Robots should be able to adapt to varying terrain and robot properties for efficient navigation. In this project, we develop methods that allow robots to learn and adapt models of their motion capabilities. Specifically, we study approaches that learn action-conditional forward models which could be used for model-predictive control and navigation planning.
In TRADYN~[ ], we base model learning on our prior work Explore-the-Context~[ ] to learn a terrain- and robot-aware dynamics model for a simulated unicycle robot. ...todo