Guanghang, Monsieur CAI (2019) Multi-task sim-to-real via distillation in Continual SRL PRE - Research Project, ENSTA.



Reinforcement learning is a state-of-art policy learning model, which is widely used in the aspect of robots and automation. During my three-month internship, I have worked with my team on the continual learning of reinforcement learning models. My job is to reproduce the experiment results on the real robot. Using ROS, we have managed to manipulate a three-wheel robot, Omnirobot, to do its tasks. We have used dynamic randomisation when training the model in a simulation environment. Then we applied the model on the real robot, and found that the result is not similar to that of simulation. We will also discuss about the factors of the discrepancy in the paper.

Item Type:Thesis (PRE - Research Project)
Subjects:Information and Communication Sciences and Technologies
ID Code:7391
Deposited By:Guanghang CAI
Deposited On:09 juin 2021 15:12
Dernière modification:09 juin 2021 15:12

Repository Staff Only: item control page