Correcting specular reflections for computer vision in collaborative robotics

Unlike traditional industrial robots, collaborative robots are meant to be used alongside workers. They can also be taught their task by the workers themselves instead of robotic engineers. This means that these robot have to be inherently safe and be able to properly perceive their environments when performing tasks such as picking parts from a conveyor belt. Because these robots can be taught in a wide variety of environments with people moving in and out, this also means that the operating conditions, especially lighting, can be quite dynamic. This produces unwanted artifacts such as reflections and highlights in images used to perceive the environment, and traditional processing systems to perform poorly. This project aims at developing an algorithm based on deep learning to correct such artifacts and allow these robots to work properly in non-ideal lighting conditions, enabling the workers to focus on the task instead of the robot’s needs

Faculty Supervisor:

François Ferland

Student:

Sebastian Pelchat

Partner:

Robotiq

Discipline:

Engineering - computer / electrical

Sector:

Manufacturing

University:

Université de Sherbrooke