Citation

Tan, Chi Jie and M.A., Munjer and Zheng, Pan Wei and Eiji, Hayashi and Lim, Way Soong (2026) LLM-Supervised Genetic Programming for Multi-Robot Behavior Tree Evolution. In: Conference on Artificial Life and Robotics, 29 January 2026 - 1 February 2026, Oita, Japan.

Full text not available from this repository.

Abstract

The use of large language models (LLMs) as adaptive supervisors for mutation rate control in genetic algorithms for evolving robot behavior trees is investigated. A controlled comparison was conducted between an LLM-supervised system (Claude Sonnet 4) and a baseline genetic algorithm with fixed parameters over 48 generations (N=30). Significant performance gains were observed under LLM supervision, including a 16.3% increase in peak fitness, 45–56% faster convergence to high-fitness thresholds, and a 153% increase in excellent solutions. Additionally, population robustness was enhanced, with an 83% higher proportion of complex behavioral specialists and a 303% improvement in clearing behavior coverage, indicating effective mitigation of premature convergence. These findings provide empirical evidence that natural-language reasoning applied to population dynamics can complement numerical fitness signals, enabling a more effective exploration–exploitation balance and yielding superior solution quality in evolutionary optimization. © The 2026 International Conference on Artificial Life and Robotics (ICAROB2026).

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Behavior Tree, Field Robotics, Large Language Models (LLMs), Multi-objective Optimization
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	03 Mar 2026 03:49
Last Modified:	06 Mar 2026 02:29
URII:	http://shdl.mmu.edu.my/id/eprint/15437

Downloads

Edit (login required)