Efficient Robot Design

with Multi-Objective Black-Box Optimization

and Large Language Models

IEEE Access (2026)

  • Kento Kawaharazuka
  • Yoshiki Obinata
  • Naoaki Kanazawa
  • Haoyu Jia
  • Kei Okada
  • EVA Robotics Laboratory (EVARL) / JSK Robotics Laboratory, The University of Tokyo, Japan

Various methods for robot design optimization have been developed so far, ranging from numerical optimization to black-box optimization. While numerical optimization is fast, it is not suitable for cases involving complex structures or discrete values, leading to frequent use of black-box optimization instead. However, black-box optimization suffers from low sampling efficiency and takes considerable iterations to obtain good solutions. In this study, we propose a method to enhance the efficiency of robot body design based on black-box optimization by utilizing large language models (LLMs). In parallel with the sampling process based on black-box optimization, sampling is performed using LLMs, which are provided with problem settings and extensive feedback. We demonstrate that this method enables more efficient exploration of design solutions and discuss its characteristics and limitations.

Overview

We set the robot design parameters, automatically generate the robot URDF, compute objective functions, and perform multi-objective black-box optimization with LLM-based sampling in parallel.

Design Parameters and Objectives

The design parameters include the base link coordinate, joint types, and link lengths. The objectives are to minimize end-effector position error and required torque for given operation points.

LLM-Based Sampling

The LLM receives the problem settings and feedback from sampled solutions to propose additional designs. This allows broader exploration while the black-box optimizer focuses on precise numeric refinement.

Experimental Setup

We evaluate three sets of target operation points (Target-1, Target-2, and Target-3) to compare optimization performance.

Results for Target-1/2/3

The proposed method explores Pareto-optimal solutions more efficiently than black-box optimization alone across all targets.

Result for Target-1

Result for Target-2

Result for Target-3

Evaluation and Ablation

Hypervolume evaluation and ablation studies show improved performance and stability when LLM-based sampling is included.

Bibtex

@article{kawaharazuka2024efficient,
  author={Kento Kawaharazuka and Yoshiki Obinata and Naoaki Kanazawa and Haoyu Jia and Kei Okada},
  title={{Efficient Robot Design with Multi-Objective Black-Box Optimization and Large Language Models}},
  journal={IEEE Access},
  year={2026},
  doi={10.1109/ACCESS.2026.3664844},
}

Contact

If you have any questions, please feel free to contact Kento Kawaharazuka (kawaharazuka@jsk.imi.i.u-tokyo.ac.jp).