AI-Based Supervising System for Improved Safety in Shared Robotic Areas

Robots were introduced into the industry to improve quality and productivity. However, questions began to arise regarding the safety of employees. A new generation of robots, called cobots, has emerged and started to gain prominence due to their characteristics, although tools and grippers have limited security to be included in collaborative working cells. In this sense, we consider that the creation of an adaptative safety system for work cells is necessary to improve safety with minimum production impact. Our goal with the creation of an intelligent vision system is to detect humans/robots in the work cell, detect their joints using neural networks, simultaneously acquire the current position of each robot axis and determine the physical distance between the human and robot by creating a virtual environment developed in Unity. Depending on the safety level determined by the system, a different action/speed is communicated to the robot. Low speeds if the safety level is low, and higher speeds if there is no danger to the human. We only intend to make changes to the robot's speed, avoiding sudden stops or emergency stops that will eventually deteriorate the correct operation and fluidity of the robot's movements. In this sense, we present some validation results of the system's operation, performing several tests with different danger and safety distances and verifying if it can accurately identify the safety level that the human is in from the robotic collaborative work cell.