XPeng Shifts to Physical AI Architecture, Aims for L4 Autonomy by 2027 and Mass Production of Humanoid Robots

XPeng has made a major strategic shift in its autonomous driving technology roadmap, abandoning its previous system that combined software engineering with AI algorithms and halting R&D efforts in that direction—into which it had already invested tens of billions of yuan last year. The company is now betting on a new "Physical AI" architecture, believing this approach is key to achieving true driverless capability. He Xiaopeng, Chairman of XPeng, stated that, based on current progress, the company expects to achieve Level 4 (L4) autonomous driving within 18 to 24 months—by the end of 2027. He emphasized that autonomous driving is just one critical application scenario for Physical AI. XPeng is now building its future around Physical AI as the foundation, identifying smart EVs, robotics, and global business as its three core growth pillars over the next decade. In robotics, XPeng is focusing on general-purpose humanoid robots, having restructured its team to adopt an AI-driven approach for perception, decision-making, and motion control—moving away from traditional software engineering methods. He predicts that 2027 could mark the first year of commercial mass production for advanced humanoid robots, noting that as the core embodiment of Physical AI, robots possess high software value and, once breakthroughs in capability are achieved, could scale even faster than the automotive industry. Furthermore, He Xiaopeng forecasts that starting in 2027, Physical AI will begin significantly impacting jobs across multiple industries, potentially replacing human labor to some extent.