Autonomous Robots for Space: Trajectory Learning and Adaptation Using Imitation.

Area of Science:

• Robotics
• Aerospace Engineering
• Artificial Intelligence

Background:

• Spacecraft missions require autonomous robotic manipulators for tasks like debris removal and servicing.
• Robotic arm movements on free-floating spacecraft induce reaction forces, causing attitude changes that strain the Attitude Determination and Control System (ADCS).
• Efficient trajectory planning is crucial for minimizing power consumption during space operations.

Purpose of the Study:

• To develop an imitation learning method for trajectory planning of robotic arms on free-floating spacecraft.
• To minimize spacecraft attitude disturbances caused by manipulator motion.
• To reduce the computational power required for trajectory planning post-deployment.

Main Methods:

• Utilizing programming by demonstration (imitation learning) for trajectory planning.
• Encoding demonstrated trajectories into a probabilistic distribution.
• Conditioning the learned distribution for planning in novel situations.
• Defining a cost function to identify trajectories minimizing attitudinal changes.

Main Results:

• The proposed method effectively learns robot arm trajectories that minimize spacecraft attitude variations.
• Offline trajectory learning significantly reduces the need for onboard computational power during missions.
• The approach enables planning for previously unseen scenarios by conditioning the learned probabilistic model.

Conclusions:

• Imitation learning offers an effective solution for autonomous trajectory planning in space robotics.
• Minimizing attitude disturbances reduces the burden on the ADCS, enhancing mission stability.
• Offline learning and probabilistic trajectory generation contribute to power efficiency and adaptability in space missions.