Tesla FSD v12 feels more human — for better or worse — as 'end-to-end AI' autonomous driving matures
Tesla and its CEO, Elon Musk, have long held that Full-Self Driving Beta version 12 would be a revolutionary, thanks to a switch to what Musk calls “end-to-end AI.” In a nutshell, Tesla's FSD is now controlled by a machine-learning algorithm that has been trained to recognise its environment and issue commands, rather than the “AI” system doing the interpreting while commands are issued from a limited number of hand-coded vehicle inputs.
The idea is that “end-to-end AI” gives Tesla's FSD system more flexibility, and that largely seems to be true, except for the occasional oddity reported by beta testers. One FSD beta tester, who goes by @MissJilianne on X (formerly Twitter), posted about her experiences with the FSD v12.2.1 update, commending it on handling complicated interactions, like pedestrians crossing at a cross-walk and merging onto busy streets more fluidly than before.
However, FSD v12 isn't perfect. In one of her videos, her Tesla Model S manages to navigate to a destination, safely get through the parking lot, and successfully find and park in an empty parking lot. It's the last step where the Tesla comically manages to park itself across two parking spaces.
Tesla FSD V12.2.1 takes me into the parking lot, finds a few available spaces and parks safely. Mind you it took two parking spaces but that’s a huge improvement. I’m confident the Tesla AI team will correct this with ease. pic.twitter.com/giUk5PgQn8
— Miss Jillybean (@MissJilianne) February 23, 2024
Another Tesla account, AI DRIVR, posted a video in which his Tesla EV deftly navigates a number of parking lots in a manner he describes as “incredible.” Impressively, the Tesla FSD system seems to behave more predictably and confidently around other vehicles and pedestrians, even when the other road users were not behaving as predictably.
AI DRIVR also demonstrated his Tesla automatically executing a U-turn with FSD beta v12 active, commending the vehicle's ability to do so smoothly, safely, and without hesitation. The new FSD v12 also seems to be more adept at handling adverse road conditions, like speed bumps, puddles, and potholes, more gracefully.
At speed bumps, the vehicle is shown slowing down, where FSD beta v11 would usually have just blown through the speed bump. FSD v12 was also shown to handle a dip that consistently caused issues with FSD v11 far more comfortably.
There are still some blunders in FSD v12, though. Specifically, the Tesla exhibits strange behaviours when the car doesn't have any other traffic or road markings to reference. When on an open road, the Tesla fails to maintain a regular pace, sporadically gaining and losing around 5–10 mph. Similarly, the FSD software seems to be confounded when placed in a vacant, unmarked parking lot, resulting in chaotic steering behaviour, as if the car was looking for something to see before it can make a decision.
Ultimately, the reports indicate that FSD beta v12 — and specifically the recent v12.2.1 on display in these posts, seems to be a step forward for Tesla's self-driving systems, even if it isn't perfect just yet.
Read about the development of FSD and the Cybertruck in Walter Isaacson's biography of Elon Musk, or get in on the Cybertruck hype with a Hot Wheels Tesla Cybertruck model from Amazon.
Source(s)
