Skills

Robotics 6: Robot learning

6 weeks · 2 milestones

Teach a robot with reinforcement learning or demonstrations, then cross the sim-to-real gap. Stage 6 of the Varsity Curriculum: Robotics. A simulation-only route counts at every step; free resources only.

Part of: Robotics · Proves for: Engineers, UG

Milestone map

2 milestones

Train a robot policy in simulation

3–5 weeks (about 8 hrs/week, est)

Build a robot policy trained in simulation with reinforcement learning or imitation from demonstrations you recorded, with domain randomisation, evaluated over at least 5 seeds. Keep it in one public repo and commit as you go. No hardware needed: this milestone is done in simulation.

Proof required

Link the public repo with the training code, the environment, learning curves over at least 5 seeds, and success rates with and without domain randomisation.

What gets checked

Curves cover at least 5 seeds.
Success rate is measured on held-out conditions.
The effect of domain randomisation is measured.

Common mistakes

One big upload with no commit history.
Results the repo cannot reproduce.

Powstik Guide

A public repo with a robot policy trained in simulation with reinforcement learning or imitation from demonstrations you recorded, with domain randomisation, evaluated over at least 5 seeds.

Steps

Make a public repo.
Set up the simulator from its free docs.
Build the smallest version that works, then measure it.
Write a README: what it is, how to launch it, what the numbers show.
Ask someone who knows robotics to check it.

Template

README
Stage 6: Robot learning
What is here:
How to launch it:
Results:
Limits:

What gets sent back

No measurements.
Code that does not work from the README.
Copied code with no history.

Resources

Foundationstart here

What a verifier looks for

Curves cover at least 5 seeds.
Success rate is measured on held-out conditions.
The effect of domain randomisation is measured.
Open the repo: does the code produce what the recording shows?

You'll sign in first, then come straight back here.

Show the policy to a person

1 week (about 4 hrs, est)

Show the trained policy acting on hardware (sim-to-real), or in a new simulated scene you did not train on, while someone watches. Do it in person or on a live video call. Simulation-only route: if you have no hardware, a live run in the simulator counts.