- Tesla's latest version of "Full Self-Driving" can back out of a spot, complete a journey and look for parking on the other end.
- The automatic parking features aren't fleshed out, though, and in one video the car attempts to park on a sidewalk ramp that resembles a driveway.
- FSD still makes plenty mistakes, so you should always be vigilant while using it.
We’ve been covering Tesla Full Self-Driving since it was launched in October 2020. In the four years since we’ve seen it both working perfectly and failing, but true full self-driving with no safety driver just seemed distant. The latest version of FSD changes that, though, and it finally makes true driverless cars seem possible.
There have always been big claims made by Elon Musk about FSD and its capabilities, but the rate at which it was progressing made some of these statements seem exaggerated. What made us question FSD was Tesla’s decision to make the system only work on a combination of cameras and neural networks, ditching other types of sensors and radar.
FSD version 13.2 is a big improvement over version 12. It not only makes the autonomous driving better and smoother, but it also adds new features. You can now enable FSD while the car is in Park, and it will reverse out of the parking spot for you and begin its journey to your set destination where it will find a suitable spot to park itself.
Dirty Tesla drove around in his Model Y running FSD V13.2, and it worked well on the driving portion. But at its destination, the vehicle attempted to park on a sidewalk ramp. It looked big enough for a car to fit, in its defense, though it's hard to imagine a human making that mistake. However, that was one of the very few times when the FSD disengaged or the driver had to intervene. Still, it highlights a big problem for automated parking: Knowing where the car will fit is one thing, but knowing where it can legally park is an entirely different issue. It may be a long time before a Tesla can sort out what a loading zone is, what permit parking means, what permits it has and what spots are time-restricted.
It also failed to find a proper parking spot after the sidewalk situation, but its ability to navigate the parking lot is still impressive. It even had to squeeze through a tight gap between a parked car and a building, and it made it look easy. Don’t forget this is a car with no ultrasonic sensors to gauge distance, which only makes its performance more impressive. Though you can see why many automakers do use them.
One of the big changes with V13.2 is that it uses higher-resolution feeds from the exterior cameras, which also run at a higher refresh rate. This allows the AI model that underpins FSD to run faster, helped by the new Cortex supercomputer at Giga Texas, which is five times quicker at computing data compared to the older supercomputer called Dojo.
Another video where Dirty Tesla tests FSD V13.2 highlights just how smooth it is with steering and brake application. Early versions would saw the wheel as they made turns, which didn’t instill confidence. But now it seems to be getting close to matching a human driver’s level of smoothness. Note, however, that early reviews of FSD systems tend to be more positive, as Tesla has been known to prioritize data access to those with friendlier approaches to the system.
Such a big improvement in FSD’s performance makes Tesla’s robotaxi plans seem more plausible. The manufacturer intends to begin production of its Cybercab sometime in 2026 or 2027. I was doubtful when I originally heard that timeline, and I'm still unsure now. But now that FSD has the ability to start and end drives on its own terms—even with notable errors—that future feels closer than ever.
