— The National Transportation Safety Board released information this week from its investigation of a fatal crash in May 2016 when a Tesla Model S with its Autopilot system engaged hit a semitrailer. The collision killed the Tesla's driver, 40-year-old Joshua D. Brown.
The agency cautions that the 538-page docket has preliminary information, is subject to change and has "only factual information collected by NTSB investigators." It doesn't provide findings or recommendations, and "no conclusions about how or why the crash occurred should be drawn," NTSB says.
The Tesla in question — a 2015 Model S 70D — struck a 53-foot refrigerated semitrailer pulled by a 2014 Freightliner Cascadia truck, NTSB said. The incident occurred at 4:36 p.m. EDT on Saturday, May 7, 2016, in the eastbound lanes of U.S. Highway 27A, west of Williston, Fla. That's a four-lane highway (two lanes in each direction) with a 75-foot-wide median and posted speed limit of 65 mph. The Model S struck the truck as the truck turned left from Highway 27A's oncoming (i.e., westbound) lane onto a local paved road.
Data downloaded from the Model S indicated a vehicle speed of 74 mph just before impact, with adaptive cruise control and lane-centering steering systems — which Tesla calls Traffic-Aware Cruise Control and Autosteer, respectively — both active. The systems are part of Tesla's Autopilot program.
NTSB has no indication that the Tesla or its driver hit the brakes in the moments before the "first fault report" — a reference to the Tesla's fault codes that indicated a collision. Another driver who witnessed the scene later told investigators that the Tesla never appeared to slow down in the moments that led up to the collision, according to NTSB. That's consistent with a separate investigation by the National Highway Traffic Administration, which noted that neither the driver nor the Model S' automatic emergency braking system took any action to decelerate before impact. NHTSA's crash reconstruction also found the tractor trailer "should have been visible to the Tesla driver for at least seven seconds prior to impact," the agency stated.
In a statement after the accident, Tesla said its Autopilot system, which uses a collection of radar, camera and ultrasonic sensors, failed to perceive the white trailer against a "brightly lit sky." Asked for comment on this week's NTSB report, a Tesla spokesperson did not immediately respond to .
NHTSA's investigation did not turn up any defects or design flaws in Tesla's automatic braking or Autopilot system; the agency notes that automatic emergency braking systems in the auto industry through the 2016 model year "are not designed to reliably perform in all crash modes, including crossing path collisions."
Prior to the accident, vehicle data indicate that the Model S had traveled for 41 minutes, according to NTSB. Its driver controlled the car for the first 2.5 minutes of that journey, plus two other occasions of about 20 to 30 seconds each. For the rest of the time — about 37.5 minutes — the Model S' TACC and Autosteer were active. NTSB also notes that for nearly that entire period — roughly 37 out of 37.5 minutes — Autosteer did not detect the driver's hands on the steering wheel.
Citing Tesla's owner's manual, NTSB notes that at highway speeds, Autosteer warns you to keep your hands on the wheel after one to five minutes, depending on a few other factors, if it doesn't detect any force applied to the wheel. After a series of three escalating warnings, it will disengage, slow the car to a stop and turn on the hazard lights. If the driver applies steering or changes the cruise-control speed, the timer resets.
The "longest period between alerts during which the system did not detect driver-applied torque on the steering wheel was about six minutes," NTSB investigators said. "The driver increased the cruise speed about two minutes before the crash, which did not require the driver to place hands on the steering wheel; however, this action reset the hands-off counter."
Indeed, NTSB says onboard data presented visual and audible warnings to the driver six times, plus visual warnings alone a seventh time. Autosteer never detected inaction long enough to disengage, the agency says.
NTSB says the Model S owner's manual has repeated warnings to pay attention and avoid sole reliance on the system to avoid collisions. It notes that Autosteer in particular is a hands-on feature and intended for use "with a fully attentive driver." Even NHTSA notes Autopilot is a Level 2 system, a reference to SAE International's widely accepted six levels of automated driving (Level Zero to Level 5). Level 2 cars can manage steering, acceleration and brakes but require "continuous attention by the operator to monitor the driving environment and take immediate control when necessary," NHTSA says.
At the time of the crash, the Model S had Tesla's latest software updates, according to NTSB. Tesla has since updated its software to reduce the maximum hands-off time at highway speeds to three minutes, NTSB says. It will also turn off Autosteer capabilities if the driver incurs three separate audible alerts within an hour.
After the impact, NTSB says the witness called 911, drove to where the Tesla ended up and waited for an hour or more while emergency personnel tended to the scene. The agency notes that the witness did not "hear or see any evidence of a video, or other audio/visual entertainment device playing." That runs counter to an earlier witness report that the Tesla driver was watching a movie at the time. Indeed, NTSB says its investigators recovered the driver's cellphone, laptop and other electronics but "did not uncover any evidence that those devices were in use at the time of the crash."
The forward collision warning system in the crashed Tesla could be set to varying levels of intensity; it works between 5 mph and 85 mph. Recorder data furnished by Tesla "does not contain values that would indicate that [forward collision warning] was disabled," NTSB said. "The data does not contain any information indicating that FCW was activated at any point during the crash trip."
The automatic emergency braking on the crashed car was running different firmware than the Model S that aced the Insurance Institute for Highway Safety's automatic braking tests, but Tesla confirmed no significant differences between the two, according to NTSB. It defaults on at the start of a trip, but the agency says the data it received "does not contain values that would indicate that AEB was disabled." The data also contains no information to indicate that AEB "was activated at any point during the crash trip."
The crash remains under investigation, with the Florida Highway Patrol and Tesla also involved. NTSB says it will issue recommendations and determinations of probable cause at a later date.