Non-Calibrated Camera-Based Robot Arm for Pick and Place Tasks

Abstract

The integration of robotic systems with machine learning and computer vision has become increasingly critical in automating systems. This is specifically essential to accomplish complex and different tasks, hence enabling intelligent systems to operate with minimal human intervention. These tasks include object recognition, classification, and manipulation. However, accurate object localization and manipulation pose challenges for robotic systems relying on non-calibrated cameras due to positional inaccuracies. This paper introduces a new non-calibrated camera mapping technique to enhance precision in robotic pick-and-place operations. To this end, a robotic arm controller is implemented to pick and place objects using efficient machine learning and computer vision techniques. The proposed approach is used to pick and place different objects that are identified and labeled by certain classes and sorted accordingly. The location of the objects is found using a proposed non-calibrated camera approach. This approach eliminates the need for expensive calibration procedures while maintaining high accuracy. The proposed method is evaluated by comparing actual object locations with those found using the camera mapping, resulting in a 2.458% error rate. In addition, a comparison with other non-calibrated camera approaches demonstrated that the proposed method achieved higher accuracy and lower errors. These results show promise for applications in industrial automation and robotics systems.