In-orbit autonomous filling control method and system based on multi-task learning

Abstract

The invention discloses an in-orbit autonomous filling control method and system based on multi-task learning, and belongs to the technical field of space. The method comprises the following steps: constructing a task expression network G and a task execution network H, and training and finely adjusting the task expression network G and the task execution network H by using a reinforcement learning method in an on-orbit filling virtual environment until parameters of the two networks converge to form a multi-task strategy network F; in the on-orbit autonomous filling control system based on multi-task learning, resetting the motion state of a real mechanical arm, and using a multi-task strategy network F for controlling the real mechanical arm, so that the real mechanical arm executes corresponding actions, and an on-orbit filling operation task is completed. Aiming at the problem of insufficient autonomy caused by independent learning of multiple operation tasks, deep reinforcement learning and a multi-task learning method are combined, unified expression and learning of multiple operation task strategy networks are realized, and the autonomy and robustness are improved compared with artificial design task state judgment and switching.

Description

technical field [0001] The invention relates to an on-orbit autonomous refueling control method and system based on multi-task learning, which belongs to the field of space technology. Background technique [0002] On-orbit fuel filling technology can effectively prolong the operating life of spacecraft, and is the precursor and basis for driving other on-orbit service technologies. For teleoperated on-orbit refueling, the communication delay will reduce the controllability and reliability of the system, so the development of independent on-orbit refueling technology is of great value. Facing the space environment with many sources of interference and great uncertainty, the use of manipulators for on-orbit refueling has greater flexibility and robustness. The learning ability determines the operation autonomy level of this system. According to the different learning principles, Table 1 summarizes several learning-based manipulator control methods. It can be seen that the d...

AI Technical Summary

Problems solved by technology

For teleoperated on-orbit refueling, the communication delay will reduce the controllability and reliability of the system, and the development of independent on-orbit refueling technology is of great value

Images