ComputeNEORACER DOCS
These docs are public and open source.Edit on GitHub
HARDWARE / COMPUTE

THE COMPUTE STACK.

One Linux host runs , your Python, the camera pipeline, and any neural net you want to deploy. A real-time microcontroller stack handles motor commands and so the high-level loop never blocks the wheels.

NVIDIA Jetson Orin Nano0 TOPS AI acceleratorUbuntu + ROS 2 HumbleWi-Fi 6
01 / WHAT THE JETSON DOES

THE JETSON HOST.

  • Runs Ubuntu (the NeoRacer image ships pre-configured).
  • Hosts the ROS 2 graph, every node from FIG. A on the ROS 2 driver page lives here.
  • Drives the camera pipeline at 640×480 / 60 fps.
  • Provides the AI accelerator for any TensorRT, ONNX, or PyTorch model you deploy.
  • Bridges Wi-Fi for SSH, ROS 2 topics, and OTA updates.
02 / WHAT THE MCU DOES

THE MICROCONTROLLER STACK.

The Jetson shines at perception and planning, but guaranteed real-time response is not its strength: a Python garbage-collection pause is longer than one control loop tick. That's where the microcontroller stack comes in. It streams serial frames to the Jetson over USB, the driver node bridges them into ROS 2, and it takes care of the time-critical loops the Jetson can't.

01

Motor control

PWM out to the ESC. Encoder counts in.
Closed-loop velocity control runs at a fixed kHz tick so you don't see torque ripple from a stalled Python loop.
02

Servo control

Steering angle commanded over PWM.
Trim and centre offsets are stored in flash so a re-flash of the Jetson doesn't lose your calibration.
03

IMU fusion

Fused orientation at 200 Hz.
The MCU (microcontroller unit) does the bias subtraction and orientation fusion so the Jetson sees clean orientation+angular-rate samples on /imu, no warm-up delay.
03 / WHO LIVES WHERE

THE WORK SPLIT.

Anything where timing has to be exact lives on the MCU, and everything else lives on the Jetson. In practice you'll spend about 99 % of your time writing code on the Jetson side. The MCU is usually something you configure rather than program.