Abstract
This paper presents a novel evasion guidance law for hypersonic morphing vehicles, focusing on determining the optimized wing's unfolded angle to promote maneuverability based on an intelligent algorithm. First, the pursuit-evasion problem is modeled as a Markov decision process. And the agent's action consists of maneuver overload and the unfolded angle of wings, which is different from the conventional evasion guidance designed for fixed-shape vehicles. The reward function is formulated to ensure that the miss distances satisfy the prescribed bounds while minimizing energy consumption. Then, to maximize the expected cumulative reward, a residual learning method is proposed based on proximal policy optimization, which integrates the optimal evasion for linear cases as the baseline and trains to optimize the performance for nonlinear engagement with multiple pursuers. Therefore, offline training guarantees improvement of the constructed evasion guidance law over conventional ones. Ultimately, the guidance law for online implementation includes only analytical calculations. It maps from the confrontation state to the expected angle of attack and the unfolded angle while retaining high computational efficiency. Simulations show that the proposed evasion guidance law can utilize the change of unfolded angle to extend the maximum overload capability. And it surpasses conventional maneuver strategies by ensuring better evasion efficacy and higher energy efficiency.
