Skip to content

Getting started with goatracer.ai

This page takes you from an invite code to your first training run.

1. Create an account

  1. Open goatracer.ai and click Register.
  2. Enter your name, your email, and a password of at least 8 characters.
  3. Paste the invite code from your instructor. The code puts you on your team.
  4. Confirm your email from the message the app sends you.

If you have no code, leave your email on the register page. The team sends a code when seats open.

Note

An instructor can also create accounts for a team. Ask your instructor if the register page says registration is closed.

2. The Labs page

Log in and open Labs. The page shows every lab as a node on a map. The map has bands. Each band has a goal.

  • A lab with a Log in to start badge needs an account.
  • A lab with a Coming soon badge has no content yet.
  • A Done badge means you finished it.

Labs do not gate each other. Pick any lab. The map suggests an order, and you can skip around.

Every lab has three tabs at the top right of the page.

Tab What it holds
Read The manual for the lab.
Play The interactive part. In the Train labs this is the launch form.
Code Real Python, edited and run in your browser. Nothing you write leaves your browser.

3. Launch a training run

Open a Train lab and go to the Play tab. The launch form has three main knobs.

Knob What it does
Parallel environments How many copies of the car train at once. A fixed menu: 256, 512, 1024, 2048, or 4096.
Iterations How many rounds of drive-then-update the run does.
Seed The random start. The same seed gives the same curve.

The form shows an estimate of the cost in pod-minutes before you click. Click Start training. The browser never talks to the GPU. The app starts the run on a GPU for you and streams the results back.

4. What a run shows

Output What it is
Reward curve The live chart. Mean reward per iteration. A healthy run climbs and then levels off.
Log tail The last lines from the trainer. Watch the iteration counter and the steps per second.
Playback The trained policy driving, played back in 3D in your browser when the run completes.
policy.onnx The exported policy file. This is the same format the car loads.

Every run is kept. Open Gym to see your previous runs with their knobs and results.

5. Credits

GPU time is metered in pod-minutes. The meter runs from the moment your GPU pod boots until it shuts down. Idle time counts. A SIL check is GPU time too.

  • A new account gets 60 free minutes. They expire after 30 days.
  • The daily cap is 120 minutes.
  • A car pack grants 600 minutes per month for 3 months. Unused minutes roll over for the 90 days.

The meter sits in the page header. It shows minutes left. Amber means one more session, then you are out. Red means a launch is refused or cut short. Click the meter to see your run audit.

TODO (Rob): confirm the free-minute and daily-cap numbers for launch. They are the app defaults today.

Next