This project explores the application of reinforcement learning for autonomous control of a drone. The primary objective is to develop an RL-based policy capable of stabilizing a drone in a fixed position in space. By optimizing the drone’s ability to maintain a particular location, we will be able to establish a foundation for more complex behaviors such as navigating dynamic environments an, hopefully, race courses.
On-policy reinforcement learning with neural networks.
Our initial focus will be on achieving a stable hovering position through RL. This will involve a state space, develolped by us, that will capture the drone’s position, velocity, and other important parameters. This will ensure the agent can maintain balance. Once stability is achieved, we will extend the policy to enable the drone to follow a moving target.
We will meet with Prof. Fox on Thursday, Feb. 6 at 4:45 PM.