Regardless of our individual feelings regarding Tesla’s controversial Full Self-Driving (FSD) suite of advanced driver assistance features, the company envisions FSD as its future.
That future is expected to materialize in the form of robotaxis and EVs that can drive like pro-human drivers. At the shareholder vote meeting today, CEO Elon Musk even said that robotaxis could rocket the company to a market valuation between five and 12 trillion dollars. That seems optimistic.
Get Fully Charged
Tesla's bet on FSD.
Tesla has pivoted its focus on AI and robotaxis. CEO Elon Musk made that goal all the more clear in the latest shareholder vote meeting that approved his colossal pay package. In reality, though, FSD beta testers are encountering experiences that are far from perfect.
Back here on the Earth and in the real world, FSD’s latest 12.4.1 version tries to improve the existing system, but misses the mark by a big margin, according to early reports from a handful of beta testers who have access to it.
Last month, Musk said that 12.4.1 was released internally (For Tesla employees) and limited external beta testers. He claimed that miles per intervention—the distance covered before the supervising human driver has to correct the car’s self-driving error—had improved by five to 10 times.
Some beta testers have said that’s nowhere close to reality. One beta tester going by the name of AI DRIVR on YouTube implied that FSD 12.4.1 has decision-making issues with lane discipline. It needlessly switches lanes, with or without traffic, and sometimes even enters merging lanes. This unnecessary back and forth between lanes is frustrating, according to him.
However, he added that he’s the only beta tester reporting this issue, so these errors could be unique to his EV.
At intersections, he faced another major glitch. When the traffic light was green, the EV zoned out and crawled into the intersection at a snail's pace instead of making a turn smoothly, despite the road being clear. In other instances, it was outright disobedient. It ignored the navigation and headed in opposite directions. Instead of turning right, it turned left.
Beyond these bugs, FSD 12.4.1 executed many things like a good human driver. It made full U-turns and even accelerated slightly before the turn was complete, which looked natural.
Before making left turns, it waited for gaps in the oncoming traffic behind the crosswalk line and not within the intersection, instead of moving into the intersection and waiting for the traffic to clear up. It’s something human drivers often do and is unsafe.
Pedestrian safety also seems to have improved. Imagine you’re in a parking lot and a person is walking to the back of their car, in the same direction as your path. FSD 12.4.1 slows down the EV considerably, in case they proceed onto your path.
The car can also detect what the driver misses. It noticed a discreet vehicle parked on the side of the road with its hazard lights on and slowed down for a man on an electric scooter approaching an intersection from far away. It moved only when the man waved. It also slows down when it sees emergency vehicles.
The screen rendering of the environment failed to display trash lying on the ground, but the car did notice it and drove around it.
Another well-known FSD 12.4.1 beta tester, going by the name of Whole Mars Catalog on YouTube and X, had similar findings.
He was parked on the side of the road when he activated FSD. The car paused a few seconds before moving because it noticed a vehicle parked directly behind him that was about to come out of the parking and onto the road. Finally, it let that vehicle pull out and drive away before it did the same. Credit where it is due, some of the stuff in the video above is impressive.
FSD 12.4.1 seems to have improved from previous versions in some areas. But not by five or 10 times like the CEO claims, maybe by a few percentage points.
All said it’s crucial to note that Autopilot and FSD are both SAE Level 2 driver assistance systems. No matter how impressive you find them, they’re not driverless systems and require constant driver attention and readiness to take control.
