Eureka: Human-Level Reward Design via Coding Large Language Models

arXiv:2310.12931v2 Announce Type: replace-cross
Abstract: Large Language Models (LLMs) have excelled as high-level semantic planners for sequential decision-making tasks. However, harnessing them to learn complex low-level manipulation tasks, such as dexterous pen spinning, remains an open problem. We bridge this fundamental gap and present Eureka, a human-level reward design algorithm powered by LLMs. Eureka exploits the remarkable zero-shot generation, code-writing, and in-context improvement capabilities of state-of-the-art LLMs, such as GPT-4, to perform evolutionary optimization over reward code. The resulting rewards …

abstract algorithm arxiv bridge coding cs.ai cs.lg cs.ro decision design eureka fundamental gap however human language language models large language large language models learn llms low making manipulation semantic tasks them type via