REASON-MC: RL for Efficient Sampling Of spatial FunctioNs with Movement Constraints

High-level planner for robots to efficently approximate a spatial function from exposure to previous samples of the function, with the robots movement constaints.

Status: Work in Progress. Will be significantly updated in the next couple of weeks

• Overall environment works, with hard-coded movement constraints and uniform distribution (i.e. Coverage Path Planning)
• Structure: Env contains env logic, train.py contains PPO stuff
• To do: implement a yaml, to specify movement constraints; implement a way to specify spatial functions; better evalution methods; experiments beyond PPO;multiple parallel environments; documentation

Image: In provided env the agent needs a lot of training to even learn how to stay in bounds. We will provide a more efficient env soon.