One of Tesla's most-followed FSD testers made a direct claim today: Tesla's Full Self-Driving software is ready to operate without human supervision on customer vehicles. Whole Mars Catalog, who logs thousands of miles on FSD weekly, backed the assertion with a clear explanation of what 'streak' metrics actually prove — and shut down the recurring argument that LIDAR is the missing ingredient.
What a 10,000-Mile Streak Actually Means
The streak counter — introduced in FSD v14.3.3 (firmware 2026.14.6.6) and rolling out to early access owners since May 17 — tracks how many miles a vehicle has driven on FSD since the last human intervention. An intervention is defined as the driver taking the wheel, pressing the brake, or hitting the accelerator for safety reasons while FSD is active. The Self-Driving App permanently logs each user's longest intervention-free run.
Whole Mars Catalog spelled out exactly what a high streak number means in practice:
That's a meaningful distinction. A streak isn't a comfort metric or a gamification feature — it's a direct measure of autonomous competence. If a vehicle accumulates 10,000 miles without a crash and without the driver ever needing to intervene, the software demonstrably handled every situation it encountered on its own. At that scale, the data becomes hard to dismiss as anecdote.
The LIDAR Debate, Settled
Critics of Tesla's vision-only approach have long argued that cameras alone can't match the spatial precision of LIDAR sensors — and that adding LIDAR would unlock unsupervised capability. Whole Mars Catalog pushed back on that framing directly:
The argument aligns with Tesla's long-standing engineering position. FSD's remaining failure modes — edge cases, rare scenarios, ambiguous road geometry — aren't problems that better ranging hardware solves. They're problems that require the neural network to generalize better from training data. Tesla's v14.3.3 update reflects exactly this: the release includes an upgraded neural network vision encoder with improved understanding of rare and low-visibility scenarios, stronger 3D geometry reconstruction, and expanded traffic sign recognition. The AI compiler and runtime were also rewritten using MLIR, delivering a reported 20% faster reaction time.
Adding LIDAR wouldn't have produced those gains. Better deep learning did.
Where Unsupervised FSD Actually Stands
The tester's assessment arrives at a moment when Tesla's unsupervised ambitions are moving from roadmap to reality — though the gap between 'ready' and 'deployed at scale' remains real.
As of mid-May 2026, Tesla's unsupervised Robotaxi fleet is operating across three Texas cities — Austin, Houston, and Dallas — with no safety monitors inside the vehicles. The fleet is currently limited to fewer than 30 cars total, with Dallas and Houston each running roughly five to six vehicles. Elon Musk has stated publicly that unsupervised FSD will be widespread in the US by the end of 2026.
For owners of consumer vehicles, the timeline is more measured. A gradual rollout of unsupervised FSD for personal cars is projected to begin in Q4 2026 at the earliest, starting in geographically validated areas and prioritizing vehicles with AI4/HW4 hardware. A lighter version for HW3 hardware — referred to as V14 Lite — is under separate development. Musk has confirmed that the FSD 14.3 software stack is architecturally sufficient for unsupervised deployment; the remaining work is validation and regulatory clearance, not fundamental capability.
On the same day as these tweets, Tesla also announced that supervised FSD is now available in China, with full regulatory approval for that market anticipated by Q3 2026.
What This Means for Owners Right Now
If you're on FSD and haven't checked your streak metrics yet, now is the time. The live counter on the main display shows your current intervention-free miles; the Self-Driving App stores your all-time best. Both are available in v14.3.3 for early access owners, with broader rollout ongoing.
The streak number isn't just a personal benchmark — it's the same type of data Tesla and independent testers are using to build the case for removing the supervision requirement entirely. Every mile you log without intervention is a data point in that argument. Whether Tesla's regulators and internal validation teams reach the same conclusion as Whole Mars Catalog by Q4 is the question that matters most for what comes next.
Marcus covers Tesla's software releases, FSD rollouts, and OTA changes. Background in automotive engineering. Based in Austin.
Sources verified at publish time. Spotted an inaccuracy? Email editorial@basenor.com.
