Feature Detection of Non-Cooperative and Rotating Space Objects through Bayesian Optimization.

Area of Science:

• Robotics and Spacecraft Engineering
• Artificial Intelligence and Machine Learning

Background:

• Spacecraft proximity operations require robust feature detection for navigation and assessment.
• Non-cooperative target feature identification poses significant challenges due to unknown characteristics.

Purpose of the Study:

• To develop an autonomous algorithm for maximizing feature detection on a non-cooperative space object.
• To optimize camera directional angles for a chaser spacecraft using limited sensor data.

Main Methods:

• Implementation of a Bayesian Optimization (BO) strategy coupled with Gaussian Processes (GP).
• Development of the Space Object Chaser-Resident Assessment Feature Tracking (SOCRAFT) algorithm.
• Utilizing a combined reward model incorporating feature detection scores and sinusoidal rewards.

Main Results:

• The SOCRAFT algorithm effectively determines optimal camera angles for feature detection.
• Simulations in 2D and 3D domains confirm enhanced feature detection within limited camera range and field of view.
• The GP models accurately predict feature locations and reward distributions for BO.

Conclusions:

• The proposed BO-based strategy with GP significantly improves feature detection capabilities for chaser spacecraft.
• SOCRAFT offers a viable solution for autonomous feature identification in close-proximity space operations.
• The method demonstrates effectiveness in maximizing feature discovery under constrained sensor conditions.