Mechanical arm manipulation skill learning method based on task decomposition

Abstract

The invention discloses a mechanical arm manipulation skill learning method based on task decomposition. The mechanical arm manipulation skill learning method based on task decomposition comprises the following steps: defining skill representations of subtasks by decomposing different complex manipulation tasks, and taking the skill representations as manipulation skill primitives; then acquiring the manipulation skill primitives and task-related strategy networks based on visual information by means of reinforcement learning and imitation learning, and packaging the manipulation skill primitives and the task-related strategy networks into task-related strategy libraries in the form of parameters respectively; and loading the task-related strategy networks and the manipulation skill primitives required for task completion through text information, and completing the complex manipulation tasks in real time according to scene information. According to the method for conducting mechanical arm complex manipulation skill learning in combination with RGB-D visual perception and the text information, the problems that the overall skill learning difficulty for completing the specific complex tasks is large, and the reusability is not high can be effectively solved.

Description

technical field [0001] The invention belongs to the field of robot operation skills learning, and relates to a method for learning manipulator skills based on task decomposition; specifically, it is a method for learning complex manipulator skills based on RGB-D visual perception and text information. It is used to solve the problem of high difficulty in learning overall skills and low reusability of specific complex tasks. Background technique [0002] The definition of a robot requires it to have the ability to sense and change its environment, so the ability to manipulate objects is crucial for an intelligent robot. Robot manipulation skill refers to the ability of a robot to operate a specific object in the environment within a limited time based on its own sensing, perception, decision-making, planning and control capabilities, so that the object can reach the target state from the initial state. Skills are ubiquitous in life and production, such as depalletizing and p...

AI Technical Summary

Problems solved by technology

[0007] Existing methods have the following problems: 1) Learning skills requires a large amount of sample data, and the training time is long, time-consuming and laborious; 2) The learned skills can only be aimed at specific environments and tasks. Facing new tasks, new skills need to be re-learned , the reuse rate of skills is low; 3) The skill learning results are only verified on simple operation tasks, and the effect on complex operation tasks is unknown

Images