XPeng Refutes Rumors Regarding Assisted Driving R&D and Confirms VLA Model Progress
XPeng Motors has officially refuted online rumors regarding the development progress of its assisted driving systems for certain car models. The company confirmed that the deployment of its second-generation Vision-Language-Action (VLA) model distillation onto vehicles is proceeding orderly as planned. This clarification reassures the market about XPeng's autonomous driving roadmap, particularly the integration of advanced VLA models. Successful deployment of distilled VLA models could significantly improve real-time decision-making and traffic efficiency in production vehicles. XPeng's second-generation VLA model reportedly improves traffic efficiency by 30% compared to traditional L2 systems and Robotaxis, and can recognize complex scenarios like traffic police gestures. The company aims to adapt this unified model for both Chinese and European markets, targeting overseas deliveries by 2027.
## BACKGROUND
A Vision-Language-Action (VLA) model is a multimodal AI that integrates visual perception, language understanding, and action outputs to control robotic systems or autonomous vehicles. Knowledge distillation is a technique used to compress these large, resource-intensive models into smaller, faster "student" models suitable for real-time execution on vehicle hardware.