Automated pneumatic vacuum suction robotic arm with computer vision

Abstract

In this paper, a 4 DOF robotic arm that made of acrylic was built and equipped with a 5 MP USB camera, a pneumatic vacuum suction cup and a small air pump. An image will be captured by the camera and then processed with computer vision algorithms. The algorithm will extract the coordinate from the object image. The extracted coordinate data from test objects then fed to the microcontroller. Pulses were generated by the microcontroller to the driver module to control the motors. Some parameters were changed and adjusted accordingly to achieve a different result for analysis purpose. Objects that are used in this research are boxes with various shapes, colors, surface sizes, and weight. The result, the robot was able to reach the object within 2 mm to 4 mm tolerance. Image processing algorithm centroid calculation errors are within 2 mm. Vacuum pickup success rate is 99%.

Intelligent sorting method for assembly line based on visual positioning and model predictive control of robotic arm

“…YOLOv5 uses GIOU_LOSS as the loss function, which is composed of bounding box confidence loss L conf , category loss L cla , and coordinate loss L GIOU [23,24] . The calculation of the loss function is shown in Equations ( 3)- (5).…”

“…Ruan et al [4] reviewed two main methods of fruit location and recognition, including digital image processing techniques and algorithms based on deep learning, and conducted target tracking in dynamic jamming environments. Do et al [5] generated both target and reference model point clouds based on a depth camera system. Lim et al [6] extracted the coordinates of feature points of boxes with different shapes and then controlled the robot.…”