A Survey: Learning Embodied Intelligence from Physical Simulators and World Models

This survey provides a comprehensive overview of research at the intersection of physical simulators and world models for advancing embodied intelligence in robots. It targets researchers and engineers seeking to understand the landscape of learning embodied AI, offering a structured perspective on simulators, world models, and their applications, ultimately aiming to accelerate the development of more autonomous and adaptive robotic systems.

How It Works

The project synthesizes research on two foundational pillars of embodied intelligence: physical simulators, which offer controlled environments for training and evaluation, and world models, which enable robots to build internal representations for predictive planning and decision-making. It explores how the synergy between these technologies enhances robot autonomy and task performance across diverse domains.

Quick Start & Requirements

The provided README outlines a survey of research and does not include installation instructions, code execution commands, or specific system requirements for a runnable project.

Highlighted Details

• Proposes a novel capability grading model (IR-L0 to IR-L4) for intelligent robots, integrating "intelligent cognition" and "autonomous behavior" dimensions.
• Features extensive tables categorizing research in robotic mobility, dexterity, and interaction techniques, including Model Predictive Control (MPC), Whole-Body Control (WBC), Reinforcement Learning (RL), and Vision-Language-Action (VLA) models.
• Provides a detailed overview of mainstream physical simulators (e.g., Webots, Gazebo, MuJoCo, Isaac Gym/Sim) and various world model architectures (e.g., RSSM, JEPA, TSSM, Diffusion).
• Includes dedicated sections on the application of world models for autonomous driving and articulated robots, with comprehensive lists of relevant papers and resources.

Maintenance & Community

The README lists multiple authors but provides no information regarding project maintenance, community channels (like Discord/Slack), or a public roadmap.

Licensing & Compatibility

No license information is specified in the provided README content. Consequently, compatibility for commercial use or closed-source linking cannot be determined.

Limitations & Caveats

As a survey paper, this repository primarily serves as a curated collection of research references and does not offer executable code or a practical framework for direct implementation. Users seeking to build or adopt specific technologies will need to consult the cited papers individually.