Safe Reinforcement Learning: Optimal Formation Control With Collision Avoidance of Multiple Satellite Systems

Area of Science:

• Aerospace Engineering
• Robotics and Control Systems
• Artificial Intelligence

Background:

• Multisatellite systems require robust collision avoidance and formation control.
• Existing methods may lack adaptability and guaranteed safety in dynamic environments.

Purpose of the Study:

• To propose a novel safe reinforcement learning (RL) algorithm for multisatellite collision avoidance and formation control.
• To enhance system safety and adaptive capabilities through innovative learning techniques.

Main Methods:

• Developed a safe RL algorithm within an adaptive dynamic programming framework.
• Integrated a barrier function (BF) into the cost function for guaranteed collision avoidance.
• Implemented an adaptive distance-varying learning method combining online and historical data.

Main Results:

• The proposed algorithm achieved effective collision avoidance and stable formation control.
• Numerical simulations verified the algorithm's safety, stability, and superiority over existing methods.
• The system demonstrated adaptive formation and self-learning capabilities.

Conclusions:

• The novel safe RL algorithm effectively addresses multisatellite collision avoidance and formation control.
• The barrier function and adaptive learning method ensure system safety and performance.
• The approach offers a promising solution for autonomous satellite operations.