Open driver-assistance
A self-driving computer where the whole system is reuse — except the eyes.
comma ships openpilot on open hardware. The compute, the CAN interface, the GPS are all standard reuse. The bottleneck — and the slice — is perception: how the system sees, and at what energy.
Exploded
Decompose the product into its stack. Each layer is colour-coded by what it is to the next build.
Bill of materials
Every part is a real, sourced component from the open baseline — no invented part numbers.
The perception front-end is the slice. openWafer’s move: event-based vision feeding an on-die spiking accelerator — compute only what moves, at microsecond latency and a fraction of the power, instead of reprocessing every pixel of every frame.
https://www.ovt.com/products/ox03c10/ →Mobile SoC reused for compute — refreshes on a generation clock.
https://blog.comma.ai/comma3X/ →Mature MCU bridging to the vehicle bus. Reuse.
https://blog.comma.ai/comma3X/ →Density refreshes each generation. Reuse.
https://blog.comma.ai/comma3X/ →Standard panel. Reuse.
https://blog.comma.ai/comma3X/ →Standard GNSS module. Reuse.
https://www.u-blox.com/en/product/neo-m8-series →Honest gap: comma’s public pages don’t name the IMU part — confirm in the github.com/commaai/hardware schematics. Reuse either way.
https://github.com/commaai/hardware →The slice openWafer designs
Driver assistance is bottlenecked on perception, and comma does it the standard way: three frame cameras stream into the Snapdragon’s NPU, burning watts to reprocess every pixel of every frame — most of which never changed. The compute box, the CAN interface, the GPS, the IMU are all reuse. openWafer’s slice is the eyes: an event-based vision sensor feeding an on-die spiking accelerator, so the system computes only what moves — microsecond latency, a fraction of the power. Perception is the slice; the rest is assembly.