Hardware-validated hybrid super-twisting, fuzzy logic, and deep reinforcement learning framework for spacecraft attitude control

Engineering notes

Comprehensive quantitative metrics (MSE/ISE/ITSE, settling time, control effort, and disturbance tolerance) indicate that fuzzy-GA-based STA designs provide the best overall trade-off between precision and efficiency, whereas DRL hybrids offer superior resilience under commanded and external disturbances.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

This study proposes a hardware-validated hybrid intelligent framework for spacecraft attitude control that integrates the super-twisting sliding-mode algorithm (STA), Type-1/Type-2 fuzzy logic control (FLC), and deep reinforcement learning (DRL). Fourteen controller configurations are developed and systematically compared across four architectural scenarios: (i) baseline STA, (ii) STA augmented with Type-1/Type-2 FLC, (iii) STA augmented with DRL (DDPG, PPO, TD3), and (iv) integrated STA–FLC–DRL schemes. Fuzzy hyperparameters are tuned via genetic algorithm (GA) and Taguchi method to balance tracking accuracy and control effort. The methodology is assessed in both high-fidelity simulations and hardware-in-the-loop (HIL) experiments on a three-degree-of-freedom (3-DoF) platform equipped with three reaction wheels. In simulation, GA-tuned hybrid fuzzy–STA controllers achieve faster settling and lower time-weighted tracking error than baseline STA, whereas Taguchi-tuned and DRL-augmented variants trade settling speed for improved disturbance rejection. HIL experiments reveal a systematic performance gap relative to simulation, with several controller rankings reversing under sensor noise and unmodeled dynamics; nonetheless, closed-loop stability is preserved across all configurations. Comprehensive quantitative metrics (MSE/ISE/ITSE, settling time, control effort, and disturbance tolerance) indicate that fuzzy-GA-based STA designs provide the best overall trade-off between precision and efficiency, whereas DRL hybrids offer superior resilience under commanded and external disturbances. The findings establish a practical pathway for tailoring intelligent attitude controllers to mission constraints and underscore the importance of joint data-driven and robust design to bridge the simulation-to-reality gap. Future work will extend the framework to fault-tolerant operation and learning-based disturbance estimation in orbit.

Links and sources

• PDF from original source

Comments