A method for end-to-end robotic arm control based on deep learning

Abstract

The invention discloses a method for conducting arm control through deep learning. The method comprises the steps that firstly, an image of the motion process of a mechanical arm is collected, meanwhile, a control command of the arm is recorded at a certain frequency, and a controller expressed through a deep neural network is obtained through an end-to-end training method. On that basis, close observation shows that the control structure is expressed through the deep neural network, the motion error of the arm can be further reduced through the end-to-end training method, and obstacle avoidance motion can still be achieved greatly under the condition that an obstacle exists. The method is flexible to implement, samples needed by training are decreased to a great extent, and the method has a great advantage for the condition that large samples cannot be obtained easily for motion of the mechanical arm.

Description

technical field [0001] The invention belongs to the field of deep reinforcement learning, and in particular relates to an end-to-end motion control method, which greatly reduces the sample data required for training. Background technique [0002] In recent years, deep learning has developed rapidly in academia, especially in pattern recognition. In many traditional recognition tasks, the recognition rate has been significantly improved. Many other fields also try to use deep learning to solve some problems in this field. [0003] There have been some studies on the application of deep learning applications in the field of control, especially the combination with reinforcement learning, showing its unique advantages. Deep reinforcement learning is a field that combines deep learning and reinforcement learning. It can realize a new algorithm for end-to-end learning from perception to action. Simply put, it is the same as humans, input perceptual information such as vision, ...

AI Technical Summary

Problems solved by technology

However, in the field of control, sometimes data is not easy to obtain or the cost of obtaining data is high

For example, the motion control data of the robotic arm mentioned in this article limits the application of deep learning.

Images