robots

Full Self-Driving (Supervised)

Source-backed map of Tesla FSD — product naming, AI stack claims, supervised vs unsupervised, and evaluation discipline.

Topics: robots autonomy

Tags: fsd tesla autonomy

Tesla markets Full Self-Driving (Supervised). The parenthetical is load-bearing. Muskularity treats FSD as a living software product with versioned behavior, not a binary “solved autonomy” switch.

Official AI stack language (Tesla)

From tesla.com/AI (checked 2026-07-12), Tesla describes work across:

AreaOfficial framing (summary)Label
Vision + planning AIPath to general solution for FSD, bipedal robotics and beyondCompany technical claim
FSD chipCustom inference silicon, perf-per-watt focus, mass production in vehiclesCompany hardware claim
Neural networksPerception, BEV networks, fleet-sourced hard cases; large training buildsCompany ML claim — numbers on page are Tesla’s
Autonomy algorithmsWorld representation, trajectory planning under uncertaintyCompany software claim
Code foundationsLatency, determinism, OTA, multi-SoC pipeliningCompany systems claim
Evaluation infrastructureOpen/closed loop, HIL, fleet clip suitesCompany eval claim

Rule: Tesla’s training-hour or network-count marketing figures are company claims with a check date — not third-party audits.

Naming discipline

TermUse
FSD (Supervised)Current consumer product naming on Tesla site
AutopilotOlder / adjacent suite — do not equate casually
Unsupervised / robotaxi autonomySeparate claim set; requires operational evidence
SAE levelsUse only with cited standard + Tesla’s own wording

Safety and regulation

  • Safety statistics need Tesla safety pages / regulators, not social clips
  • Disengagement narratives need methodology
  • Regulatory approvals are jurisdiction-specific — never globalize one city’s pilot

Source anchors