Debate2Create: Robot Co-design via Multi-Agent LLM Debate

TL;DR

Debate2Create (D2C) employs multi-agent LLM debate grounded in physics to improve robot co-design, outperforming prior methods in locomotion benchmarks by iterative structured discussion and multi-objective feedback.

Contribution

D2C introduces a novel multi-agent debate framework for robot co-design, integrating physics-based evaluation and multi-objective feedback to enhance morphology and control optimization.

Findings

Achieves up to 3.2x and 9x improvements on MuJoCo benchmarks.

Iterative debate yields 18-35% gains over zero-shot methods.

Rewards transfer successfully to default morphologies in most tasks.

Abstract

We introduce Debate2Create (D2C), a multi-agent LLM framework that formulates robot co-design as structured, iterative debate grounded in physics-based evaluation. A design agent and control agent engage in a thesis-antithesis-synthesis loop, while pluralistic LLM judges provide multi-objective feedback to steer exploration. Across five MuJoCo locomotion benchmarks, D2C achieves up to $3.2\times$ the default Ant score and $\sim9\times$ on Swimmer, outperforming prior LLM-based methods and black-box optimization. Iterative debate yields 18--35% gains over compute-matched zero-shot generation, and D2C-generated rewards transfer to default morphologies in 4/5 tasks. Our results demonstrate that structured multi-agent debate offers an effective alternative to hand-designed objectives for joint morphology-reward optimization.