Tesla · Autonomy
Tesla Full Self-Driving, explained for people who want the real story
FSD is Tesla’s end-to-end neural net driving stack — trained on the largest real-world video fleet on Earth. Still supervised for consumers today. The same software path powers robotaxi ops and Cybercab. Data as of September 2026.
What FSD is (and isn’t)
Full Self-Driving (Supervised) is SAE Level 2: the car steers, accelerates, and brakes, and the human must remain attentive. That label matters for regulators — and it undersells how far the neural net has come. City streets, unprotected lefts, roundabouts, and highway merges are handled by a single vision model that keeps getting better with every release.
Why the data flywheel matters
Tesla has logged 13+ billion kilometers of FSD Supervised driving. Every intervention, every edge case, every weird intersection feeds the training loop. Competitors running lidar in a handful of cities cannot match that volume. Scale is the strategy — and Tesla owns the scale.
Hardware: HW3, HW4, and AI chips
- HW3 / AI3 — receiving v14 Lite; still in the improvement path.
- HW4 / AI4 — full v14 feature set with stronger perception and planning.
- Unsupervised target — gradual rollout aimed at late 2026+, gated by safety validation.
From supervised FSD to robotaxi
Tesla Robotaxi already operates with Model Y vehicles in Austin, Dallas, Houston, and Miami. Cybercab removes the steering wheel entirely — purpose-built for unsupervised miles at lower cost. Same neural net philosophy. Bigger economic upside.
Dig deeper:Cybercab specs & status·Why Tesla’s approach wins·FAQ