Efficient Robot Task Allocation and Navigation in Dynamic Decluttering via Multi-Objective Optimization and Reinforcement Learning

Abstract

For a successful deployment of self-driven multi-robot systems in various and constantly changing environments, two of the most important things are efficient task scheduling and route optimization. This work proposes an extensive strategy for addressing these concerns, including optimized task scheduling; model-based reinforcement learning (MBRL) path planning approach has been employed while developing advanced object detection using transformers technique so that they can operate autonomously. The hybrid Prairie dog and Wobat optimization algorithm process encompasses total task completion time, path length, energy consumption, robot idle time, and urgency as metrics to enhance overall system performance. Path planning utilizes MBRL and DQN which are trained with a physics simulator to enable realistic navigation that adapts to real-time environmental changes. Furthermore, the inclusion of ODT enables accurate object detection, which is important for avoiding obstacles in a moving environment. By using a lot of simulations, we have been able to show that task effectiveness, navigation accuracy and general system functionality has been significantly increased. As a result, the improvement allows this to offer an effective response across different parts for deployment on ground using multiple robots in terms of task allocation and movement capacities.