Layered task planning method and system for spatial fine operation

Abstract

The invention discloses a hierarchical task planning method and system for spatial fine operation, and belongs to the technical field of space. In order to solve the problems of large gradient estimation variance and poor algorithm convergence when a strategy network learns multiple tasks, high-level strategy planning based on a dynamic estimator and Monte Carlo tree search is constructed, and the method specifically comprises the following three steps of offline data collection, asynchronous high / low-level strategy optimization and model evaluation. According to the method, the algorithm convergence is improved, satellite-borne computing resources of on-orbit service space operation can be saved, and actual engineering requirements are met.

Description

technical field [0001] The invention belongs to the field of space technology, and in particular relates to a hierarchical task planning method and system for fine operation in space. Background technique [0002] Spatial fine-tuning requires the ability to learn long, sequential multitasking. For example, in order to realize the on-orbit refueling mission of non-cooperative targets, the service spacecraft needs to first capture the target spacecraft, reposition, cut the wrapping film, trim the wire, open the cover, etc. before performing the refueling operation. In practice, it is difficult to perform the above multiple operational tasks with a single policy network optimized by reinforcement learning. In addition, for multi-task / multi-MDP reinforcement learning, the number of samples to be collected is large, and the parameter dimension of the policy network needs to be increased accordingly, resulting in problems such as large gradient estimation variance in the optimiza...

AI Technical Summary

Problems solved by technology

In practice, it is difficult to perform the above multiple manipulation tasks with a single policy network optimized by reinforcement learning

In addition, for multi-task / multi-MDP reinforcement learning, a large number of samples needs to be collected, and the parameter dimension of the policy network needs to be increased accordingly, resulting in problems such as large variance of gradient estimation in the optimization process and poor algorithm convergence.

Images