Tesla Model S In Fatal Autopilot Crash Was Going 74 MPH In a 65 Zone, NTSB Says (latimes.com)

An anonymous reader quotes a report from Los Angeles Times: The Tesla car involved in a fatal crash in Florida this spring was in Autopilot mode and going about 10 miles faster than the speed limit, according to safety regulators, who also released a picture of the mangled vehicle. Earlier reports had stated the Tesla Model S struck a big rig while traveling on a divided highway in central Florida, and speculated that the Tesla Autopilot system had failed to intervene in time to prevent the collision. The National Transportation Safety Board released a preliminary report Tuesday that confirms some details of the May 7 collision, along with a photo that shows the car with its windshield flattened and most of its roof sheared off. The federal agency also included a photo of the big rig, circling an area on the right side of the tractor-trailer that showed the light damage the truck received from the collision. The 2015 Model S was moving at 74 mph, above the posted 65 mph speed limit, when it struck a 53-foot trailer being pulled by a Freightliner Cascadia truck. Tesla's semi-autonomous Autopilot driving feature was engaged, the report says.

Comments

Re:I thought Autopilot was not on...

[Areyoukiddingme]

I remember reading something from Tesla saying they found autopilot was not on, and had it been it would have stopped the car.

Different incident. There have been three in recent weeks. This is the fatality, where autopilot was on, didn't detect the truck, and the jackass was watching a Harry Potter DVD in the driver's seat. Hopefully he didn't have children, for their sake and for ours, so no one has lost their father and we get a Darwin Award nominee.

What about the truck?

[Stu Fuller]

So, did the truck turn in front of oncoming traffic? If so, why is this the Tesla's fault?

Re:What about the truck?

[beanpoppa]

Autopilot could see through the under-carriage of the semi. It's programmed to ignore obstacles it considers over the top of car so that it doesn't stop at every overpass and road sign. Also, the trailer was gray in color, which matched the color of the sky at the time.

Re:74 at time of crash

[is as us Infinite]

It's a Cruise-Control-like system so I would assume it goes at the speed that the driver sets it at and then seeks to keep that speed. 74mph is 1mph under 10 over, which everybody knows is the 'real' speed limit and I bet is a common setting for people using Cruise-Control.

Re:74 at time of crash

[Anonymous Coward]

I think obstacles that are empty below 3 ft confuse the car.

http://bgr.com/2016/05/11/tesla-model-s-summon-crash/

I have also heard of the car running into 1/2 open garage doors.

Re:74 at time of crash

[Anonymous Coward]

Apparently brakes were not applied. They believe it was a combination of the trailer being a solid light gray color that tended to visually blend in with the sky, coupled with the radar being designed to ignore large flat signs that cross above the road. So the trailer managed to be filtered out as an hazard and was ignored by the software.

If the trailer had adhered to european safety regulation it would have at least side rails under to prevent cars being stuck underneath it. Not only would it have saved the car's driver it would also prevent the trailer from being detected as sign.

Re:74 at time of crash

[Jeremi]

So the road under the truck looked clear (sort of)

If only the autopilot system had been calibrated to take into account the exact height of the Tesla's roof. If that had been done, then there would have been no accident in this case (the Tesla would have stopped until the truck was out of the way), but when encountering a somewhat higher truck, the Tesla would pass cleanly underneath it, with the driver probably never even noticing what had happened. And that would have been rather awesome.

Re:74 at time of crash

[Rei]

I'm not so sure that a simple software fix can fix it. Some key notes:

About 4:40 p.m. eastern daylight time on Saturday, May 7, 2016, a 2015 Tesla Model S, traveling eastbound on US Highway 27A (US-27A), west of Williston, Florida, struck and passed beneath a 2014 Freightliner Cascadia truck-tractor in combination with a 53-foot semitrailer. At the time of the collision, the combination vehicle was making a left turn from westbound US-27A across the two eastbound travel lanes onto NE 140th Court, a local paved road. As a result of the initial impact, the battery disengaged from the electric motors powering the car. After exiting from underneath the semitrailer, the car coasted at a shallow angle off the right side of the roadway, traveled approximately 297 feet, and then collided with a utility pole. The car broke the pole and traveled an additional 50 feet, during which it rotated counterclockwise and came to rest perpendicular to the highway in the front yard of a private residence. The 40-year-old male driver and sole occupant of the Tesla died as a result of the crash.

US-27A is a four-lane highway with a posted speed limit of 65 mph. A 75-foot-wide median separates the two eastbound lanes from the two westbound lanes. Additionally, at the uncontrolled intersection with NE 140th Court, both eastbound and westbound lanes incorporate left turn lanes, allowing for a median opening of about 132 feet. At the time of the crash, it was daylight with clear and dry weather conditions.

Eastbound. Afternoon. May. Aka, the sun was right behind him. Clear and bright outside. This is a perfect recipe for light-colored objects ahead to be overexposed, against other overexposed objects, potentially including the road and the sky. If you have a big block of RGB(255,255,255), how do you determine the boundaries? The best you can do is recognize that it's a threat and disable autopilot, while warning the driver.

A more appropriate solution, if this was indeed the case, would be a hardware fix: read the *raw* data from the camera. A potential alternative, if the frame exposure time can be adjusted, would be to read out alternating short and long exposure frames and combine them.