A Comparative Study of Simulators for Sim-To-Real Deep Reinforcement Learning
Note that this project will be co-supervised by Suzan Ece Ada, Phd student in Colors Lab.
This project aims to compare the accuracy of two physics-based simulators, namely MuJoCo and pybullet, for a sim-to-real deep reinforcement learning transfer task. A UR10 robot will be designed in MuJoCo and pybullet. Then, a policy will be trained for each simulator using a policy gradient algorithm of your choice.
The robot will be trained for a manipulation task (Robot arm reacher, grasping etc.) Finally, a comparative evaluation study of the policies trained in different simulators will be conducted on the real UR10 robot.