SurveyBrainBody — An Annotated Survey Hub for Embodied Co-Design

What it is. This GitHub repository hosts a comprehensive academic survey paper titled "Embodied Co-Design for Rapidly Evolving Agents: Taxonomy, Frontiers, and Challenges" (arXiv:2512.04770). The paper has been accepted with minor revisions by IEEE Transactions on Evolutionary Computation (impact factor 15.9). The repository serves as both the paper's companion site and a living, curated list of related research.

Core idea. The survey focuses on Embodied Co-Design (ECD): instead of designing a robot's body (morphology) and its control software (brain) separately, ECD optimizes both together. The README explains that this mirrors how animals in nature evolve body and brain in tandem, and it argues that joint optimization leads to better environmental interaction and task performance than optimizing control alone.

What the survey covers.

• Definitions of ECD and its relation to fields like evolutionary robotics, artificial life, and computational robot design.
• A new hierarchical taxonomy that classifies design spaces and representation methods for control, morphology, and tasks.
• Analysis of over one hundred recent ECD studies.
• Reviews of benchmarks and applications in both simulation and the real world.
• Identification of open challenges and future research directions.

Repository structure. The README organizes the literature into several categories:

• Bi-Level Co-Design — methods that treat control learning and morphology evolution as two nested levels (e.g., evolutionary reinforcement learning, surrogate-assisted methods).
• Single-Level Co-Design — methods that optimize morphology and control in one unified process (evolutionary, reinforcement-learning, physics-based, diversity-based, and other approaches).
• Generative Co-Design — using rule-based, latent-based, or LLM/VLM-based generation for designs.
• Open-Ended Co-Design — including brain-body-environment co-evolution and developmental co-evolution.
• Theoretical and Experimental Analysis — studies that analyze ECD from a formal or empirical standpoint.
• Simulated Benchmarks and Real-World Embodied Co-Design — practical testbeds and deployments.
• Other Surveys Recommended — pointers to further reading.

Each category contains a table of papers with titles, publication venues, agent types, and links to code or project pages where available.

Key components explained in the README.

• Controlling Brain — the software that processes sensory input and generates motor commands.
• Body Morphology — the physical form: shape, materials, sensor placement, etc.
• Task Environment — the conditions and feedback that shape learning.
• Co-Design Algorithms — described as "a machine that can design other machines," these algorithms simultaneously refine control, morphology, and environmental interaction.

How to contribute. The README invites pull requests to add missing papers. Contributors should fork the repo, add entries in a specified table format (Title, Paper link, Published in, Agent Type, Code/Project link), and submit a pull request.

Extras. The repository links to a YouTube video, several articles (including one from MIT Technology Review), and provides a citation format for the paper. The update log shows the paper was posted to arXiv in December 2025 and accepted in June 2026.

Bottom line. This is a well-organized, actively maintained survey resource for researchers and engineers interested in the emerging field of co-designing robot bodies and brains together — useful both as a reference and as a starting point for finding relevant papers and code.