Towards a Deep Reinforcement Learning Solution to the Coverage Path Planning Problem

Shaza I. Kaoud Abdelaziz, Aboelmagd Noureldin, Sidney Givigi

Abstract: Coverage path planning is an intriguing problem that applies to many everyday tasks, such as vacuum cleaning, lawn mowing, snow blowing, and, on a much larger scale search and rescue. This prevalent task has a complex nature that requires optimal solutions. Contrary to our homes, where we could accept the sub-optimality of autonomous vacuum cleaners, in case of search and rescue, that delay could be the difference capable of saving lives. In this research, we investigate Deep Q-Networks, elaborate reward shaping, and the effect of action dimension on the solution’s optimality.