An educational series exploring the foundations of embodied artificial intelligence - systems that perceive, reason, and act in the physical world.
This repository contains a comprehensive learning path that connects mathematical foundations to real-world robotic systems. Using a quadruped robot as a running example, the material progresses from basic machine learning concepts through to autonomous navigation and control.
The content is organized into three phases:
Phase I: Foundations
Core concepts in machine learning—data representations, neural networks, optimization algorithms, and training methodologies.
Phase II: Building Intelligence
Advanced topics including computer vision, natural language processing, multimodal learning, and agent architectures that enable perception, language understanding, and planning.
Phase III: Integration & Reality
Reinforcement learning for locomotion, bridging simulation and real-world deployment, and complete system integration.
This material is designed for students and practitioners seeking to understand how modern AI systems - from language models to autonomous robots - work from the ground up. Each section builds on previous concepts, creating a coherent path from theoretical foundations to practical implementation.
- Neural networks and deep learning fundamentals
- Self-supervised learning and foundation models
- Computer vision (CNNs, Vision Transformers, segmentation)
- Natural language processing and large language models
- Instruction following and alignment techniques
- Multimodal perception and reasoning
- Agent systems and planning
- Reinforcement learning for continuous control
- Sim-to-real transfer and robustness