This project integrates Large Language Models (LLMs) and Vision Language Models (VLMs) into Task and Motion Planning (TAMP) to enhance robotic planning. Traditional TAMP requires explicit logical specifications, often limited by closed-world assumptions. We propose a framework where LLMs generate action sequences that are verified by a constraint satisfier (CSP) for physical and geometric feasibility. This iterative process refines plans, ensuring they align with real-world constraints, thereby improving the adaptability and effectiveness of autonomous robots in dynamic environments.

I am participating in SuperUROP to gain hands-on experience and apply my machine learning knowledge to robotics. I aim to deepen my expertise in AI by working on this project, where I’ll develop a framework to enhance robotic planning and execution. This project excites me because it represents a significant step toward creating adaptable, intelligent robotic systems capable of solving complex, real-world challenges.