Learned Perceptive Forward Dynamics Model on arXiv

Our paper Learned Perceptive Forward Dynamics Model for Safe and Platform-aware Robotic Navigation is now available on arXiv, and is published at Robotics: Science and Systems (RSS) 2025.

The model predicts the robot’s future state and failure probability from the surrounding geometry and proprioceptive history, and is used inside an MPPI planner for safe navigation on ANYmal. Joint work with Jonas Frey, Cesar Cadena and Marco Hutter.

Pascal Roth
Pascal Roth
Ph.D. in Robot Learning & Software Engineer

Research on learning-based navigation and autonomy for legged robots, with focus on perceptive planning and sim-to-real transfer.