Consider the task of a problem attempting to follow a path in a constrained environment with only a few lines to follow. We attempt this using end-to-end reinforcement learning and explore two algorithms for doing so: Deep Deterministic Policy Gradients (DDPG) and Proximal Policy Optimisation (PPO). We further explore different problem formulations to learn a path-following controller or the velocities of the agent directly, and report our findings.
The PDF containing the project's Report can be found here.
Slides used for the project's presentation can be found here.
- Juraj Micko ([email protected])
- Kwot Sin Lee ([email protected])
- Wilson Suen ([email protected])