World-Model-Based Driving Skill Learning for Humanoid Robots and Real-Vehicle Validation in a V2X Testbed
A personal NRF-funded project that I lead as principal investigator. Rather than making the vehicle intelligent, it replaces the driver: a humanoid robot operates the steering wheel and pedals of an ordinary car. The driving policy is learned through world-model–based reinforcement learning in a coupled robot–vehicle simulation, then transferred to a real humanoid and validated on an actual vehicle at the KAIST V2X testbed, where the robot itself acts on the infrastructure's cooperative messages.