Adding Terrain Height to Improve Model Learning for Path Tracking on Uneven Terrain by a Four Wheel Robot

Closely tracking a defined path by a wheeled mobile robot on a three-dimensional surface is important for accurate movement on uneven terrain. Conventional methods in two dimensions are difficult to extend to three dimensions due to the computational complexity in finding wheel-terrain interactions. Learning based methods bypass the need for explicit modelling and can accurately predict these dynamic relations. We use learning based Model Predictive Controller (MPC) for path tracking by a four-wheel robot. A neural network is used as a model due to its capability for learning complex state transition dynamics. Learning terrain height information aids the MPC on uneven terrain. The algorithm is rigorously tested in simulation on a variety of terrain profiles to track paths by a four wheel robot's center of mass. Results show the method is robust to model errors and that our novel method of incorporating terrain height information significantly improves performance on terrains with high frequency surface profile changes.

Automated summary

This study utilizes a learning based Model Predictive Controller (MPC) for path tracking by a four-wheel robot on uneven terrain. A neural network is employed as a model with the capability to learn complex state transition dynamics. Incorporating terrain height information improves performance significantly.