Hold the Line: Video From Location of Deadly Tesla Crash Shows Weird Autopilot Behavior

Steph Willems
by Steph Willems

We all play amateur detective whenever a Tesla crashes or does something wonky while operating on Autopilot (or in its absence), and last week was no exception.

The death of Wei Huang following his Model X’s collision with a lane divider on California’s US-101 freeway in Mountain View prompted Tesla to issue two statements concerning the incident. In the second, the automaker admitted, after retrieving digital logs from the vehicle, that the vehicle was in Autopilot mode and that the driver did not touch the wheel in the six seconds leading up to the March 23rd impact.

Retracing the last few hundred yards of Huang’s journey on Google Streetview led this author to make a very obvious observation: that the paint marking the left-side boundary of the lane Huang was presumably driving in was faded and half missing as it approached the barrier. As it turns out, the condition of that not-so-solid white line caused another Tesla’s Autopilot to act strangely, but this time the driver corrected in time. He also has a video to show what happened.

Shantanu Joshi, who filmed the video and posted it to Reddit’s Tesla Motors forum (later to be picked up by Electrek), claims he travels down the same stretch of US-101 on his way to work every day.

After hearing about the crash, he decided to test his own car’s Autopilot on the section where the exit lane breaks off from the southbound US-101 and carries over to Highway 85. Joshi kept his car in the left lane of US-101, which is accompanied (on the left) by the 85 exit ramp leading up to the barrier.

“The results are kinda freaky,” Joshi said.

As seen in the video, the marker between these two lanes splits approaching the barrier, separating the exit lane from the left lane of US-101. That split grows wider as it approaches the barrier, but there’s a stark contrast between the paint on either side. On the US-101 side, the paint is, in many parts, barely there — especially at the beginning of the split. Meanwhile, the paint on the exit side is uniform.

As the two lanes diverge, Joshi’s Tesla follows the left-most lane marker, pulling the vehicle into what is not actually a lane and sending it on a collision course with the barrier. It seems the car’s sensors latched onto the most prominently painted line, and the lane-holding electronics ensured the vehicle charted this new course. On a fast-lane freeway journey, this may be the only spot capable of tricking Autopilot. However, it’s clearly a situation that can turn deadly if the driver’s attention is distracted for just a few seconds.

Note that, after taking his hands off the wheel to record the video and allow the car to approach the split under Autopilot control, we can see and hear the Tesla emit a visual and audio warning a split second before Joshi retakes the wheel.

Huang’s family told local media that the victim had complained about his Model X veering off the road on the same stretch of highway, but didn’t offer further details. For its part, Tesla claims the victim never approached them with an Autopilot-related complaint, just a “navigation” one. This author will admit to being suspicious as to the nature of this vague navigation concern.

Both the National Highway Traffic Safety Administration and National Transportation Safety Board have opened investigations into the crash, and Tesla says it’s cooperating fully. Without a preliminary report to go on, this video, coupled with the crash details already released by Tesla and local media, paints a pretty scary picture of why Tesla drivers can’t let their guard down.

[Image: Shantanu Joshi/ YouTube]

Steph Willems
Steph Willems

More by Steph Willems

Comments
Join the conversation
2 of 59 comments
  • Islander800 Islander800 on Apr 04, 2018

    I think a good wake-up call for Elon Musk would be a charge for criminal negligence leading to death. Why him? Because, as he is so eager to point out at every opportunity, these initiatives are all mandated from the very top of the company, and that would be Elon. This stuff is obviously NOT ready for prime time. To cynically and recklessly use his vehicles, his customers and the general public as guinea pigs for his "beta" testing, where "software errors" result in death, seems to me the definition of criminal negligence. And by the way, the same goes for regulatory officials that gave the green light to this idiocy on the public highways.

  • Master Baiter Master Baiter on Apr 05, 2018

    An "autopilot" system that can kill you inside of six seconds is less than worthless. . .

  • Kmars2009 I rented one last fall while visiting Ohio. Not a bad car...but not a great car either. I think it needs a new version. But CUVs are King... unfortunately!
  • Ajla Remember when Cadillac introduced an entirely new V8 and proceeded to install it in only 800 cars before cancelling everything?
  • Bouzouki Cadillac (aka GM!!) made so many mistakes over the past 40 years, right up to today, one could make a MBA course of it. Others have alluded to them, there is not enough room for me to recite them in a flowing, cohesive manner.Cadillac today is literally a tarted-up Chevrolet. They are nice cars, and the "aura" of the Cadillac name still works on several (mostly female) consumers who are not car enthusiasts.The CT4 and CT5 offer superlative ride and handling, and even performance--but, it is wrapped in sheet metal that (at least I think) looks awful, with (still) sub-par interiors. They are niche cars. They are the last gasp of the Alpha platform--which I have been told by people close to it, was meant to be a Pontiac "BMW 3-series". The bankruptcy killed Pontiac, but the Alpha had been mostly engineered, so it was "Cadillac-ized" with the new "edgy" CTS styling.Most Cadillacs sold are crossovers. The most profitable "Cadillac" is the Escalade (note that GM never jack up the name on THAT!).The question posed here is rather irrelevant. NO ONE has "a blank check", because GM (any company or corporation) does not have bottomless resources.Better styling, and superlative "performance" (by that, I mean being among the best in noise, harshness, handling, performance, reliablity, quality) would cost a lot of money.Post-bankruptcy GM actually tried. No one here mentioned GM's effort to do just that: the "Omega" platform, aka CT6.The (horribly misnamed) CT6 was actually a credible Mercedes/Lexus competitor. I'm sure it cost GM a fortune to develop (the platform was unique, not shared with any other car. The top-of-the-line ORIGINAL Blackwing V8 was also unique, expensive, and ultimately...very few were sold. All of this is a LOT of money).I used to know the sales numbers, and my sense was the CT6 sold about HALF the units GM projected. More importantly, it sold about half to two thirds the volume of the S-Class (which cost a lot more in 201x)Many of your fixed cost are predicated on volume. One way to improve your business case (if the right people want to get the Green Light) is to inflate your projected volumes. This lowers the unit cost for seats, mufflers, control arms, etc, and makes the vehicle more profitable--on paper.Suppliers tool up to make the number of parts the carmaker projects. However, if the volume is less than expected, the automaker has to make up the difference.So, unfortunately, not only was the CT6 an expensive car to build, but Cadillac's weak "brand equity" limited how much GM could charge (and these were still pricey cars in 2016-18, a "base" car was ).Other than the name, the "Omega" could have marked the starting point for Cadillac to once again be the standard of the world. Other than the awful name (Fleetwood, Elegante, Paramount, even ParAMOUR would be better), and offering the basest car with a FOUR cylinder turbo on the base car (incredibly moronic!), it was very good car and a CREDIBLE Mercedes S-Class/Lexus LS400 alternative. While I cannot know if the novel aluminum body was worth the cost (very expensive and complex to build), the bragging rights were legit--a LARGE car that was lighter, but had good body rigidity. No surprise, the interior was not the best, but the gap with the big boys was as close as GM has done in the luxury sphere.Mary Barra decided that profits today and tomorrow were more important than gambling on profits in 2025 and later. Having sunk a TON of money, and even done a mid-cycle enhancement, complete with the new Blackwing engine (which copied BMW with the twin turbos nestled in the "V"!), in fall 2018 GM announced it was discontinuing the car, and closing the assembly plant it was built in. (And so you know, building different platforms on the same line is very challenging and considerably less efficient in terms of capital and labor costs than the same platform, or better yet, the same model).So now, GM is anticipating that, as the car market "goes electric" (if you can call it that--more like the Federal Government and EU and even China PUSHING electric cars), they can make electric Cadillacs that are "prestige". The Cadillac Celestique is the opening salvo--$340,000. We will see how it works out.
  • Lynn Joiner Lynn JoinerJust put 2,000 miles on a Chevy Malibu rental from Budget, touring around AZ, UT, CO for a month. Ran fine, no problems at all, little 1.7L 4-cylinder just sipped fuel, and the trunk held our large suitcases easily. Yeah, I hated looking up at all the huge FWD trucks blowing by, but the Malibu easily kept up on the 80 mph Interstate in Utah. I expect a new one would be about a third the cost of the big guys. It won't tow your horse trailer, but it'll get you to the store. Why kill it?
  • Lynn Joiner Just put 2,000 miles on a Chevy Malibu rental from Budget, touring around AZ, UT, CO for a month. Ran fine, no problems at all, little 1.7L 4-cylinder just sipped fuel, and the trunk held our large suitcases easily. Yeah, I hated looking up at all the huge FWD trucks blowing by, but the Malibu easily kept up on the 80 mph Interstate in Utah. I expect a new one would be about a third the cost of the big guys. It won't tow your horse trailer, but it'll get you to the store. Why kill it?
Next