Deep Learning Robot Brain

A foundation model is a single brain that can drive many different robot bodies. Build a tiny version of one.

35 min+160 XP

Watch

See it happen in the real world.

Sergey Levine's lab is building models that learn from videos of people doing things. Once the model has watched enough humans pour coffee, it can guide a robot arm to pour coffee — even if the arm is a different shape from the human's. This is what people mean when they say 'foundation model for robots.' One brain, many bodies. The same idea that lets a single AI model answer about cooking and chemistry from the same network.

Watch this

Think

A question worth sitting with.

If one robot brain runs both a vacuum and a robot arm, what would it need to know about each body?

Build

Make something with your hands.

In Python or Scratch, write a tiny brain that takes three inputs (battery, distance, light) and outputs one action (move, stop, turn). Run it on five different sets of inputs and see what it does.

Step-by-step

Pick three inputs your tiny brain will read: battery level (0–100), distance to nearest object (0–500 cm), and light level (0–100).
Pick the four actions your brain can output: MOVE, STOP, TURN, or RECHARGE.
Write the rules. Example: 'If battery < 20, RECHARGE.' 'If distance < 30, TURN.' 'If light < 10, STOP.' Otherwise MOVE.
Run the brain on five made-up scenes. (Battery 80, distance 100, light 70 → ?)
Pick the scene where your brain's answer felt wrong. Add or change one rule. That edit-and-retest loop is how every real robot brain gets smarter.

A four-rule robot brain

Python

def brain(battery, distance, light):
    if battery < 20:
        return 'RECHARGE'
    if distance < 30:
        return 'TURN'
    if light < 10:
        return 'STOP'
    return 'MOVE'

scenes = [
    (80, 100, 70),
    (15, 200, 50),
    (60,  20, 80),
    (90, 150,  5),
    (50, 300, 60),
]
for s in scenes:
    print('inputs:', s, '->', brain(*s))

Toolkit

Python
Scratch

Play

Test it. See what it does.

Be the brain. A friend describes a robot scenario in one sentence. You give the next action in one word. Do 10 rounds and see how often your one-word answer is the right call.

Challenge

Push it a little further.

Add a fourth input — sound. Update your tiny brain so it reacts when something loud happens nearby.

Reflect

Notice what your robot taught you.

What did your robot learn that surprised you?

Also ask yourself

What surprised you?

Reward

Mission outro

+0XP

“Next week your robot will learn to see.”

Skills advanced: foundation models, machine learning

Badge

AI Apprentice

Explore another mission →