A Motion Camouflage-Inspired Path Planning Method for UAVs Based on Reinforcement Learning

Abstract

Inspired by the phenomenon of motion camouflage, this article proposes a fast path planning method for uncrewed aerial vehicles (UAVs) by dimensionality reduction for both efficiency and accuracy. First, a virtual motion camouflage (VMC)-based environment model is constructed, where the UAV is considered as a virtual predator and its states are dimensionally reduced to form a set of 1-D path control parameters. Then, based on the constructed environment model, the UAV path planning process is formulated as a Markov decision process, and then VMC-based Q-learning is introduced to optimize the path control parameters to achieve effective and fast path planning. In particular, a reference point selection rule based on obstacle configuration is established to improve algorithm efficiency. This is because different reference points in the VMC-based environment model result in state space with varying density distributions. The simulation results suggest that due to the reduction in dimensionality, the proposed method achieves superior performance by providing better path planning result, reducing the total running time, and reaching convergence more quickly.