A Distributed Learning Framework for Energy-Aware Cooperative Navigation in Multi- Robot Systems
Abstract Cooperative Navigation Problem with Multiple Robots is known to be a quite challenging task, especially under unknown and dynamic traffic conditions. This happens since robots operating under these circumstances usually suffer from a lack of communication, inability to observe the whole environment, and a short lifetime due to a low level of available battery power. Traditional approaches to centralized control and planning for multi-robot navigation often fail here as they do not scale well and carry extremely large computational costs. A team of researchers has come forward to offer a new solution based on the concept of Energy-Efficient Distributed Deep Reinforcement Learning (ED-DDRL). In particular, their method involves developing a much-decentralized decision-making process, which means that each robot independently learns its own policy while communicating with other neighboring ones through the minimal amount of shared local data. To make sure all operations work smoothly without wasting too much electricity, the authors proposed a unique reward function taking into account various factors affecting robot movement. Thanks to an advanced lightweight communication scheme developed specifically for this system, ED-DDRL significantly reduces inter-robot bandwidth utilization levels. All in all, tests performed extensively within a physical prototype showed that compared with state-of-the-art solutions like DQN, PPO, MAPPO, and even MADDPG, ED-DDRL excels in several critical areas – achieving a higher overall success rate alongside enhanced energy efficiency metrics. The latter was most evident during trials involving environments packed with lots of bots where ED-DDRL easily managed to keep ahead.
Authors
- Mohamed Shili (ORCID: https://orcid.org/0000-0003-0311-3724)
Institutions
- Tunisia Polytechnic School (TN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-11578692/v1
- Primary Topic
- Reinforcement Learning in Robotics
- Type
- preprint