FSD (Full Self-Driving)

FSD stands for “Full Self-Driving.” It refers to a software package from automaker Tesla that is purchased as an add-on to the car for an extra fee. When active, the car steers itself, keeps distance, changes lanes, turns, and stops at red lights. However, the name promises more than the system is allowed to deliver: the person behind the wheel must continuously monitor the drive and be able to intervene immediately. Tesla itself now calls the feature “FSD (Supervised),” meaning supervised driving. Legally, the driver remains responsible for everything that happens.

Why the name causes controversy

Experts divide automated driving into levels from 0 to 5. Level 2 means: the car steers and brakes, while the human monitors permanently. Only from level 3 onward is one allowed to take one’s eyes off the road temporarily. Despite its name, FSD is a level 2 system. This is exactly what fuels the criticism, since the product name suggests level 5.

This is not merely a play on words. If drivers trust the system too much, they react too late to situations the software misjudges. US traffic authorities have therefore launched investigations into Tesla multiple times, for instance after accidents in poor visibility. Advertising claims about FSD have also already been the subject of proceedings.

For investors, FSD is also a financial matter. Tesla has been selling the package for years for several thousand euros or via subscription. A portion of this revenue may only be booked once the promised features have actually been delivered. And the expectation of later turning FSD into a network of driverless robotaxis is embedded in a significant part of the company’s stock market valuation.

Cameras instead of lasers: the technical approach

FSD works almost exclusively with cameras around the car. Other manufacturers additionally rely on lidar, a sensor that measures distances using laser pulses. Tesla deliberately forgoes this and argues that humans, after all, drive using only their eyes as well. Cameras are cheaper but more sensitive to fog, glare, and darkness.

A neural network analyzes the images. This is a program that was not programmed rule by rule, but instead learned from huge amounts of example videos. Older versions of FSD combined this learning with hand-written traffic rules. Newer versions operate largely “end-to-end”: a single network receives the camera images and directly outputs steering and braking commands.

The advantage of this approach is fluidity, the disadvantage is opacity. If the car brakes for no apparent reason, it’s hard to say which rule just applied, because there is none. Training is done using driving data from Tesla’s own fleet, which gives the company a significant data advantage. The computing work inside the car is handled by a purpose-built chip that must process the images in real time.

Where FSD shows up on the road and in the headlines

In Germany, FSD has so far not been approved in its full scope. Only limited assistance functions are permitted here, because European regulations are stricter than those in the US. In the US and in China, on the other hand, hundreds of thousands of vehicles drive with the system. Anyone watching videos of it should keep in mind: they mostly show successful drives, not the cases involving intervention.

In business news, you mainly encounter FSD in three contexts. First, in quarterly earnings, when Tesla books revenue from the package. Second, in regulatory announcements about recalls or investigations. Third, in announcements surrounding robotaxis, meaning ride services without a driver.

A common misconception is confusing FSD with “Autopilot.” Autopilot comes standard on Teslas and can essentially maintain lane and distance on the highway. FSD is the expensive upgrade that also operates in city traffic. And both are clearly distinct from systems like Waymo, which genuinely operate without a driver within defined areas.