OpenNovelty.org

RoboMoRe: LLM-based Robot Co-design via Joint Optimization of Morphology and Reward

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 6.0Download Report PDF
robot co-designLLMreward shaping
Abstract:

Robot co-design, the joint optimization of morphology and control policy, remains a longstanding challenge in the robotics community. Existing approaches often converge to suboptimal designs because they rely on fixed reward functions, which fail to capture the diverse motion modes suited to different morphologies. We propose RoboMoRe, a large language model (LLM)-driven framework that integrates morphology and reward shaping for co-optimization within the robot design loop. RoboMoRe adopts a dual-stage strategy: in the coarse stage, an LLM-based Diversity Reflection mechanism is proposed to generate diverse and high-quality morphology–reward pairs and Morphology Screening is performed to reduce unpotential candidates and efficiently explore the design space; in the fine stage, top candidates are iteratively refined through alternating LLM-guided updates to both reward and morphology. This process enables RoboMoRe to discover efficient morphologies and their corresponding motion behaviors through joint optimization. The result across eight representative tasks demonstrate that without any task-specific prompting or predefined reward and morphology templates, RoboMoRe significantly outperform human-engineered design results and competing methods. Additional experiments demonstrate robustness of RoboMoRe on manipulation and free-form design tasks.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper introduces RoboMoRe, a framework that jointly optimizes robot morphology and reward functions using large language models. It resides in the 'LLM-Driven Reward and Morphology Co-Design' leaf, which contains only two papers including this one. This is a notably sparse research direction within the broader taxonomy of 46 papers across 36 topics, suggesting that LLM-based approaches to simultaneous morphology-reward optimization represent an emerging rather than crowded area of investigation.

The taxonomy reveals several neighboring research directions. The sibling leaf 'Reward-Policy Co-Evolution' addresses reward-policy iteration without morphology changes, while parent-level branches like 'Hierarchical and Evolutionary Co-Optimization' and 'Gradient-Based and Differentiable Co-Design' tackle morphology-control coupling through traditional optimization methods. RoboMoRe diverges by substituting evolutionary or gradient-based search with LLM-guided generation, positioning it at the intersection of reward shaping and morphology optimization but using fundamentally different exploration mechanisms than gradient or evolutionary baselines.

Among 10 candidates examined, the core RoboMoRe framework contribution shows one refutable candidate from three examined, while the Diversity Reflection mechanism also identifies one refutable candidate from six examined. The coarse-to-fine optimization paradigm appears more novel, with zero refutable candidates among one examined. These statistics reflect a limited semantic search scope rather than exhaustive coverage, indicating that within the top-10 nearest neighbors, some overlapping prior work exists for the LLM-driven co-design concept and diversity generation, but the specific coarse-to-fine structure may be less anticipated.

Based on the top-10 semantic matches examined, the work appears to occupy a relatively novel position in applying LLMs to joint morphology-reward optimization, though the limited search scope means broader prior art may exist beyond these candidates. The sparse taxonomy leaf and modest refutation counts suggest incremental novelty over the single sibling paper, with the coarse-to-fine paradigm potentially offering the most distinctive contribution within the examined literature.

Taxonomy

Core-task Taxonomy Papers
46
3
Claimed Contributions
10
Contribution Candidate Papers Compared
2
Refutable Paper

Research Landscape Overview

Core task: joint optimization of robot morphology and reward functions. The field encompasses several major branches that reflect different emphases in co-design research. Morphology-Control Co-Optimization Frameworks focus on tightly coupling physical structure with control policies, often through evolutionary or gradient-based methods such as those seen in Data-efficient co-adaptation of morphology[2] and Co-optimization of Morphology and[7]. Reward Function Learning and Shaping addresses how to specify or discover objectives that guide both morphological and behavioral adaptation, including approaches that leverage large language models or human feedback. Task-Specific Co-Design Applications target domains like locomotion, manipulation, or social interaction, exemplified by works such as Co-design of a Social[4] and Task-driven co-design of mobile[19]. Representation and Search Space Design explores how to encode morphological parameters and control strategies efficiently, while Theoretical Foundations and Benchmarking provide principled analysis and standardized evaluation environments. Human-Centered and Participatory Co-Design emphasizes stakeholder involvement, particularly in assistive or companion robotics contexts like Robots for elderly care[9] and Co-Designing Companion Robots for[29]. Recent work has increasingly explored the interplay between automated design search and interpretable reward specification. A small handful of studies employ large language models to generate or refine both morphology proposals and reward signals, aiming to reduce manual engineering effort. RoboMoRe[0] sits within this LLM-driven strand, using language-based reasoning to co-design morphology and rewards in a unified framework. This contrasts with neighboring approaches like Debate2Create[37], which also harnesses LLM-based multi-agent debate but may emphasize different aspects of the design loop or task domains. Compared to more traditional evolutionary or Bayesian optimization pipelines such as Efficient Bayesian exploration for[13], RoboMoRe[0] leverages semantic priors from pre-trained models to guide exploration, potentially accelerating convergence on novel tasks. Open questions remain around how to balance automated LLM suggestions with domain constraints, ensure physical realizability, and generalize across diverse robotic platforms and objectives.

Claimed Contributions

RoboMoRe framework for joint morphology and reward optimization

The authors introduce RoboMoRe, an LLM-driven framework that simultaneously optimizes both robot morphology and reward functions in the co-design loop, addressing the limitation of existing methods that rely on fixed reward functions and only optimize morphology.

3 retrieved papers
Can Refute
Coarse-to-Fine optimization paradigm with Diversity Reflection and Morphology Screening

The authors propose a two-stage optimization strategy: a coarse stage that uses Diversity Reflection to generate diverse high-quality samples and Morphology Screening to filter unpromising candidates, followed by a fine stage that alternately refines morphology and reward to converge on optimal pairs.

1 retrieved paper
Diversity Reflection mechanism for generating diverse morphology-reward pairs

The authors introduce a task-agnostic Diversity Reflection mechanism where the LLM reflects on previously generated samples and deliberately produces new candidates that maximize diversity, improving both sample quality and exploration of the design space without requiring task-specific modifications.

6 retrieved papers
Can Refute

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

RoboMoRe framework for joint morphology and reward optimization

The authors introduce RoboMoRe, an LLM-driven framework that simultaneously optimizes both robot morphology and reward functions in the co-design loop, addressing the limitation of existing methods that rely on fixed reward functions and only optimize morphology.

Contribution

Coarse-to-Fine optimization paradigm with Diversity Reflection and Morphology Screening

The authors propose a two-stage optimization strategy: a coarse stage that uses Diversity Reflection to generate diverse high-quality samples and Morphology Screening to filter unpromising candidates, followed by a fine stage that alternately refines morphology and reward to converge on optimal pairs.

Contribution

Diversity Reflection mechanism for generating diverse morphology-reward pairs

The authors introduce a task-agnostic Diversity Reflection mechanism where the LLM reflects on previously generated samples and deliberately produces new candidates that maximize diversity, improving both sample quality and exploration of the design space without requiring task-specific modifications.

Table of Contents