This week, Tesla and investigators made public that the May crash of a Tesla Model S in Florida resulted in the first-ever fatality for an operator of a vehicle driving autonomously. There is early evidence that the driver may have been watching a movie prior to the crash, flouting Tesla’s requirement that Autopilot users “maintain control and responsibility” for their vehicle at all times.
However, the crash, when compared to another recent non-fatal Autopilot incident, suggests that a very specific weakness of Tesla’s Autopilot may also have been at fault. The fatal May crash occurred when a large tractor-trailer turned in front of a Model S travelling on a divided highway. According to Tesla, “the high ride height of the trailer combined with its positioning across the road” contributed to Autopilot not ‘seeing’ it or applying the brakes. The Model S passed under the trailer, which impacted its windshield.
Get Data Sheet, Fortune’s technology newsletter.
That scenario is eerily similar to a crash in Utah in which a Model S’s ‘Summon’ feature led it to pull forward into the overhanging tail of a tractor-trailer, which impacted its windshield. This was a low-speed collision, and the Utah car was unoccupied, but the two events are otherwise nearly identical.
Jalopnik’s Jason Torchinsky argues that this points to a possible ‘blind spot’ above the car’s hood. The Model S’s primary sensor for detecting forward obstacles is a radar unit located in the car’s nose, while a camera above the windshield handles primarily information about lanes and speed. This means the Model S is much better equipped to ‘see’ obstacles on the plane of the car’s nose.
For more on Tesla, watch our video:
Tesla’s own literature seems to confirm this, as it warns that the car “may not detect” obstacles “hanging from the ceiling.” Truck trailers, at least at the end and middle, would seem to present a similar problem for the Model S sensor array.
Trucks make up as much as 1/4th of the traffic on congested highways, meaning that the Model S may not be fully aware of a major element of its surroundings in a scenario where Autopilot is most appealing. Torchinsky suggests that Tesla may need to add a second, high-mounted radar unit to address the apparent design flaw. With Tesla’s upcoming Model 3 strongly rumored to include autonomous features, and its design nearly finalized, a rethink on that front may be due as well.