Development of a digital system of MyCobot 280 Pi

Reinforcement learning has caught significant attention in the past decade with landmark successes including deep Q-net applied to video games and AlphaGo for the game of go. However, the gap is still wide when it comes to application to real world problems mainly due to high training cost. We propose to develop a digital model of a physical system for the purpose of training an autonomous agent. This way, reinforcement algorithms can be trained on the digital model at a fraction of cost. This project will focus on a specific system: MyCobot 280 Pi from elephantRobotics. The digital model will mimic the dynamics of the physical system, and the calibration of the digital model will be done by a reinforcement learning algorithm itself. This project will be a groundwork for subsequent investigation.

Faculty Supervisor:

Chi-Guhn Lee

Student:

Partner:

Cobionix

Discipline:

Engineering

Sector:

Manufacturing

University:

University of Toronto

Program: