Online tree-based planning for active spacecraft fault estimation and collision avoidance

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Autonomous robots require robust fault identification in uncertain environments.
• Coupled fault modes and state uncertainty pose challenges for existing methods.
• Time-optimal fault isolation is critical for safety-constrained operations.

Purpose of the Study:

• To develop a method for autonomous robots to identify and isolate faulty components under safety constraints.
• To address the challenge of ambiguous faults intertwined with robot state estimation.
• To enable motion plans that discriminate between simultaneous actuator and sensor faults.

Main Methods:

• Combined belief-space tree search, marginalized filtering, and concentration inequalities.
• Developed safe fault estimation via active sensing tree search (s-FEAST) planner.
• Utilized active sensing to gain informative observations while enforcing probabilistic state constraints.

Main Results:

• Theoretically demonstrated s-FEAST's convergence to optimal policies.
• Experimentally validated s-FEAST in a robotic spacecraft simulator, successfully performing fault estimation.
• Successfully navigated a collision course with a model comet while identifying faults.

Conclusions:

• s-FEAST enables safe and efficient fault estimation in autonomous systems.
• The method is effective even with coupled fault modes and state uncertainty.
• Validated through simulations and a robotic spacecraft scenario, showing robust performance.