Mechanical arm grabbing planning method and system combined with self-supervised learning

Abstract

The invention discloses a mechanical arm grabbing planning method and system combined with self-supervised learning, and the method comprises the steps: generating a plurality of grabbing postures through the complete point cloud of an object, obtaining a pair of contact points between the grabbing postures and the complete point cloud of the object, comparing the included angles between the normal vectors, pointing to the interior of the object, of the pair of contact points and the unit vectors, serving the tangent value of the large included angle as the quality coefficient of the grabbing posture, and generating a category label for the grabbing posture; serving the multiple grabbing postures with the category labels as training data, training the classification neural network to be converged, and obtaining a grabbing quality classification model; and generating a plurality of candidate grabbing postures through the complete point cloud of the object to be grabbed, scording and classifying the candidate grabbing postures through the grabbing quality classification model, and serving the candidate grabbing posture with the highest score and the category of the candidate grabbing posture as the optimal grabbing plan of the mechanical arm. The self-supervised learning mode is introduced to construct the training data, robust grabbing of unknown objects in the multi-target stacking scene is achieved, and the grabbing success rate of the mechanical arm is increased.

Description

Technical field [0001] The present invention belongs to the field of robot applications, and more particularly to a mechanical arm gripping planning method and system that combines self-supervised learning. Background technique [0002] With the advancement of industrial automation, robotic automation technology has accelerated, and industrial robots are now widely used in industrial production process. At present, robots are gradually moving from unstructured environments such as home services, warehousing logistics, and intelligent has become a new direction of development of robots. The intelligence of the robot requires it to perceive the environment and interact with objects in the environment. A typical way to grabbing the robot is interacting with objects, and the intelligent grabbing of robots will improve production efficiency and improve human-computer interaction. It is of great significance. [0003] Existing research shows that deep learning technology can indeed hel...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the above defects or improvement needs of the prior art, the present invention provides a method and system for grasping planning of a manipulator combined with self-supervised learning, thereby solving the problem of unreasonable grasping strategies, poor algorithm robustness, and grasping problems in the prior art. Technical issues with low success rate

Images