Next machine
A new physical-intelligence mission will join the squad.
INCOMING BUILD LOG 001 · THE PHYSICAL WORLD
Intelligence,
made physical.
Learning projects for people who want to turn real vehicles and devices into capable intelligent systems—built for dirt, weather, work, and consequence.
PHYSICAL_INTELLIGENCE.pipelineRUNNING
01
SENSEObserve the real world
↓
02
UNDERSTANDEstimate state and intent
↓
03
PLANChoose a safe useful action
↓
04
ACTChange the physical world
SAFETY GATE: ARMEDHUMAN: ON LOOP
FIELD ROBOTICSAUTONOMOUS VEHICLESEDGE AIMACHINE CONTROLRUNTIME SAFETY
01 / MISSION
// WHY THIS EXISTS
Built for the real world.
Systems that do useful work.
AI leaves the screen when it commands a track, blade, motor, pump, wheel, or actuator. Then intelligence must become more than impressive—it must become measurable, observable, safe, and useful.
AI Squad Hub breaks that journey into open learning projects. Each project begins with a physical mission and builds the complete path from sensing to action.
02ACTIVE PROJECTS
45LAB LESSONS
∞REAL PROBLEMS
02 / PROJECTS
// ENTER THE LAB
Choose a machine.
Every project is a complete learning track with concepts, runnable practice, progress, and a measurable real-world destination.
PROJECT / 001
GNSS FIXED
GRADE Δ 0.042m
MODE AUTO
GRADE Δ 0.042m
MODE AUTO
ACTIVEEARTHMOVING · AUTONOMY
DODO
Teach a bulldozer to understand terrain, plan useful passes, control its tracks and blade, and safely grade a real worksite.
ENTER PROJECT →
PROJECT / 002
POLICY 50 HZ
OBS 61-D
MODE SIM
OBS 61-D
MODE SIM
ACTIVEBIPED · REINFORCEMENT LEARNING
MICRO
DUCK
Train a tiny open-source biped in MuJoCo, close the actuator reality gap, export its policy, and earn a careful path from simulation to hardware.
ENTER PROJECT →
“If it cannot sense failure, stop safely, and prove useful work, it is not ready.”THE SQUAD RULE
03 / PRINCIPLES
// HOW WE BUILD
The squad standard.
Shared constraints across every vehicle, robot, and intelligent device.
Mission before model
Begin with useful physical work and measurable acceptance—not an algorithm looking for a demonstration.
Simulation earns hardware
Prove interfaces, invariants, failures, and recovery before increasing mass, force, speed, or risk.
Safety is architecture
Independent limits, watchdogs, safe states, and human authority are part of the system from day one.
Evidence over spectacle
Repeatable tests, metrics, logs, and limitations matter more than one polished successful run.
AI
↘
WORLD
↘
WORLD