Tesla Inc. terminated an employee six days after he posted a YouTube video of his car running into a traffic pylon while using Full Self-Driving, or FSD, the carmaker’s controversial driver-assistance system.
John Bernal, who worked on the data-annotation team for Tesla’s Autopilot system, received a separation agreement from the company on Feb. 11, just under a week after he posted a video that now has more than 180,000 views. About 3 1/2 minutes in, Bernal’s Model 3 makes a right turn too sharply and runs into a green pylon separating a road and bike lane in downtown San Jose, California.
Here is the video. The collision occurs at the 3:30 mark.
Bernal, 26, said in a phone interview that while his manager refused to put the reason for his firing in writing, he was told it was in part due to improper use of FSD. Tesla said in January the beta software was running on almost 60,000 vehicles in the U.S.
The carmaker, which disbanded its public relations department in 2020, didn’t respond to a request for comment.
At 2:40 in the video, the car does a rolling right turn against a red light which appears to be in violation of California’s rules for turns:
Right turn against a red traffic signal light – Signal and stop for a red traffic signal light at the marked limit line. If there is no limit line, stop before entering the crosswalk. If there is no crosswalk, stop before entering the intersection. You may turn right if there is no sign to prohibit the turn. Yield to pedestrians, motorcyclists, bicyclists, or other vehicles moving on their green traffic signal light.
There is an exception for turning into a dedicated lane for right turns, but that does not appear to be the case here.
Here is a frame-by-frame step-through of the accident commented by Bernal, showing when he reacted to the impending collision and when the autopilot disconnected in response.
Watching this, I don’t think I would have taken that corner anywhere near as fast as “full self-driving” did.
That is consistent with many FSD videos I’ve seen. Drivers say that Tesla is going faster than or surprisingly nearly as fast as they would have driven themselves.
In contrast, I have not used the radar cruise control in my car for a long time because it is so timid. It leaves too large a gap to be useful in even very light highway traffic.
Not an FSD accident, but reports out of Florida have Tesla taking a more conventional products liability hit in this case:
Waiting for verdict to hit the database.
Based on the complaint, the accusations involved 1) flammable batteries, 2) doors that did not unlock as well as 3) Tesla removing a speed limiter (software setting?) that the owner had them originally install due to his irresponsible son driving the car.
Apparently has some interesting damages apportionment.
Thank you for sharing. My reaction is the bollard delimiters must be an edge-case or outlier situation for FSD.
This article indicates near obstacle detection is based on sensor fusion across optical and ultrasound sensors (source)
The sensor package designed into Tesla Model 3 includes: eight cameras which provide 360-degree visibility around the car within a radius of 250 meters; 12 ultrasonic sensors that complete this vision system. Together, they allow the detection of hard and soft objects at a distance and with almost twice the accuracy of the previous system. The package also incorporates a forward-facing radar system with improved processing capabilities. It provides additional data about the surrounding environment on a redundant wavelength that can see through heavy rain, fog, dust and even beyond previous cars.
On the camera front, there are four cameras facing the front that support the radar and have different characteristics. The main one, covers 250 meters but with a very narrow-angle of view, and there are others that cover shorter distances (150-, 80- and 60 meters) but with a wide-angle view of the environment around the car and are those used to read the road signs. The other four cameras face the sides and rear of the car and can see up to 100 meters away.
Sonar, on the other hand, uses ultrasound to detect obstacles within a radius of 8 meters around the car. It works at any speed and also controls the blind spot. The data collected by the sonar is also used by Autopilot to manage the automatic lane change during overtaking. Finally, GPS is used to detect the position of the car concerning the road.
I’ve watched the so-called “frame by frame” collision analysis and while realizing the perception of how fast a vehicle is going is different when you are in the car vs. when you watch video footage, it does not seem like the car is going that quickly around the corner.
Lots of opportunity to speculate as to the driver’s motivation…
But the apportionment of responsibility means Tesla will only be liable for $105,000, which is equivalent to 1% of the negligence as determined by the verdict, according to Curt Miner, an attorney representing the Riley family.
Presumably, the car and driver were registered for the FSD Beta program. Tesla can figure out the location, then identify the car, then identify why it failed or appeared to fail.
