Go back to projects

Lightweight World Model and Planning for Autonomous Driving

#Autonomous Driving #BEV #World Model #Planning

Summary

This project explores a lightweight conditional world model for autonomous driving with a strong focus on downstream planning.

Technical Direction

The current research path includes:

  • multi-camera BEV representation learning,
  • temporal world modeling with ConvLSTM,
  • staged training toward end-to-end joint modeling for perception and planning.

My Role

I serve as a core research member and contribute to both algorithm exploration and system-level design. The work is positioned around building a practical planning-oriented stack rather than an isolated perception benchmark.