reinforcement learning for robot routing and testing against uncertainty

this paper discusses the routing of a robot in an environment (4x12) by reinforcement learning. in this environment, there are 48 locations (squares) where 10 squares of these 48 squares are burned and the agent must learn to choose a path that does not hit the flames. our expectation of the agent is that it can recognize two paths. the first path is a safe path to avoid flames as far as possible and the second path is the fastest path. simulation of this issue has been done with matlab software. after learning the agent, uncertainty has been added to the environment. the uncertainty in this problem is defined as adding new fire squares, and the simulation results show that, ultimately, the agent has been able to detect this change in the environment and suggest another new safe path.