A Decentralized Partially Observable Markov Decision Process for complete coverage onboard multiple shape changing reconfigurable robots

Achieving complete area coverage by multiple collaborative robots is an essential aspect of productive deployment for applications such as cleaning, maintenance, and patrol. However, in real-world scenarios, physical constraints such as structural obstacles in the environment that generate narrow spaces, hinder the area coverage achieved by the fixed morphology robots, resulting in ineffective deployments. This paper proposes a Multi-Agent Reinforcement Learning (MARL) framework leveraging the Decentralized Partially Observable Markov Decision Process (Dec-POMDP) for Complete Coverage Planning (CPP) of multiple shapereconfigurable robots. This decentralized framework runs onboard each robot to mitigate the physical limitations while achieving complete area coverage. The proposed approach was trained on diverse map environments with different tight spaces to generate policies that enabled the robots to learn and coordinate their joint actions. The framework was further demonstrated on multiple unseen test environments to evaluate the generalization capabilities and area coverage performance. Additionally, baseline comparisons with the different combinations of multiple fixed morphology robots were conducted to validate the area coverage effectiveness of the proposed approach. In all scenarios, the proposed framework achieved 100% area coverage while the baselines only achieved partial area coverage ranging from 46.21% to 96.55%. The improved coverage performance compared to the baselines demonstrates the potential and usefulness of such an approach in the deployment of multiple reconfigurable robots.