Jérémy Ferrara

Project: Sim-to-Real using reinforcement learning applied to collaborative robots

My research project is taking place in the 3IT of the University of Sherbrooke under the supervision of Pr Wael Suleiman.
The main objective of my research will be to reduce the gap between simulation learning and real learning in the field of collaborative robot control. To achieve this, I will implement a software architecture that is as generic as possible to allow as many collaborative robots as possible to use it.
I will first, based on deep learning methods (GraspGAN, RL-CycleGAN, RetinaGAN, CycleGAN), study the possibility of training the robot using images and commands both in a simulated environment and on the real one. After the training is completed, we will use the resulting decision
policy to test it in a real environment and compare the accuracy between the simulation and the real world. And then figure out possible improvements to further minimize this difference. The tests will be carried out on Sawyer robot from Rethink Robotics.


Supervisor(s):
Jérémy Ferrara
Jérémy Ferrara

Université de Sherbrooke

Master

Robotics

[email protected]