Reinforcement Learning: Controlling a Physical System

Problem
Reinforcement learning has been used to control physical systems such as robots. In this project you will use reinforcement learning to control a simulated physical system. This involves the following steps:

1. Choose a physical system: Physical systems used in the reinforcement learning literature include an inverted pendulum on a cart and a robot arm. I recommend starting with a very simple system, a one link robot arm moving in a gravitational field.
2. Choose a task: Figure out what you want the physical system to do. What is known, and what isn't, at the start of the task? How is the task organized (a set of trials, one continuous attempt to achieve a goal, ...)?
3. Define a reward function: You will need to define the function the system is trying to optimize.
4. Implement a simulation: I can provide you with simulations of a variety of physical systems.
5. Implement reinforcement learning: You will need to implement some form of reinforcement learning.