Compute

SidewalkPilot splits work across three computers. The Jetson Orin Nano is the AI brain for the current camera-based self-driving system; the other computers connect that intelligence to the physical car and its display.

Board	Current responsibility	Link
Jetson Orin Nano	AI Model Manager: Series 1/2 PyTorch inference and Series 3/4 ONNX Runtime inference on the GPU	Direct Ethernet at 10.42.0.2:8770
Raspberry Pi 5	Camera capture, Xbox input, sensors, final safety arbitration, motors, steering, logs, and dashboard sender	Hardware buses, USB, Ethernet
Zero 2 W	Receives telemetry and renders the 64x32 HUB75 dashboard	USB Ethernet at 192.168.10.2:8765

Why It Is Split

The Jetson Orin Nano supplies the GPU performance that makes the current 320x180 Series 3/4 models practical at the selected live rate. If its result is missing, stale, or for the wrong model, current autonomy requests a stop. The Raspberry Pi 5 keeps the established Camera Module 3 Wide, steering/motor, sensor, and safety integration, while the Zero 2 W isolates panel rendering from the control loop.

The unified field runtime sends every steering-model family to the Jetson Orin Nano. Series 1/2 use PyTorch CUDA; Series 3/4 use ONNX Runtime CUDA. The larger 320x180 Series 3/4 networks ran much more slowly on the Raspberry Pi 5 CPU, while the Jetson Orin Nano GPU can run the selected models near the camera rate in the current deployment. Exact inference rate remains runtime telemetry and should be reported with the model, provider, software build, and power mode used for the measurement.

The Zero 2 W does not command the steering servo or motors. Losing the display removes observability but does not transfer control authority.