Residual error network deep learning method for mechanical arm grabbing pose estimation

Abstract

The invention discloses a residual error network deep learning method for mechanical arm grabbing pose estimation, which comprises the following steps of: initializing a mechanical arm, and adjustinga mechanical arm wrist camera to enable the mechanical arm wrist camera to be positioned at a known height above a vertical X0Y plane; Acquiring a depth image of a to-be-grabbed object of the mechanical arm; performing mapping processing on the depth image by using a pre-trained improved GG-CNN model, and outputting four 300 * 300 pixel captured information images, including a capture success rate, a capture angle cosine value, a capture angle sine value and a capture width; obtaining the grabbing angle and width information of the position with the highest success rate; and the grabbing angleand width of the target object under the mechanical arm base coordinate system are obtained through coordinate transformation according to the grabbing information obtained in the grabbing success rate image. Through establishing the residual network, the improved GG-CNN model in the above method enhances fitting effect and the learning capability of the convolutional neural network, and the grabbing precision of the generated grabbing pose is higher.

Description

technical field [0001] The invention belongs to information control technology, and in particular relates to a residual network deep learning method for grasping pose estimation of a mechanical arm. Background technique [0002] In recent years, vision-based robotic grasping has become a hotspot of current research. Generally, when performing a grasping action, it is first necessary to achieve accurate target detection and positioning. Traditional target detection is usually static detection, and the target is single. Target detection is affected by factors such as changes in shape, size, viewing angle, and changes in external lighting. Therefore, the generalization ability of the extracted features is not strong and the robustness is poor. The development of deep learning algorithms has facilitated the progress of object detection and localization tasks. The research community generally believes that a deep network is generally better than a shallow network, but the depth...

AI Technical Summary

Problems solved by technology

However, due to the above-mentioned GG-CNN's excessive pursuit of the speed of recognition and grasping, the recognition accuracy of the neural network is reduced, which makes the application of this network model in the grasping of manipulators have certain limitations.

Images