Robot out-of-order workpiece grabbing method based on deep reverse reinforcement learning

Abstract

The invention discloses a robot out-of-order workpiece grabbing method based on deep reverse reinforcement learning. A visual sensor shoots image information o a workpiece to be grabbed, and the information is transmitted to an information processing unit; the information processing unit generates three-dimensional point cloud of target workpiece extracting for the shot image; the treated three-dimensional point cloud data are input to deep reverse reinforcement learning, and a robot motion path is calculated; and the robot carries out workpiece grabbing according to the motion path obtained through calculation. The industrial production needs can be met, little expert demonstration data need to be demonstrated, grabbing programming of specific workpieces can be rapidly achieved, and the defects that the deep reinforcement learning. method strategy is poor in adaptation, limited in grabbing capacity, long in training period, low in efficiency and the like are overcome.

Description

technical field [0001] The invention relates to a robot workpiece grasping method belonging to artificial intelligence, in particular to a robot out-of-sequence workpiece grasping method based on deep reverse reinforcement learning. Background technique [0002] As one of the world's top five industrial robot consumers, China's installation volume increased to 36.0% of the world in 2018. A total of 138,000 industrial robots were installed, a year-on-year increase of 59%. The consumption volume has exceeded the sum of Europe and the United States. Intelligent manufacturing is the main direction of Made in China 2025, and there is a huge demand for intelligent industrial robots. The application of robots for handling and loading and unloading accounts for more than two-thirds, and the added value brought by intelligent upgrading is obvious. [0003] With the development of artificial intelligence deep learning, it has begun to study the deep learning of workpiece vision based...

AI Technical Summary

Problems solved by technology

However, traditional reinforcement learning methods have great limitations when solving high-dimensional state and action space problems. Under the condition of limited samples and computing units, the ability to express complex functions is limited, and the performance in practical applications is often not ideal.

At the same time, the traditional deep reinforcement learning algorithm needs to provide a large amount of data for training. During the training process, the robot needs to continue to grasp and try and make mistakes, so that it is possible to obtain a stable grasping ability.

This training method has a long cycle and low efficiency, and there are safety hazards in the actual training process, which often cannot meet the needs of industrial production applications.

Moreover, in the actual multi-step reinforcement learning process, the design of the reward function is very difficult

Images