Yuxing-Wang-THU/SurveyBrainBody

What it solves

This project addresses the challenge of developing intelligent agents by moving away from traditional methods that focus solely on control optimization. Instead, it promotes "Embodied Co-Design" (ECD), which jointly optimizes an agent's physical body (morphology) and its controlling brain (software) to improve environmental interactions and overall task performance.

How it works

ECD treats the creation of an agent as a simultaneous optimization of four critical components:

1. Controlling Brain: The software that handles perception-action coupling to generate motor responses.
2. Body Morphology: The physical hardware, including shape, materials, and sensor placement.
3. Task Environment: The specific conditions and feedback loops that shape the agent's learning objectives.
4. Co-Design Algorithms: The optimization engines (such as Evolutionary Reinforcement Learning, Generative models, or Physics-based methods) that refine the brain, body, and environment interactions together.

Who it’s for

This resource is designed for researchers and developers in the fields of robotics, artificial life, and embodied AI who want to understand the taxonomy, current benchmarks, and state-of-the-art methods for jointly designing robot bodies and controllers.

Highlights

• Comprehensive Taxonomy: Provides a hierarchical classification of design spaces and representation methods for control, morphology, and tasks.
• Broad Scope: Covers a wide array of agents, including rigid robots, soft robots, aerial robots, and virtual creatures.
• Diverse Methodologies: Categorizes co-design approaches into Bi-Level, Single-Level, Generative, and Open-Ended co-design.
• Extensive Literature Review: Analyzes over one hundred recent studies and provides a curated list of latest works with links to code and papers.