Deep Reinforcement Learning Applied to Dynamic Systems

Deep learning is typically used for pattern classification on static datasets, such as face recognition from streaming video, based on prior experience. We are working on implementing deep learning methods in the framework of reinforcement learning, applied to dynamic systems, such as robot manipulators. We have obtained good results with simulations.

In this project we will apply our algorithms to a physical two link manipulator (robot) in the lab. The students will be responsible of porting our algorithms to the robot controller (based on Matlab), running the algorithm and analyzing the data.

At the end, we expect a method which learns to control the robot, so that the robot will follow the motion paths that are supplied to it.