Tesla FSD ingests 37 billion pixels per inference cycle, says engineer

SchmidhuberAI · x · 2026-09-23

In a thread with Schmidhuber, Nico Chauvin points out autonomous driving has a tiny output domain (2 DoF: steering and speed) against an enormous input domain of vision, physics and audio. Example: Tesla FSD uses 7,500 frames of 5MP at each inference cycle — about 37 billion pixels.

Original post →

More from Embodied

Embodied channel →