Abstract:
We present a system that addresses the challenge of concurrently mapping, scheduling, and deploying a team of energy-constrained robots to persistently cover an unknown and potentially dynamic environment. This system can passively maintain an accurate representation of occupied space, allowing robots reliable access for monitoring, study, or search and rescue. Current state-of-the-art algorithms only address relaxed versions of this problem by neglecting energy constraints, considering only single-robot problems, or addressing simpler objectives. Typical long horizon schedulers have difficulty responding to rapid changes in the environment, while greedy mapping techniques fail to take proper advantage of cooperation in multi-robot teams. We show the proposed system deploying teams of robots to rapidly explore an environment while making efficient use of their limited energy capacity. Additionally, we introduce confidence decay to gradually reduce model certainty over time and passively induce revisiting of locations that may have changed since last observed. We reserve the exploitation of this feature for future work, but show in this presentation how we are still able to reduce uncertainty despite the constant influx of entropy.

Committee:
Nathan Michael
Katia Sycara
Maxim Likhachev
Wenhao Luo