A Deep Deterministic Policy Gradient‐Compensated Backstepping Controller for Adaptive Tracking and Aiming in Aerial Two‐ DOF Gimbal System

ABSTRACT To improve the tracking and aiming performance of the aerial two‐DOF gimbal system, this paper proposes a backstepping‐based control strategy compensated by the Deep Deterministic Policy Gradient (DDPG) algorithm to address the system's strong nonlinearity and susceptibility to external disturbances. The backstepping control (BSC) method constructs a hierarchical control law to ensure system asymptotic stability, whereas the DDPG algorithm compensates for modeling errors and external disturbances via a deep reinforcement learning process based on policy optimization. A mathematical model of the aerial two‐DOF gimbal system is first established. Two DDPG‐compensated backstepping controllers are designed: a centralized single‐agent compensation structure (BSC‐DDPG‐C) and a decentralized multi‐agent compensation structure (BSC‐DDPG‐D). Simulation experiments were conducted to analyze the training behaviors of the two BSC‐DDPG controllers and reinforcement‐learning‐based PID (RL‐PID) controller, followed by performance comparisons with three conventional controllers: a PID controller, a backstepping controller, and an adaptive backstepping controller. The results show that both BSC‐DDPG controllers significantly outperform the traditional ones, while the RL‐PID controller provides moderate improvement but remains inferior to the DDPG‐compensated methods. Among all, BSC‐DDPG‐D achieves the best yaw and pitch tracking accuracy and stability, validating the effectiveness and superiority of the proposed approach.