Reinforcement Learning for Multi-robot Field Coverage Based on Local Observation

Field coverage is a representative exploration task that has many applications ranging from household chores to navigating harsh and dangerous environments. Autonomous mobile robots are widely considered and used in such tasks due to many advantages. In particular, a collaborative multirobot group can increase the efficiency of field coverage. In this paper, we investigate the field coverage problem using a group of collaborative robots. In practical scenarios, the model of a field is usually unavailable and the robots only have access to local information obtained from their on-board sensors. Therefore, a Q-learning algorithm is developed with the joint state space being the discretized local observation areas of the robots to reduce the computational cost. We conduct simulations to verify the algorithm and compare the performance in different settings.