EILLM: efficient imitate with large language models

Minghao Qiu; Hankz Hankui Zhuo

doi:10.1117/12.3039320

4 September 2024 EILLM: efficient imitate with large language models

Minghao Qiu, Hankz Hankui Zhuo

Proceedings Volume 13259, International Conference on Automation Control, Algorithm, and Intelligent Bionics (ACAIB 2024); 1325947 (2024) https://doi.org/10.1117/12.3039320
Event: Fourth International Conference on Automation Control, Algorithm, and Intelligent Bionics (ICAIB 2024), 2024, Yinchuan, China

Abstract

Rapid evolution is underway in the fields of nature language processing and robotics. Our research endeavors to integrate Large Language Models (LLMs) with imitation learning, aiming to enhance the efficiency of action generation in robotics. The proposed model, Efficient Imitate with Large Language Models (EILLM), utilizes the vast knowledge of LLMs, thus improving the limitations that have traditionally plagued imitation learning approaches, which often falter when confronted with limited expert data. Employing a multi-pronged approach, EILLM adopts LLMs for task understanding, state-action pair analysis, and Monte Carlo search tree pruning, thereby simultaneously raising the accuracy of action generation and efficiency of search process. Furthermore, our experiments demonstrated the superior performance of EILLM compared to baseline models, corroborating its efficacy in achieving sample-efficient robot motion by integrating LLMs and imitation learning.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Minghao Qiu and Hankz Hankui Zhuo "EILLM: efficient imitate with large language models", Proc. SPIE 13259, International Conference on Automation Control, Algorithm, and Intelligent Bionics (ACAIB 2024), 1325947 (4 September 2024); https://doi.org/10.1117/12.3039320

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available