Learning Improvement Heuristics for Multi-Unmanned Aerial Vehicle Task Allocation

Abstract

Nowadays, small UAV swarms with the capability of carrying inexpensive munitions have been highly effective in strike missions against ground targets on the battlefield. Effective task allocation is crucial for improving the overall operational effectiveness of these UAV swarms. Traditional heuristic methods for addressing the task allocation problem often rely on handcrafted rules, which may limit their performance for the complicated tasks. In this paper, a NeuroSelect Discrete Particle Swarm Optimization (NSDPSO) algorithm is presented for the Multi-UAV Task Allocation (MUTA) problem. Specifically, a Transformer-based model is proposed to learn design NeuroSelect Heuristic for DPSO to improve the evolutionary process. The iteration of DPSO is modeled as a decomposed Markov Decision Process (MDP), and a reinforcement learning algorithm is employed to train the network parameters. The simulation results are provided to verify the effectiveness of the proposed method.

Keywords

• unmanned aerial vehicle (UAV)
• multi-UAV task allocation
• discrete particle swarm optimization
• reinforcement learning
• transformer