Tesla released Full Self-Driving (FSD) Beta V12.2.1 earlier this week to a limited number of testers in the US. Now that the software has been tested for a few days, it is exhibiting a brand new, and impressive behaviour that moves Tesla one step closer to full autonomy.
FSD Beta V12 is a significant release because it is the first version of the software featuring what Elon Musk calls “end-to-end neural nets.” This approach is a shift from the previous method where the vehicle’s perception of its surroundings through cameras (a neural net) was processed and interpreted by over 300,000 lines of traditional code to make driving decisions like accelerating, decelerating, or turning (not a neural net). Now, both the input (the car’s visual perception) and the output (the car’s driving actions) are managed by neural networks.
In early tests last month with the first semi-public release of V12.1.1, this new approach resulted in more “human-like” behaviours, with the software being much smoother, and less jerky, in its movements, according to those who have used it. It also had a new behaviour in that the car would attempt to pull over to the side of the road and park once it reached the destination.
Now with the more public (but still limited) release of V12.2.1, the software is exhibiting even more human-like behaviours when it comes to finding a parking spot, as noticed by X user Arthur (@ArthurFromX).
The destination on this particular route was an alley in downtown Sacramento, with parking spots on both sides of the alley. The car reached the destination, but instead of stopping in the middle of the alley like it would have in V11, the car actually continued through and the navigation re-routed the car around the block to seemingly try again.
The car then entered the alley for a second time, and instead of driving half-way up the alley to reach the actual destination point on the map, it pulled into an open, perpendicular, parking spot well before that point.
As noted by Arthur, this is exactly how a human would have handled the same situation. First driving to the actual destination point, realizing there are no parking spots, and then circling around again to park in a spot close to, but not at the actual destination point.
The experience wasn’t perfect and there is still work to be done, as the car parked between two parking spots, a behaviour that Arthur said that was consistent with his other tests where the car parked itself at the end of a route. It is however a significant step forward from previous versions and shows the potential of Tesla’s new approach to reaching full autonomy.
You can watch the full video below.
FSD Beta 12: The One Where It Parks Itself
— Arthur (@ArthurFromX) February 21, 2024
After consistently having the car either park itself into a stall (if my destination was a parking lot) or pull to the curb out of the roadway (if the destination was on a street), I experienced this…
Set the destination to a random… pic.twitter.com/yCOFk9DB94
