GOAT Racer¶

GOAT Racer is two things that work together.

The car. A 1/10 scale race car with a robot brain. It carries an NVIDIA Jetson, a depth camera, a lidar, and a VESC motor controller. It drives itself with a policy that you train.
goatracer.ai. The cloud app. It teaches the labs and trains a driving policy on a GPU with one click. You watch the reward curve climb live. When the run ends you get a 3D playback of the policy driving and the policy file.

The loop is sim-to-real. A digital twin of the car trains in simulation on a GPU. The trained policy goes to the Jetson on the real car. On the car the policy turns lidar readings into steer and drive commands.

Getting started¶

Create an account, open the Labs page, and launch your first run.

Labs¶

What each Train lab observes, rewards, and passes.

The car¶

Safety, power-up, Wi-Fi, the dashboard, and the controls.

Send a policy¶

Put your trained policy on the car and drive the lap.

How the pieces fit¶

Piece	What it does
Labs	Short readings, a playground, and real Python in your browser.
Training run	One click starts a headless simulation on a GPU. 1024 copies of the car learn at once.
Policy	A small neural network. Observation in, steer and drive out. Exported as `policy.onnx`.
SIL check	The car's own runner software drives one simulated car with your policy. PASS means the software chain works.
Send to car	The policy goes to the car over the air. It appears in the car's policy library.

Support¶

Write to support@goatracer.ai. See FAQ and support for common questions.