Physical AI & Humanoid Robotics
The Technology Behind Autosapien G1
No humanoid has yet demonstrated 8-hour unsupervised household operation. G1 will. This is the curriculum that gets us there.
Research-grade curriculum covering the 2024–2026 frontier. From QDD actuators to π0 VLA foundation models — every module informs a design decision in G1.
Modules: 10 · Lessons: 42 · Total hours: 68+ · Author: Dilawar Gopang
Who this book is for
- Robotics Engineers
- ML Researchers
- G1 Development Team
- Physical AI Students
Each lesson opens with learning objectives, builds the ideas in Overview and Key concepts, then asks you to apply them in a hands-on lab and test recall with Check your understanding. Work the labs — they are where the learning sticks.
Modules
Module 1: The Humanoid Landscape
2024–2030 Platform Analysis — Map the competitive terrain. Understand what Figure, Tesla, Boston Dynamics, and Unitree built — and what it means for G1.
5 hours · Foundational · 6 lessons
Module 2: Actuator Architecture
Electric Motors, Compliance & Power Systems — Every physical action G1 takes flows through its actuators. Understand the tradeoffs between torque density, backdrivability, safety, and energy efficiency.
6 hours · Intermediate · 5 lessons
Module 3: Bipedal Locomotion & Whole-Body Control
Walking, Balance & Loco-Manipulation — From ZMP to reinforcement learning: build a deep understanding of how modern humanoids achieve robust, dynamic locomotion across real-world terrain.
8 hours · Advanced · 5 lessons
Module 4: Perception & Spatial Intelligence
Sensors, SLAM & 3D Scene Understanding — G1 must understand its world in real-time: where objects are, what they are, and how to interact with them. Build the full perception stack from sensor fusion to semantic scene graphs.
7 hours · Intermediate · 4 lessons
Module 5: Foundation Models & VLA Architecture
π0, OpenVLA, Diffusion Policy & Robot Brains — The most consequential shift in robotics since deep learning: general-purpose neural policies that turn language + vision into robot actions. Understand the architecture powering the next generation.
9 hours · Advanced · 5 lessons
Module 6: Dexterous Manipulation
22-DOF Hands, Tactile Sensing & Grasp Planning — G1's 22-DOF hands must thread a needle, pick up an egg, and handle medication. Build the complete manipulation stack from hand design to learned in-hand policies.
7 hours · Advanced · 3 lessons
Module 7: Simulation & Digital Twins
NVIDIA Isaac Sim, MuJoCo & Sim-to-Real — Train G1's locomotion and manipulation policies in simulation before deployment. Build physics-accurate digital twins and master the sim-to-real transfer techniques that make it work.
6 hours · Intermediate · 3 lessons
Module 8: Safety & Human-Robot Interaction
ISO Standards, Compliant Control & Ethical AI — A humanoid that cannot safely share space with humans will never be approved for hospital or home deployment. Build G1's safety architecture from standards to implementation.
5 hours · Intermediate · 4 lessons
Module 9: Edge AI & On-Board Intelligence
NVIDIA Orin, Neuromorphic Chips & Real-Time Inference — G1 cannot depend on the cloud. Build a complete on-board AI stack that runs foundation models, locomotion controllers, and safety monitors within the compute and power budget of a mobile humanoid.
6 hours · Advanced · 3 lessons
Module 10: G1 Development Roadmap
Architecture, Milestones & 2030 Strategy — Synthesize everything. Define G1's technology stack, resolve the key architectural decisions, and lay out the concrete development roadmap from 2026 to commercial launch.
4 hours · Foundational · 4 lessons
👉 Start with the Humanoid Landscape: The Humanoid Landscape