![[2510.25850] Debate2Create: Robot Co-design via Multi-Agent LLM Debate](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.25850] Debate2Create: Robot Co-design via Multi-Agent LLM Debate
Summary
The paper introduces Debate2Create, a framework for robot co-design that utilizes multi-agent LLM debate to optimize robot morphology and reward structures, achieving significant performance improvements over traditional methods.
Why It Matters
This research is significant as it explores innovative approaches to robot design using debate mechanisms, which could lead to more efficient and effective optimization strategies in robotics. By leveraging multi-agent systems, the study opens new avenues for enhancing robot capabilities and performance in real-world applications.
Key Takeaways
- Debate2Create employs a structured debate framework for robot co-design.
- The approach yields substantial performance improvements over existing methods.
- Iterative debate enhances multi-objective feedback for design optimization.
- D2C demonstrates effective transfer of rewards to default robot morphologies.
- This method presents an alternative to traditional hand-designed objectives.
Computer Science > Robotics arXiv:2510.25850 (cs) [Submitted on 29 Oct 2025 (v1), last revised 22 Feb 2026 (this version, v2)] Title:Debate2Create: Robot Co-design via Multi-Agent LLM Debate Authors:Kevin Qiu, Marek Cygan View a PDF of the paper titled Debate2Create: Robot Co-design via Multi-Agent LLM Debate, by Kevin Qiu and 1 other authors View PDF HTML (experimental) 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. Subjects: Robotics (cs.RO); Machine Learning (cs.LG); Multiagent Systems (cs.MA) Cite as: arXiv:2510.25850 [cs.RO] (or arXiv:2510.25850v2 [cs.RO] for this version) https://doi.org/10.48550/arXiv.2510.25850 Focus to learn more arXiv-issued DOI via DataCite Submission history Fr...