← Teacher guides

Teacher guide: AI robots

Drive a robot with a model you trained, on the computer or on your wrist. Turn labels into moves, keep the robot safe when the model is unsure or goes quiet, and make it more careful where a mistake costs more.

Open the student course

Each lesson works two ways. Students can work through the online tutorial at their own pace (it checks their answers and saves their progress in their browser), or you can teach it to the whole class from the slides, then send students into the tutorial for the Modify and Make steps.

Every lesson guide has plain-language notes, the answers to every check, a solution for every task, and slides and a worksheet for Waypoints Prepare. These guide pages are not linked from student pages, but they are not secret either: keep printed Teacher editions for answers you do not want students to find.

Whole-course cheat sheet: Open to print (every command, grouped by unit: a handout for the final challenge)

Drive with a model

Labels from a webcam model or a wrist model become modes the robot acts on, with a safe action when the model is unsure or silent.

A robot driven by a model in 5 minutes (read before teaching this unit)

The idea in plain words

A model trained on gestures sends a word (Forward, Left, Stop) through the bridge to the robot by radio. The robot's program does the rest:

  • on radio received string remembers the word as the robot's mode; forever acts on the mode until a new word arrives.
  • The model can be on the computer (a webcam model and a bridge) or on a wrist micro:bit (a CreateAI model sending words by radio). The robot's program is the same either way: it only sees words.
  • So when the model goes quiet, the robot keeps going. Two safety nets fix that:
    • Unsure: the board sends it when no label is confident enough, and the robot stops.
    • A watchdog: the robot remembers when the last word arrived and stops if too long passes. The board repeats its label every second so a working robot never trips it.

Why programmers use it

Gesture-driven wheelchairs, drones and delivery robots all separate "what the model thinks" from "what the machine does", and all need a safe action for doubt and for silence.

An everyday picture, and where it breaks

A train driver must keep pressing a pedal (a dead man's switch): if they stop, the train brakes. The watchdog is that pedal. Where it breaks: if the signal is only sent when something changes, a steady "all fine" looks like silence, which is why the board repeats.

Try it yourself (5 minutes)

  1. In Drive with gestures, press A+B on one simulated micro:bit: the other robot drives and keeps going.
  2. In A wrist remote, open the editor full screen and choose actions in the ML panel.
  3. In When the model is unsure, press A+B once: the robot stops by itself after two seconds.

Questions students ask

  • "Why does it keep going?" Forever acts on the last mode until a new word arrives.
  • "Why does it stop while I'm still showing Forward?" The board isn't repeating, or the radio groups differ.

How it connects

Robotics challenges taught modes and radio; micro:bit and the computer taught the model and the bridge. Here they meet.

Cheat sheet: Open to print (A4, with a picture of what each command draws) · Markdown

Costly mistakes

Some mistakes matter more than others, so some actions need more confidence.

Costly mistakes in 5 minutes (read before teaching this unit)

The idea in plain words

Every model is sometimes wrong, and some mistakes cost more than others. A robot that mistakes Stop for Go can drive off a table; one that mistakes Left for Right only wastes time. So:

  • A higher confidence threshold for the risky action: the board here sends Go only at 95% (others at 70%).
  • The robot adds its own caution: it drives only after Go twice in a row, while Stop acts at once.
  • The cost of caution is more Unsure, so more stops. Put caution where mistakes are costly, not everywhere.

Why programmers use it

Self-driving cars, medical screening and spam filters all set thresholds by asking which mistake is worse: missing something, or a false alarm.

An everyday picture, and where it breaks

You cross a road only when you're sure it's clear, but you'll step back at the first hint of a car. Where it breaks: be too cautious everywhere and you never cross.

Try it yourself (5 minutes)

  1. Print the signs from Signs: some mistakes cost more and train Go, Left, Right and Stop.
  2. Hold Go far away or tilted: the board says Unsure while Left still counts.

Questions students ask

  • "Why won't it send Go?" It needs 95%: retrain Go with clearer, more varied photos.
  • "Why does the robot wait?" It needs two Gos in a row: about a second of steady Go.

How it connects

The delivery robot challenge adds the final rule: the robot's own sensor overrules the model.

Cheat sheet: Open to print (A4, with a picture of what each command draws) · Markdown

Final challenge: Delivery robot

A sign-driven delivery robot whose own sensor can overrule the model, built in milestones.

Cheat sheet: Open to print (A4, with a picture of what each command draws) · Markdown