Automating Reward Function Design Using LLMs for Robotic Disassembly of EV Batteries

• EPFL - Ecole Polytechnique Fédérale de Lausanne
• Automatic Control Lab

This project explores the integration of large language models (LLMs) into a reinforcement learning (RL) pipeline to automate reward function design for the robotic disassembly of end-of-life electric vehicle (EV) batteries. By iteratively refining reward functions using LLMs, the approach aims to enhance training efficiency, accelerate learning, and improve the precision and safety of complex disassembly tasks. Utilizing Nvidia Isaac Sim for simulation and transferring skills to real-world robots, the research seeks to reduce human intervention in reward engineering, providing scalable solutions for advanced robotic manipulation in battery recycling and beyond.

• Intelligent Robotics, Mechanical Engineering, Systems Theory and Control, Text Processing, Virtual Reality and Related Simulation
• Master Thesis