Detection method for target object grab position based on deep learning robot

Abstract

The invention is applicable to the field of robot grab and provides a detection method for a target object grab position based on a deep learning robot. The method comprises the steps that an RGB-D image containing a target object is collected through a sensor; candidate grab regions are divided in a target region of the RGB-D image; the length-width ratio of each candidate grab region is kept unchanged, the size of each candidate grab region is magnified to the size required for input of a neural network; input vectors are constructed for the magnified candidate grab regions; the input vectors are whitened, and the whitened input vectors are input into the trained neural network; and the score of each candidate grab region is acquired, and the candidate grab region with the highest score is determined as the grab position. The grab position of the target object can be determined just by acquiring the RGB-D image of the target object, the robot can grab any target object through the grab position, and manual intervention is not needed.

Description

technical field [0001] The invention belongs to the field of robot grasping, and in particular relates to a method for detecting the grasping position of a target object based on a deep learning robot. Background technique [0002] In order to care for the elderly, the disabled and other people with disabilities, the grasping of common objects in the home environment, such as teacups, beverage bottles, books, etc., has become an indispensable and important functional requirement for home service robots. Different from the grasping of workpieces by industrial robots in a structured environment, the intelligent grasping of service robots in a home environment faces many challenges, such as dynamic environments, lighting changes, dozens or even hundreds of target objects, complex backgrounds, Mutual occlusion between objects, etc. [0003] At present, the robot grasping detection technology includes the following types: artificially design the grasping characteristics of the t...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present invention provides a method for detecting the grasping position of an object based on a deep learning robot, which aims to solve the existing method of manually designing the grasping characteristics of an object, which is time-consuming and requires a large amount of manual participation, and for The target object that the robot has not seen cannot accurately detect the grasping position, and cannot execute the grasping action

Images