— Framework / Vol. I / Stack
The Physical AI Stack.
From raw reality to intelligent action — and continuously better robots.
Five interacting layers turn the world into structured understanding, understanding into action, and action back into experience. Every card opens. Every component has a position.
Layer 01
World Model
Action
Embodiment
The robot exists in the physical world.
- Sensors
- Actuators
- Power
- Mechanics
- Mobility
- Tactile Systems
Perception
The robot builds a usable state of reality.
- Computer Vision
- SLAM
- Sensor Fusion
- Object Detection
- Localization
- Affordances
- State Estimation
World Model
& Cognition
The robot builds internal reality.
- World Models
- Memory
- Spatial Reasoning
- Semantic Reasoning
- VLA
- LLM Planning
- Scene Graphs
- Task Abstraction
Action
& Control
The robot changes the world.
- Planning
- Manipulation
- Motion
- Policies
- Control
- Execution
- Grasping
- MPC
- RL Control
Learning Loop
The system improves from experience.
- Feedback
- Telemetry
- Data Engine
- Simulation
- Fleet Learning
- RL
- Imitation Learning
- Continuous Adaptation
The Closed-Loop Flow
Flow · Real world → real world
Real World
Embodiment
Perception
World Model & Cognition
Action & Control
Real World Changes
Feedback + Data
Learning Loop
Orchestration & Runtime — The Nervous System
Infrastructure · cross-cuttingOrchestration & Runtime · the substrate every layer runs on
Closed-loop intelligence: Perceive → Understand → Act → Learn → Improve.