Strategy for adaptive rover behavior based on experiment

Due to the fundamental nature of exploration in rough-terrain, a rover accomplishing this task is naturally confronted with an unknown environment. It is especially true regarding its interaction with the soil, as the nature of it is uncertain. This work aims at creating a strategy allowing the rover to learn from its interaction with the terrains encountered, with the goal of optimizing its behavior. In practice, the information characterizing the terrains, obtained form remote sensors (such as camera), is associated with local sensors (e.g. IMU), characterizing the rover-terrain interaction. Correspondence between those data is learned and used, through the path planner E*, to influence the rover trajectory. The CRAB platform is used in this project for the implementation and testing of this approach.