An overview of embodied AI approaches

Start with the introduction to understand robot tasks, platforms, and the relationships between perception, decision making, control, and engineering support.

Three perspectives on a technique

Perspective	Question	Examples
Task	What should the robot accomplish?	Grasping and manipulation, navigation, legged locomotion, mobile manipulation
Method	How can we solve the problem?	Perception and state estimation, planning and control, imitation learning, reinforcement learning, VLA, world models
Engineering support	How do we train, run, and validate the system?	Data, simulation, evaluation, robot bodies, sensors, communication, ROS2

These perspectives overlap. For example, a robot arm can grasp objects using motion planning and feedback control, or by learning a policy from demonstrations. Both approaches need clearly defined observations, actions, and success criteria.

From understanding to practice

1. Start with the introduction to understand the complete loop of perception, decision making, and action.
2. Choose a project organized into chapters or a standalone experiment in projects. Record the results and analyze failures.
3. When you find a gap in your knowledge, look up the relevant topic in foundations.

Reference guides

• Embodied-AI-Guide: an index of algorithms, infrastructure, control, and hardware.
• PKU EPIC Lab · embodied-ai-start: an introductory guide from task definitions and robot skills to research methods.

This site uses these references to organize learning entry points and original explanations. Consult the linked tutorials and original projects for details.