Optimal Tasking of Ground-Based Sensors for Space Situational Awareness Using Deep Reinforcement Learning.

Area of Science:

• Space Situational Awareness (SSA)
• Artificial Intelligence
• Robotics and Control Systems

Background:

• Increasing numbers of resident space objects (RSOs) complicate space situational awareness (SSA).
• Ground-based sensors face challenges in tracking numerous RSOs due to narrow fields of view and slew rate limitations.
• The SSA sensor tasking problem presents a complex combinatorial challenge.

Purpose of the Study:

• To apply deep reinforcement learning (DRL) to optimally task ground-based SSA sensors.
• To overcome the curse of dimensionality in sensor tasking.
• To improve the efficiency and effectiveness of RSO detection and tracking.

Main Methods:

• Development and training of DRL agents using proximal policy optimization and population-based training.
• Simulation of a realistic SSA environment for agent training and evaluation.
• Comparison of DRL agent performance against myopic policies.

Main Results:

• DRL agents demonstrated superior performance in reducing RSO state uncertainties.
• DRL agents observed a greater number of unique RSOs within a 90-minute window compared to myopic policies.
• Agents showed robustness across various RSO orbital regimes, observation durations, locations, and sensor characteristics.

Conclusions:

• Deep reinforcement learning offers an effective solution for the complex SSA sensor tasking problem.
• DRL agents provide a robust and adaptable approach for optimizing sensor operations in diverse scenarios.
• The developed DRL framework can be applied to arbitrary locations and SSA mission requirements.