Tesla FSD V15 Launches with 25x Improvement in Safety Handling Capability
nashnova research
Tesla rolled out FSD V15 Supervised on November 21, boosting safety-critical scenario handling by 25× and sharing its vision encoder with Robotaxi for the first time — passenger-car driving data will now feed directly into the driverless-taxi pipeline.
What actually changed in V15?
The core upgrade is a fully retrained vision encoder — the module that converts camera feeds into machine-readable signals. Tesla trained it on millions of internet videos to improve generalization in unfamiliar road conditions.
V15 introduces Tesla Reality View: the car's screen overlays live camera footage with FSD's driving plan, letting the driver see the system's decisions in real time.
This means → FSD's planning logic was previously a black box. V15 puts the reasoning on screen — trust-building shifts from "trust me" to "watch me do it."
How much did everyday driving improve?
Detection coverage expanded significantly: road signs, small animals, road debris, and standing water are now actively avoided. A new feature auto-decelerates when a speed camera is detected.
Speed control in school zones and severe weather received dedicated tuning — both were high-complaint scenarios.
Navigation now draws on fleet route-preference data, using end-to-end learning to optimize origin-destination choices and recommending parking spots based on a driver's history. In plain terms = the more the fleet drives, the smarter each car's navigation gets.
Why is "sharing architecture with Robotaxi" the key signal?
V15's vision encoder will be shared across three product lines: FSD passenger cars, Robotaxi driverless taxis, and the Optimus humanoid robot.
This means → driving data from millions of Teslas commuting daily doesn't just improve their own assisted driving — it feeds directly into Robotaxi training, forming a data flywheel (more users → more data → stronger models → more users).
This reflects Tesla's core bet: scale with passenger cars to harvest data, then use that data to accelerate driverless-taxi commercialization. Whether the flywheel actually spins is the key variable for markets assessing Tesla's autonomous-driving timeline.
Are buyers on board? What do subscription numbers say?
In Q3 2026, over 75% of North American new-vehicle deliveries included an FSD subscription — up nearly 20 percentage points from the prior quarter.
Global active FSD subscriptions reached 1.74 million by quarter-end, up 67% year-over-year.
In plain terms = three out of every four new Teslas delivered in North America now carry FSD. Adoption is shifting from "optional add-on" to "expected default."
Where does the China rollout stand?
FSD Supervised is now live in 20 markets including the US, Canada, China, Australia, South Korea, and the Netherlands.
According to China's *Daily Economic News*, Tesla China's customer service said on November 22 that the company is "actively advancing preparations for its driver-assistance software in line with national regulations."
Tesla emphasized that FSD Supervised remains a driver-supervised assistance system — the driver must stay attentive at all times. This classification is especially significant under China's regulatory framework.
What comes next?
Musk said V15 will roll out to a wider user base in December, calling this version "something special."
Future updates plan to extend fleet route-preference learning from origin-destination pairs to full driving routes.
This means → the market's near-term focus is on two things: user-feedback quality after the December-wide push, and how fast the shared architecture actually accelerates Robotaxi progress.
市场有风险,内容仅供研究参考,不构成投资建议。
