Boosting Robot Behavior Generation with Large Language Models and Genetic Programming

Aaron Verdaguer Gonzalez, Luis Merino Cabañas, Magí Dalmau-Moreno, Nestor Garcia

Keywords: LLM, Genetic Programming, Behavior Trees, Robot, Task Planning, Mobile Manipulation, Orchestration, HRI

TL;DR: Automatically generating behavior trees from task descriptions with a combination of LLMs and Genetic Programming.

Abstract: Mobile robots are increasingly ubiquitous in modern society, necessitating more human-like interaction capabilities, such as following operator instructions or collaborating with humans. Conventional robot programming methods often fall short in achieving these complex behaviors. Behavior Trees (BTs) offer a promising alternative due to their modularity, scalability and reactivity. We propose using Large Language Model (LLM) assistants to decompose task descriptions into executable BTs. The BTs are then refined using Genetic Programming and a low-resource simulator, eliminating the need for fine-tuning LLMs. Our approach accelerates behavior generation, enhances applicability in diverse environments, and democratizes the process for non-experts. Besides, it enables the generation of adaptable behaviors tailored to various scenarios.

Submission Number: 19