Observer-Based Deep Reinforcement Learning for Robust Missile Guidance and Control

Abstract

Conventional missile guidance and control systems are difficult to adapt to complex environments, and new missile guidance systems with fast and accurate decision-making capabilities and high robustness are urgently needed to meet the needs of current and future military and aerospace missions. This study presents a deep reinforcement learning (DRL)-based missile guidance system that integrates an extended state observer (ESO) using a leaky proximal policy optimization algorithm. The ESO estimates the state and disturbance information for the missile attitude control system, which is then used as an input for the DRL control. This information helps the agent gain an enhanced understanding of the system and environmental states, leading to more accurate decision making. This approach improves the convergence speed and robustness in controlling the missile attitudes during the interception of maneuvering targets. Furthermore, the system demonstrates robust performance with up to a 30% variation in aerodynamic parameters, maintaining stable control at all times. This approach performs well in uncertain and high-speed environments and is highly relevant to modern aerospace applications. This study lays the foundation for further research on integrating intelligent control algorithms with advanced learning techniques, as it presents a novel and promising missile guidance control strategy based on integrating ESO and deep reinforcement learning.

Keywords

• Deep reinforcement learning
• missile guidance
• extended state observer
• leaky proximal policy optimization