IJCAS

This study presents an integrated algorithm designed for real-time task assignment, collision-free path planning, and control. We introduce two algorithms: MPPI-GA for single-agent systems and generalized MPPIGA for multi-agent systems. Our approach leverages model predictive path integral control (MPPI) and genetic algorithms (GA) as the foundational framework. Both MPPI and GA, being sampling-based algorithms, utilize a paired concept of samples and individuals, respectively. In contrast to the existing methods that adopt a hierarchical structure for task management, path planning, and control in a sequential manner, our proposed algorithm executes such operations concurrently, ensuring real-time performance. Moreover, the generalized MPPI-GA addresses the traveling salesperson problem with self-interest task selection and decentralized conflict resolution, even within multi-agent systems. Consequently, it can autonomously schedule tasks on each computational hardware unit without requiring computation of other agents’ behaviors. We validate the efficacy of MPPI-GA through numerical simulations implemented in Python. Furthermore, we conduct experiments employing two four-wheel-drive mobile robots equipped with the proposed algorithm, thus substantiating its real-time performance.

Keywords: Decentralized multi-agent systems, genetic algorithm, model predictive path integral control (MPPI), path planning, task assignment.