Teach with it

Run this in your classroom.

Every lab here needs one thing: a browser. No accounts for your students, no software to install, no lab data leaving their machines. Below are lesson plans, ready-to-hand-out assignment links, and how a student proves they finished — without you keeping a roster.

Lesson plans

Five labs, ready to run.

Each is a self-contained session on one device. Time, what it teaches, what to watch for as students work, and questions that make the idea land.

WIRE · one period

Light an LED without the smoke

◷ 25–35 min

Ohm's law and why every LED needs a current-limiting resistor. Students place a battery, a resistor, and an LED, and a nodal-analysis solver reads the current live.

WATCH FOR — The moment a student drops the resistor to 47 Ω and the verdict flips to BURNING OUT — that is the lesson landing. Push them to predict the current before they read the meter.

DISCUSSION

  • Why does the same LED survive 470 Ω but cook at 47 Ω?
  • What happens to the current if you double the battery voltage? Predict, then check.
  • The button in series does nothing until pressed — what is it, electrically?
Open the lab →
SENSE→ACT · one period

The whole loop, in one screen

◷ 20–30 min

That a robot is a loop: read a sensor, decide, move, repeat. Students set the rule that turns a distance reading into a motor command and watch the agent behave.

WATCH FOR — Whether students can point at where 'sense' ends and 'act' begins. The magnet card reminds them the physical layer is real, not abstract.

DISCUSSION

  • Where in the loop would a delay make the robot crash?
  • What is the simplest rule that keeps the agent off the wall?
  • Which of the four verbs is this lab — and which come next?
Open the lab →
SENSE · one period

Teach a camera two words

◷ 25–35 min

Supervised learning from the student's own webcam — show examples of two classes, and a tiny classifier learns the boundary on-device. Perception as a learned function, not a rule.

WATCH FOR — The temptation to over-train on one lighting condition. Have them test in a new pose or light and watch accuracy drop — the generalization lesson, felt.

DISCUSSION

  • Why does it fail when you move to a new background?
  • How few examples can you get away with? Find the edge.
  • Nothing left your device. Why does that matter for a camera?
Open the lab →
ACT · one–two periods

Clone an expert arm

◷ 35–50 min

Behavior cloning: record a few demonstrations of reaching a target, train a small policy to imitate them, then watch it generalize to new targets. The core recipe behind the LeRobot generation.

WATCH FOR — Students expecting one demo to be enough. Let it fail sparse, then add demonstrations and watch the reach get reliable — data, not cleverness.

DISCUSSION

  • Why does the cloned policy reach targets it never saw?
  • What kind of target would break it — and why?
  • This is the SO-101 recipe. What would deploying it to a real arm add?
Open the lab →
IMAGINE · one period

Plan inside a dream

◷ 30–40 min

World models: the agent learns to predict what happens next, then plans by imagining rollouts instead of acting for real. Why 'learn a simulator' beats trial-and-error.

WATCH FOR — The contrast panel — the dreamed policy solves a rotated task the model-free one flees. That gap is the whole argument for imagination.

DISCUSSION

  • Why can planning in imagination beat acting in the world?
  • Where would a wrong world-model send the plan off a cliff?
  • What did the agent have to learn before it could dream?
Open the lab →
Assign a sequence

Hand out a link, not a login.

Each of these is just an ordered set of lab links. Share the list — students click through in order. Nothing to enroll in, nothing that tracks them.

SEQUENCE

One-period intro to Physical AI

A single class that goes from a circuit to a learning loop. No install, no hardware.

  1. Warm up: light an LED
  2. See the whole loop
  3. Teach a camera
SEQUENCE

The four verbs, in order

A short unit — one lab per verb — that builds the sense → think → act → imagine arc.

  1. Sense
  2. Act (clone an arm)
  3. Imagine (plan in a dream)
  4. The full framing
SEQUENCE

From perception to a policy (course)

The graded course, PAI-220 — seven lessons, each with a live in-browser bench and a verifiable certificate at the end.

  1. Start the course
  2. The credential path

Students can hand their work back the same way — a circuit folds into a shareable link in the gallery. No upload, nothing to collect.

Proof of completion

The credential is the receipt.

You do not need a dashboard of student data to know who finished. When a student completes a graded course, they earn a verifiable credential — an Open Badges 3.0 / W3C credential signed by the Institute. They show it to you; anyone can verify it cryptographically, and it carries the standards it aligns to. No roster, no server, nothing of theirs for you to hold.

The credential ladder →The standards these align to →