suggest that the need for L3 depends on the enterprise’s own technological path, research and development strength and data accumulation. Mainstream firms lack the capacity to develop Tesla and Waymo, and L3 remains a sound path to technology leapfrogging; head technologists can move L4 directly through L2+.