LangProp: A code optimization framework using Large Language Models applied to driving – Shu Ishida

LangProp: A code optimization framework using Large Language Models applied to driving

An overview of the LangProp trainer. The LLM generates variations of code, which is then evaluated on the training dataset. Codes with high scores are kept. The LLM is provided with information about the failure modes of the code and rewrites them to achieve higher performance on the training metric.

Autonomous Driving, Embodied Agents, Large Language Models, Machine Learning, Navigation, Robotics