Connect language to the world
Define tasks, frames, objects, state estimates, and success predicates precisely enough for a machine to act on them.
Embodied AI · from first principlesGround intelligence.
Ground intelligence.
Close the physical loop.
Start with a concrete sensor-to-action task, then add perception, demonstrations, feedback control, recovery, and evidence. Frame notation and implementation stay optional until the physical loop is clear.
30 lessons · about 14 hours
THE EMBODIED LOOP
sensors + instruction→state + policy
applied actionobserved consequenceillustrative flow · not a measurement
Bring intelligence out of the screen.
Embodied AI forces every abstract capability to meet time, geometry, uncertainty, and consequence. A robot must connect language to perception, perception to state, state to action, and action to a physical outcome that can fail in ways a text demo never reveals.
Close the loop under uncertainty
Build policies, feedback, recovery, authority handoffs, and timing contracts that respond to what actually happened.
Evaluate the system, not the demo
Track requested versus applied actions, interventions, failures, latency, and matched experiments in a bounded simulation.
The finish lineComplete the full perception–language–action loop and learn what it takes to make intelligence survive contact with reality.
From one sensor-action contract to an operated embodied research system.
Six territories make task semantics, state estimation, data lineage, policy decoding, feedback, authority, and experimental evidence observable before any physical-world claim.
NEURAL FIELD GUIDE / COURSE ARC01 · 06
Embodied Problems
Write a complete observation-action-task contract.
You unlockSimulation task with assertions and evidence boundary
Perception & State
Build a calibrated multimodal state estimator.
You unlockCalibrated estimator with changed-sensor tests
Demonstrations & Data
Create and audit trajectory data for policy learning.
You unlockTrajectory policy with compounding-error test
Language-Conditioned Policies
Build transformer and diffusion-style action policies.
You unlockTransformer/diffusion policy comparison
Planning, Control & Transfer
Operate a controller with feedback, recovery, and transfer evidence.
You unlockClosed-loop controller with safe fallback
Embodied Evaluation & Research
Run a bounded, reproducible embodied-system intervention study.
You unlockReproduced policy plus controlled cross-course intervention
Inside the course
One course. 6 connected frontiers.
Lessons 01–30 build a bounded perception-language-action system from first principles, finish each territory with an inspectable synthesis, and culminate in an original matched intervention study.
01
Specify body and taskEmbodied Problems
Write a complete observation-action-task contract. Each lesson extends one bounded perception-language-action system.
The payoffWrite a complete observation-action-task contract.
02
Turn sensors into statePerception & State
Build a calibrated multimodal state estimator. Each lesson extends one bounded perception-language-action system.
The payoffBuild a calibrated multimodal state estimator.
03
Record behaviorDemonstrations & Data
Create and audit trajectory data for policy learning. Each lesson extends one bounded perception-language-action system.
The payoffCreate and audit trajectory data for policy learning.
04
Map intent to actionLanguage-Conditioned Policies
Build transformer and diffusion-style action policies. Each lesson extends one bounded perception-language-action system.
The payoffBuild transformer and diffusion-style action policies.
05
Close the loopPlanning, Control & Transfer
Operate a controller with feedback, recovery, and transfer evidence. Each lesson extends one bounded perception-language-action system.
The payoffOperate a controller with feedback, recovery, and transfer evidence.
06
Test the whole systemEmbodied Evaluation & Research
Run a bounded, reproducible embodied-system intervention study. Each lesson extends one bounded perception-language-action system.
The payoffRun a bounded, reproducible embodied-system intervention study.
30connected lessons
30hands-on labs
30code notebooks
No black boxes. Build intuition, see the mechanism, then make the real engineering trade-offs.
Ready when you are
Complete the full perception–language–action loop and learn what it takes to make intelligence survive contact with reality.
30 connected lessons, hands-on labs, and a complete end-to-end build.