Why did a Tesla veer left before a fatal crash?
David Wagman | February 12, 2020Data retrieved from a Tesla that was involved in a fatal crash in 2018 showed that the driver did not have his hands on the steering wheel in the seconds just before the crash.
Data also showed that the car's Autopilot feature was engaged continuously for the final 18 minutes and 55 seconds of the crash trip. Phone records also suggested the driver had been playing a video game on his iPhone for at least part of the fatal drive.
The information was made available in mid-February by the National Transportation Safety Board (NTSB), which is investigating the fatal accident. The NTSB investigation is ongoing and no conclusion was reached regarding the accident's cause.
Autopilot engaged
According to the NTSB, Tesla Carlog data showed that the ignition cycle that preceded the crash lasted about 28 minutes and 33 seconds. The vehicle's Autopilot was engaged for a total of 21 minutes and 53 seconds during the last ignition cycle. During the final Autopilot segment, the system did not detect driver-applied steering wheel torque for 34.4% of the time.
During this last Autopilot segment, the NTSB said the system issued two visual alerts for hands-off driving operation and one first level auditory alert. Roughly six seconds before the crash, Autosteer did not detect driver-applied steering wheel torque. NTSB said this lack of hands-on detection continued until the time of the crash. During this phase of travel, the Tesla entered the gore area and traveled toward a previously damaged crash cushion "with no evidence of braking or evasive steering action by the driver."
Mysterious turn
NTSB said that the driver's family related that the driver had experienced issues with the Autopilot steering system at the accident location on previous occasions. Many times, when the driver went past the crash location in the left lane, the Tesla would steer left toward the gore point area and he would have to manually take control to stay within the left lane. The family explained that this happened so often that he had told both his brother and his wife about the problem.
Image showing the fatal accident scene. Credit: NTSBNTSB said its investigators reviewed 15 days of historical Tesla Carlog data and focused on the time period when the driver was making his trip to work. NTSB said it identified another similar incident on Feb. 27, 2018. At about 9:31 a.m., the Tesla was being operated with Autopilot activated and the driver’s hands were detected on the steering wheel. Autosteer applied a -6º steering wheel angle toward the gore area, which was followed two seconds later by a +1.3º corrective steering wheel angle away from the gore.
NTSB offered no further details.
Accident timeline
On Friday, March 23, 2018, at about 9:27 a.m., the electric-powered SUV, driven by a 38-year-old, was traveling south on U.S. Highway 101 (US-101) in Mountain View. At the location, US-101 has six southbound traffic lanes, including a high occupancy vehicle (HOV) exit lane to State Route 85 (SR-85) southbound on the far left.
As the Tesla approached the US-101−SR-85 interchange, it was traveling in the lane second from the left, which was an HOV lane for continued travel on US-101. According to data downloaded from the vehicle, the driver was using two advanced driver assistance system (ADAS) features: Traffic Aware Cruise Control and Autosteer (a lane-keeping assist system), which Tesla refers to as Autopilot.
As the Tesla approached the paved area dividing the main travel lanes of US-101 from the SR-85 left exit ramp, it moved to the left and entered the dividing area, known as the gore area. The Tesla continued through the gore area and struck a nonoperational crash attenuator at about 71 mph. The crash attenuator was positioned on the end of a concrete median barrier.
The impact rotated the SUV counterclockwise and caused the front body structure to separate from the rear of the vehicle. The Tesla hit two other vehicles, a 2010 Mazda 3 and a 2017 Audi A4. During the accident the Tesla’s battery was breached and a post-crash fire erupted.
After the crash, bystanders found the driver in his seat with his lap/shoulder belt buckled. They removed him before the vehicle was engulfed in fire. The driver was taken to a local hospital, where he died from blunt force trauma injuries. The driver of the Mazda sustained minor injuries. The driver of the Audi was uninjured, NTSB said.
Just further proof that autonomous cars are a long way off. Not to say that driver assist technologies won't continue to evolve, because they will. But Level 5 is decades away. If ever.
In reply to #1
Was there ever an FMEA (Failure Modes Effects Analysis completed and tests done to confirm those imposed faults??
In reply to #12
Since I don't work for Tesla, I have no idea of what went into their FMEA. But I suspect the NTSB (National Transportation Safety Board) is looking at their FMEA or FMECA. ("C" for Criticality)
In reply to #14
An FMEA is required when an electronic assembly is part of a military system purchased by the Government. It requires that each component in the electronic assembly be analysed to determine what happens to the system. The analysis of the system performance is done for the failure of that component which includes the failure defined as the component changing to the out-of-specification
The end result is a high reliability system which may be redesigned to reduce component and or reduce manufacturing cost.
Yes, it is a very expensive procedure, but also important when it contributes to the success of a flight mission.
Consider this: An aircraft takes off from an aircraft carrier for a mission of destruction of accurately defined target. The guidance system is set for the T.O. position and for the return position of the rapidly moving aircraft carrier. The aircraft must complete the mission then return to the carrier that may have changed its position while the sun has set so the aircraft must locate it in darkness.
Hopefully, now you might have some idea as to why the FMEA which helps define the Critical in FMECA.
In reply to #15
But really, they are in a FRACAS mode now.
It's interesting that the Tesla reportedly tried to have an accident at the same location several times before it was successful. Somebody should check that out.
There was a reason I didn't drop the cash for the Full Self Drive feature. I have driven a lot of miles in my life and have seen so many completely unexpected bonehead moves by human drivers that it was obvious no computer would be able to deal with that any time soon. I think it helped that I have the background and education I have also.
A computer is given the same base line we are. The rules of the road. What the computer doesn't understand is humans will intentionally violate them. Something a computer can not do. A computer will not speed up, cut in front of you, then slam on the brakes. humans do it all the time. A computer is not going to expect that because it is illogical, irrational, irresponsible, and quite dangerous. Score one for humans! My car has NEVER sped up to block the lane when a car puts on its turn signal for a lane change. Never. The car is doing its thing and trusting you, the supposedly smarter, more capable, human, to keep it and you safe. You are the primary it is the secondary. Much the same responsibility split as pilot / copilot when the Co has the stick.
In reply to #3
I still think it's easier to drive the car yourself than to remain alert watching that the autopilot does it correctly. With nothing to do, the human brain succumbs to boredom. It's "highway hypnosis" on steroids.
In reply to #5
That was my initial assumption as well. However, experience is a great teacher. The trip from Madison Wi to Yellowstone is one long, dull, boring, drive across the vast nothingness of South Dakota with very little traffic.
As I have aged, my ability to drive long distances has grown shorter. I originally allowed two days out and two days back. After the first day I had learned several invaluable lessons that changed the way I ran the rest of the trip which allowed for the side trip down to the Tetons, a full day that didn't exist in the original plan. That is how different this experience really is. The car added a full day to my vacation.
Autopilot was instrumental in this. I discovered after the first day that I was not tired or looking to stop by 6pm, we had dinner, topped up the battery cancelled the reservation and drove another two hours. I wasn't even tired then and went swimming to unwind.
It really does take less effort to monitor. Once you have become familiar with the technology and how it works you can anticipate and predict its response fairly accurately. I was not uncomfortable letting the car handle the driving and I got to sight see. You see, most of the "monitoring" is done by feel. you feel the car moving, you feel the wheel moving and know if its feels right for the circumstance. take a quick glance at the road and then back to contemplating the life of those antelope over there.
In Yellowstone itself the car did the driving and I was a spectator. THANK GOD! for the very first time I got to enjoy the scenery instead of the traffic. And Yellowstone is one of those places where you really don't want to watch the road. Again, fatigue from driving was not an issue. The car is flawless at following and lane keeping in 95% of the situations it encounters, It even displayed (as a motorcycle) and responded to the Buffalo in the road.
SO yes, I am very comfortable with it. The car is now approaching 24,000 miles and at least a third of those are on autopilot.
I think it's proper for the NTSB to be investigating this because autonomous driving is still so new and the cause of regular left steer input at this specific location should be determined but IMO this driver is completely at fault for the accident.
I might think differently (probably not) if the steering error was a random event but this driver knew there was an issue with his car's autopilot at this location yet he still played with his phone instead of monitoring and controlling his car. Playing video games in the driver's seat of a moving car when you are responsible for your own safety and that of others is just stupid.
In reply to #4
This old advertising campaign sings true . . . .
In reply to #7
But did they really Fart Fig Newtons?
In reply to #8
Not that I know of, but I do recall some rather ribald parodies of the Farfegnügen ad campaign.
I'm not worried about self driving vehicles being perfect. I just want them to be better than humans. At least when we have a lot of them on the road they will be more predictable.
Sorry. But. This should have immediate legal changes in our laws, I don’t feel comfortable out on the roads anymore with all these stupid ignorant people behind the wheels. Doing dumb things vs actual driving of their vehicles I would put all the CEO’s
of any driverless vehicle in jail immediately no trial no nothing. They are contributing
‘to these accidents and this. B/S needs to stop. Now. I don’t need to be on the roads
with these pos. Toy cars. With assholes. Playing games
An autonomous vehicle is like any programmed automation. It could work and do what’s its suppose to do 99.9999% of the time,...then for what ever reason, some how a bit gets landed where it’s not suppose to be and goes haywire for one moment.
thing about automation,... you reset it, and you spend some time to try to track it down... but it comes down to working 99.9999% accuracy is acceptable in your process, where autonomous driven vehicles is not. There may need some type of redundancies in the actually program itself, and that may not correct it.