Creating a Custom Reinforcement Learning Environment for Coding Agents

This project explores the development of a small language model using self-attention mechanisms that comprehends Python and is able to produce code capable of solving grid-based autonomy tasks. This project features a custom reinforcement learning environment, which grades the quality of the generated code, allowing the coding agent to improve performance through a robust reward system. We evaluate the use of reinforcement learning for more refined code generation and task-solving and compare against traditional language learning models on their own.
Intern: Kaelyn Sun

Mentors: Zachary Goddard (FPS/KVM), Tanner Fowler (FPS/KVM)