BTGenBot: Behavior Tree Generation for Robotic Tasks with Lightweight LLMs

TL;DR

This paper introduces a method for generating robot behavior trees using lightweight LLMs, demonstrating that compact models fine-tuned on specific datasets can produce effective behaviors suitable for real-world deployment.

Contribution

It presents a novel fine-tuning dataset and compares multiple lightweight LLMs for behavior tree generation across various tasks, highlighting practical deployment potential.

Findings

Lightweight LLMs can generate effective behavior trees.

Fine-tuned models perform well in simulation and real robots.

Compact models enable practical on-robot deployment.

Abstract

This paper presents a novel approach to generating behavior trees for robots using lightweight large language models (LLMs) with a maximum of 7 billion parameters. The study demonstrates that it is possible to achieve satisfying results with compact LLMs when fine-tuned on a specific dataset. The key contributions of this research include the creation of a fine-tuning dataset based on existing behavior trees using GPT-3.5 and a comprehensive comparison of multiple LLMs (namely llama2, llama-chat, and code-llama) across nine distinct tasks. To be thorough, we evaluated the generated behavior trees using static syntactical analysis, a validation system, a simulated environment, and a real robot. Furthermore, this work opens the possibility of deploying such solutions directly on the robot, enhancing its practical applicability. Findings from this study demonstrate the potential of LLMs with a limited number of parameters in generating effective and efficient robot behaviors.