Human bodies have been carefully refined through the long process of evolution, enabling us to utilize our bodies to solve a multitude of tasks skillfully. This capability, natural to humans, remains challenging for embodied agents, such as robots. The difficulty arises because successful interactions are highly dependent on the synergy between brain, body, and environment. Recent breakthroughs in vision, language, and robotics have significantly improved the sensorimotor learning capabilities of embodied agents. Alongside the great potential of sensorimotor learning for understanding the world, the concept of Brain-Body Co-Design (BBCD), namely, Automated Agent Design has attracted considerable cross-disciplinary attention. Unlike traditional agent design approaches, which focus primarily on optimizing an agent’s control mechanism ("brain") while keeping its morphological structure ("body") and task configurations ("environment") fixed, BBCD emphasizes their simultaneous coordination, to design embodied agents that are not only structurally sound but also behaviorally adept.
🔑 Contributions to Embodied AI: This paper surveys recent advancements in BBCD within the context of embodied intelligence. We introduce a novel taxonomy that provides a structured analysis of the representations, co-design spaces, and optimization frameworks employed in state-of-the-art BBCD methods. We review notable benchmarks and applications of BBCD in simulated and real-world environments. We identify significant open challenges and offer insights into promising future research directions.
We hope that this survey can serve as a reference-worthy and stimulating contribution to the advancement of embodied intelligence while also providing valuable perspectives for related fields.
2025/01/01: Happy New Year! I am updating the survey on this topic and will fully update the site when the survey is updated to Arxiv.
Feel free to pull requests or contact us if you find any related papers that are not included here. The process to submit a pull request is as follows:
- a. Fork the project into your own repository.
- b. Add the Title, Paper link, Published in, Agent Type, Page/Code link in
README.mdusing the following format:
|[Title](Paper Link)|Conference/Journal/Preprint|Agent Type|[Code/Project](Code/Project Link)|- c. Submit the pull request to this branch.
- We will update this page on a regular basis! So stay tuned~ 🎉🎉🎉. If you do find our survey or the repository helpful, please consider kindly giving a ⭐, 谢谢你, Thanks a lot, Спасибо, ありがとう, 감사합니다, Merci, Grazie, Obrigado, Danke, شكراً.
Here is a quick menu .^_^. :
- 🚀 Brain-Body Co-Design in Embodied Intelligence: Taxonomy, Frontiers, and Challenges
- 📋 Update List
- 🔥 Comments
- 🎥 Overview
- ⭐ Latest Works
- 1️⃣ Bi-Level Brain-Body Co-Design
- 2️⃣ Single-Level Brain-Body Co-Design
- 3️⃣ Generative Brain-Body Co-Design
- 4️⃣ Open-Ended Co-Design Methods
- 5️⃣ Theoretical and Experimental Analysis of Brain-Body Co-Design
- 💻 Simulated Benchmarks for Brain-Body Co-Design
- 🚪 BBCD in Real World
- 📖 Other Surveys Recommended
- ✉️ Contact Information
The creation of an embodied agent hinges on three critical components: Brain (control), Body (morphology), Environment (task), and the co-design algorithm (optimization) that optimizes these components.
1. Brain (Control): The controller (software) is responsible for perception-action coupling, enabling an agent to process sensory information and generate appropriate motor responses to control its body.
2. Body (Morphology): The physical embodiment (hardware) of an agent, including its shape, material properties, sensor placement, etc. Morphological design determines the agent’s physical capabilities and constraints, thereby impacting its performance and adaptability.
3. Environment (Interaction constraints): This includes both task-specific conditions and broader environmental configurations. It not only defines the challenge faced by an agent but also provides the necessary feedback for learning and adaptation. Task requirements and environmental dynamics are crucial for shaping the agent’s behavior and learning objectives.
4. Co-Design Algorithm (A machine that can design other machines): An effective co-design algorithm integrates the optimization of control, morphology, and environmental interactions. Unlike traditional methods that treat these components separately, co-design algorithms aim to simultaneously refine all aspects, leading to more efficient and adaptive agents. These algorithms leverage advanced techniques to explore and exploit the co-design space, optimizing for performance across diverse tasks and environments.
📖 We provide a detailed analysis of BBCD methods based on the proposed taxonomy. Please refer to our paper when it is available online.
The following picture introduces 4 general BBCD frameworks, with the last sub-figure showing an example of the BBCD process for creating a BipedalWalker agent.
TLDR: Methods that focus on the bi-level nature of control learning together with morphology evolution (Baldwin Effect)
TLDR: Methods that focus on using surrogate models to reduce the computational burden of the co-design process (how to efficiently evaluate a morphology without learning a specific controller?)
TLDR: Methods that focus on generating morphological diversity (how to effectively keep the morphological diversity?).
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| A ’MAP’ to find high-performing soft robot designs: Traversing complex design spaces using MAP-elites and Topology Optimization | IROS 2024 | Rigid Manipulator | N/A |
| Controller Distillation Reduces Fragile Brain-Body Co-Adaptation and Enables Migrations in MAP-Elites | GECCO 2025 | Modular Soft Robot | Code |
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Task-Agnostic Morphology Evolution | ICLR 2021 | Rigid-bodied Robot | Code&Page |
| Codesign of humanoid robots for ergonomic collaboration with multiple humans via genetic algorithms and nonlinear optimization | Humanoids 2023 | Humanoid Robot | Code |
| Evolution and learning in differentiable robots | RSS 2024 | Modular Soft Robot | Code&Page |
| Improving Efficiency of Evolving Robot Designs via Self-Adaptive Learning Cycles and an Asynchronous Architecture | GECCO 2024 | Modular Rigid Robot | - |
TLDR: Methods that focus on evolving morphology and control simultaneously (how does natural evolution inform the co-design process?).
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Evolving 3D morphology and behavior by competition | ALife 1994 | Rigid-bodied Robot | - |
| Automatic design and manufacture of robotic lifeforms | Nature 2000 | Rigid-bodied Robot | - |
| Evolving Complete Agents using Artificial Ontogeny | Morph 2003 | Rigid-bodied Robot | - |
| Generative representations for the automated design of modular physical robots | TRA 2004 | Rigid-bodied Robot | - |
| An Improved System for Artificial Creatures Evolution | ALife 2006 | Rigid-bodied Robot | - |
| Evolving virtual creatures and catapults | ALife 2007 | Rigid-bodied Robot | - |
| Solving deceptive tasks in robot body-brain co-evolution by searching for behavioral novelty | ICISDA 2010 | Rigid-bodied Robot | - |
| A Unified Substrate for Body-Brain Co-evolution | ICLRw 2022 | Vitual Creature | Code |
| Evolution of Developmental Plasticity of Soft Virtual Creatures in Changing Environments | CEC 2024 | Modular Soft Robot | N/A |
TLDR: Methods that focus on optimizing morphology and control simultaneously using Reinforcement Learning (physics-model free).
TLDR: Methods that focus on optimizing morphology and control simultaneously using Differentiable Simulation (physics-model based).
TLDR: Methods that focus on generating agent morphologies using rule-based systems (grammar, L-systems, etc.).
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Generative representations for the automated design of modular physical robots | TRA 2004 | Rigid-bodied Robot | - |
| Robogrammar: graph grammar for terrain-optimized robot design | TOG 2020 | Rigid-bodied Robot | Code&Page |
| Automatic Co-Design of Aerial Robots Using a Graph Grammar | IROS 2022 | UAV | - |
| Synergizing Morphological Computation and Generative Design: Automatic Synthesis of Tendon-Driven Grippers | IROS 2024 | Tendon-Driven Grippers | Code&Page |
TLDR: Methods that focus on generating agent morphologies using latent-based systems (GAN, VAE, etc.).
| Paper | Published in | Co-designed Agent | Code&Page |
|---|---|---|---|
| GLSO: Grammar-guided Latent Space Optimization for Sample-efficient Robot Design Automation | CoRL 2022 | Rigid-Bodied Robot | Code |
| Modular Robot Design Optimization with Generative Adversarial Networks | ICRA 2022 | Modular Rigid Robot | - |
| MorphVAE: Advancing Morphological Design of Voxel-Based Soft Robots with Variational Autoencoders | AAAI 2024 | Modular Soft Robot | Code |
| RoboNet: A Sample-Efficient Robot co-design Generator | CoRLw 2024 | Rigid-bodied Robots | - |
| Generating Freeform Endoskeletal Robots | ICLR 2025 | Endoskeletal Robots | Code&Page |
TLDR: Methods that focus on generating agent morphologies using large models (Diffusion Model, LLM, etc.).
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Evolution Through Large Models | Arxiv 2023 | Robot Gripper | Code |
| How can LLMs transform the robotic design process? | Nature Machine Intelligence 2023 | Robot Gripper | - |
| DiffuseBot: Breeding Soft Robots With Physics-Augmented Generative Diffusion Models | NIPS 2023 | Soft Robot | Code&Page |
| LLM-POET: Evolving Complex Environments using Large Language Models | GECCO 2024 Companion | Modular soft robot | N/A |
| RoboMorph: Evolving Robot Morphology using Large Language Models | Arxiv 2024 | Rigid-bodied Robot | - |
| LASeR: LASeR: Towards Diversified and Generalizable Robot Design with Large Language Models | ICLR 2025 | Modular Soft Robot | Code |
| RoboMoRe: LLM-based Robot Co-design via Joint Optimization of Morphology and Reward | Arxiv 2025 | Rigid-bodied Robot | - |
| Large Language Models as Natural Selector for Embodied Soft Robot Design | Arxiv 2025 | Modular Soft Robot | Code |
TLDR: Methods that focus on brain-body-environment co-optimization.
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Co-Designing Manipulation Systems Using Task-Relevant Constraints | ICRA 2022 | Rigid Manipulator | N/A |
| Curriculum Reinforcement Learning via Morphology-Environment Co-Evolution | Arxiv 2023 | Rigid-bodied Robot | N/A |
| Task2Morph: Differentiable Task-inspired Framework for Contact-Aware Robot Design | IROS 2023 | Rigid-bodied Robot | Code |
| Brain–body-task co-adaptation can improve autonomous learning and speed of bipedal walking | Bioinspir. Biomim. 2024 | Legged Robot | N/A |
| LLM-POET: Evolving Complex Environments using Large Language Models | GECCO 2024 Companion | 2D Modular Soft Robot (Voxel-Based Soft Robot) |
N/A |
| Evolving Complex Environments in Evolution Gym using Large Language Models | ICASSPW 2024 | 2D Modular Soft Robot (Voxel-Based Soft Robot) |
N/A |
| Task-Based Design and Policy Co-Optimization for Tendon-driven Underactuated Kinematic Chains | Arxiv 2024 | Rigid Manipulator | Page |
TLDR: Actuating Shape-Changed robots.
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| Shape change and control of pressure-based soft agents | ALIFE 2022 | Pressure-Based Soft Robot | Code&Page |
| DittoGym: Learning to Control Soft Shape-Shifting Robots | ICLR 2024 | Soft Shape-Shifting Robots | Code&Page |
TLDR: Papers that investigates the synergy of Brain, Body, and Environment.
| Paper | Published in | Co-Designed Agent | Code&Page |
|---|---|---|---|
| A comparative analysis on genome pleiotropy for evolved soft robots | GECCO 2022 | 3D Modular Soft Robot | Code |
| Subtract to adapt: Autotomic robots | RoboSoft 2023 | Modular Soft Robot | N/A |
| A comparison of controller architectures and learning mechanisms for arbitrary robot morphologies | SSCI 2023 | Modular Rigid Robot | N/A |
| How Perception, Actuation, and Communication Impact the Emergence of Collective Intelligence in Simulated Modular Robots | ALIFE 2024 | Modular Robot | N/A |
| Co-Optimization of Robot Design and Control: Enhancing Performance and Understanding Design Complexity | Arxiv 2024 | - | Code |
| Investigating Premature Convergence in Co-optimization of Morphology and Control in Evolved Virtual Soft Robots | EuroGP 2024 | Modular Soft Robot | N/A |
- Embodied Intelligence: A Synergy of Morphology, Action, Perception and Learning
Huaping Liu, et al., 2025, ACM Computing Surveys - Evolutionary robotics and open-ended design automation
Hod Lipson, 2005, Biomimetics - Exploring Embodied Intelligence in Soft Robotics: A Review
Zikai Zhao, et al., 2024, Bio-Inspired and Biomimetic Intelligence in Robotics - Collective Intelligence for Deep Learning: A Survey of Recent Developments
David Ha and Yujin Tang, 2022, Collective Intelligence - Bridging evolutionary algorithms and reinforcement learning: A comprehensive survey on hybrid algorithms
Pengyi Li and Jianye Hao, et al., 2022, IEEE Transactions on Evolutionary Computation - Design Optimization of Soft Robots: A Review of the State of the Art
FeiFei Chen and Michael Yu Wang, 2022, IEEE Robotics & Automation Magazine
This repo is developed and maintained by Yuxing Wang. For any questions, please feel free to email wyx20@tsinghua.org.cn.