Low Cost Vision-Based Gripper - Camera based gripper to locate, localize and grasp object
Date
July 2021
Service
Employer
Centre of Excellence in Advanced Manufacturing Technology, IIT Kharagpur
Project Overview
This project focused on automating object detection and grasping for a KUKA robot in an assembly line. The objective was to identify whether an object was cylindrical or cuboid to select the appropriate gripper type (three-fingered for cylindrical objects, parallel plate for cuboids). YOLOv4 was used to identify the cuboid and cylindrical objects.
Watch the video :
https://www.youtube.com/watch?v=g1MSxzrRrLM
Key Highlights
To pick up an object, it is essential to find whether the object is cylindrical or cuboid so that the appropriate gripper is used for grasping. The three-fingered gripper is generally used for cylindrical surfaces whereas parallel plate grippers are used for cuboids.
YOLOv4 was used for object detection and to differentiate between cylindrical and cuboid.
In addition, the vertices of a cuboid were found using segmentation of the object and it was used to estimate pose of the cuboid. OpenCV was used to perform segmentation and automate the KUKA robot.
The camera mounted on top of the KUKA robot, aided in calculating the position and the pose for perfect grasp.
Used Python to connect the KUKA robot controllers through the cloud to feed the 6D coordinates and automate the process in an assembly line.
