Intelligent cooperative guidance method for spacecraft under multi-information incompleteness
Abstract
<p indent="0mm">In multi-role game scenarios involving spacecraft, such as “target-pursuer-defender”, the primary challenge lies in the collaborative guidance game where the target and defender work together against the pursuer. Given the incomplete prior and observational information in such scenarios, including uncertainties in opponent guidance strategies, missing detection data, and measurement noise, this paper proposes an intelligent cooperative guidance method based on an adaptive dueling double deep Q-network (AD<sup>3</sup>QN). Firstly, to address the issue of sparse rewards during agent training, a continuous reward function is constructed using the environmental potential function, ensuring the optimality of the derived guidance strategy concerning the reward function. Secondly, leveraging the feature extraction capabilities of convolutional neural networks and the sequential data modeling capabilities of gated recurrent unit networks, a novel hybrid network architecture is developed to enhance the agent’s ability to estimate observational information and identify prior information. Furthermore, a mechanism for handling incomplete information is designed based on the spatiotemporal continuity of the spacecraft’s motion state and the operational characteristics of the network architecture. By expanding the set of incomplete information over the time dimension, this mechanism fully characterizes the real flight state and unknown strategy trajectory features, achieving efficient convergence of the decision network under various incomplete information conditions. Finally, numerical simulations validate the effectiveness of the proposed method, and comparative simulations demonstrate its superiority in spacecraft evasion performance.</p>
