Tempe Police Vehicular Crimes Unit is actively investigating
the details of this incident that occurred on March 18th. We will provide updated information regarding the investigation once it is available. pic.twitter.com/2dVP72TziQ

— Tempe Police (@TempePolice) March 21, 2018

The car’s software detected Herzberg more than five seconds before the crash but did not identify her as a pedestrian with a bike crossing the street away from a crosswalk. However the NTSB report went into depth saying that while the backup driver could have avoided the crash if they had been paying attention, it was “the last link in a long chain of actions and decisions made by an organization that unfortunately did not make safety the top priority.” That included Uber deactivating the Volvo SUV’s built-in automatic emergency braking system, which put all pressure on the backup driver to intervene in a situation where the self-driving rig failed.

The report says one of the contributing factors to the crash was that Tesla’s Autopilot software wasn’t designed to handle the type of road conditions it encountered the night of the collision. The five-mile stretch of highway where the Delray Beach crash occurred has approximately 34 intersections and private driveways where cars can join traffic. The agency says Tesla’s analysis found Autopilot didn’t treat the truck as a “threat” as it moved into the path of the Model 3 because it was designed to navigate highways with limited access where crossing traffic isn’t an issue. Had the Model 3 driver not been able to turn on Autopilot in that specific instance or had the system made sure they paid attention to the road, the accident may have been avoided.

“Contributing to the crash was the operational design of Tesla’s partial automation system, which permitted disengagement by the driver, and the company’s failure to limit the use of the system to the conditions for which it was designed,” the report says.

The agency also had some choice words for the National Highway Traffic Safety Administration (NHTSA). It once again criticized the regulator for not developing a way to ensure automakers include the right safety features in their Level 2 autonomous vehicles. The NHTSA can issue recalls for unsafe vehicle tech, something the NTSB can’t do.

The NTSB was similarly critical of Tesla last month when it announced the initial findings of the crash that killed Apple developer Walter Huang. At the time, it said it had contacted automakers as far back as 2017 about making autonomous driving systems harder to misuse. “Sadly, one manufacturer has ignored us, and that manufacturer is Tesla,” said NTSB chair Robert Sumwalt. “We ask that recommendation recipients respond to us within 90 days. That’s all we ask. But it’s been 881 days since these recommendations were sent to Tesla and we’ve heard nothing. We’re still waiting.”

The incident — which took place in Columbus, Ohio — saw a passenger fall from their seat after the shuttle made an emergency stop. The vehicle has been traveling at seven miles per hour (11 kilometers per hour). EasyMile, which has pods operating in various locations, says that the vehicles are still permitted on US roads, but won’t be able to carry passengers while the NHTSA conducts its review.

It’s not the first time EasyMile has faced safety criticism. As Reuters reports, an elderly passenger was injured in a similar braking incident in Utah last year. Despite EasyMile’s comments that its vehicles are simply performing as they are programmed to, it’s clear that the NHTSA is not prepared to take any chances over safety issues, especially since it so recently came under fire from the National Transportation Safety Board (NTSB) for its role in 2018’s fatal Tesla crash.

“Government regulators have provided scant oversight” over the semi-autonomous driving systems that are quickly becoming standard features on modern automobiles, NTSB chair Robert Sumwalt declared. While the NTSB does not have the authority to enforce safety measures, the National Highway Traffic Safety Administration can issue recalls for unsafe vehicle tech.

The NTSB also determined via cellphone records and device data that Huang’s phone was running a mobile game at the time of the crash. While the NTSB could not determine whether Huang had the phone in his hands when the crash occurred, the Tesla’s data logs indicate that his hands were not on the steering wheel in the six seconds preceding the accident. This reportedly wasn’t the first time that Huang’s vehicle had swerved towards a median divider.

After the accident Huang’s brother told reporters that the vehicle had attempted to drive into that section of wall the previous March and that he’d complained to the dealership about it. Tesla maintains that Huang himself is responsible for the crash that took his life.

A Good Samaritan at the scene of the Tesla Model X car crash described the car to be “actively emitting full flames from the battery bank.” https://t.co/n78v5ekcgV pic.twitter.com/EVGqKJnhcR

— NBC Bay Area (@nbcbayarea) March 24, 2018

Sumwalt also pointed out Tuesday that the NTSB has, as far back as 2017, been contacting automakers regarding the safety implications of these autonomous driving systems. According to Sumwalt, VW, Nissan and BMW all promptly responded to the NTSB’s recommendations about making these systems harder to misuse.

“Sadly, one manufacturer has ignored us, and that manufacturer is Tesla,” Sumwalt said. “We ask that recommendation recipients respond to us within 90 days. That’s all we ask. But it’s been 881 days since these recommendations were sent to Tesla and we’ve heard nothing. We’re still waiting.”

We’ve reached out to Tesla for comment and will update this post when the company responds.

After the crash, Huang’s brother said Huang had talked about the car swerving toward the highway barrier. Huang supposedly complained to the Tesla dealership but said that they could not replicate the issue. Tesla previously confirmed that Autopilot was engaged at the time of the crash. The company has insisted that Huang was at fault, saying that the driver received several “hands-on warnings” and that the only way the crash could have happened was if Huang wasn’t paying attention.

Huang’s family has sued Tesla, alleging that the Model X was “defective” in its design, and a preliminary investigation report by NTSB found that the vehicle sped up just before hitting the road barrier. The report also said the driver’s hands were not detected on the steering wheel in the six seconds before the crash.

Engadget has reached out to Tesla for comment.

According to the NTSB’s findings, the Uber vehicle detected Herzberg 5.6 seconds before the crash, but did not identify them as a pedestrian. “Had the operator been attentive,” it added, they would “likely have had enough time to detect and react to” the events, potentially avoiding disaster. Compounding the issue was that Uber’s self-driving software disabled the Volvo’s own automatic collision-avoidance systems. This placed more onus on the safety driver, whose phone was placed on the bottom of the center console, ensuring her eyeline was nowhere near the windshield.

The lack of a second test driver in the car was also identified as an issue, as was Uber’s failure to check if drivers were following operating procedure. “Inadequate oversight of vehicle operators” is seen as a contributing factor here, and something that led to the driver’s “extended distraction.”

Herzberg was also described as having used drugs that “diminished perception and judgment,” which caused their “unsafe behavior.” The report also says that crossing the street not at a crosswalk “violated Arizona statutes.”

One of the reasons that Uber was testing in Arizona at all was because California had objected to the use of test vehicles on public roads without a permit. The regulation-fearing rideshare outfit then relocated to the laxer Arizona, encouraged by republican governor Doug Ducey. The NTSB report says that Arizona’s failure to develop rules and regulations for self-driving vehicles both before and after the incident is an issue. It goes on to say that this failure “demonstrate the state’s shortcomings in improving the safety of ADS testing and safeguarding the public.”

In summary, the board recommends the establishment of a rigorous process to identify if proper safety procedures are being followed. Arizona particularly needs to address its failures to create a system for licensing such vehicles, while Uber itself needs to fix its broken safety culture.

The driver reportedly had Autopilot engaged and was following closely behind a large SUV or truck. The lead vehicle changed lanes to move around a fire truck that was parked in the lane ahead. The Tesla driver claimed he was drinking coffee and eating a bagel and did not see the firetruck. When the lead vehicle changed lanes, the Model S accelerated. About .49 seconds before the crash, the vehicle detected a stationary object in the road and displayed a warning, but it was too late.

The vehicle’s Autopilot didn’t detect driver-applied steering wheel torque for the last three minutes and 41 seconds before the crash. And given the driver’s admitted distractions, NTSB says the driver was likely over-reliant on the vehicle’s driver assistance system.

Driver errors, Advanced Driver Assistance system Design, led to Jan. 22, 2018, Culver City, CA, highway crash, according to NTSB Highway Accident Brief 19/07 issued Wednesday; https://t.co/hozLB1zA7F pic.twitter.com/iyQNc2HdhT

— NTSB_Newsroom (@NTSB_Newsroom) September 4, 2019

According to Reuters, the Center for Auto Safety, a consumer watchdog group, said the NTSB report should prompt the National Highway Traffic Safety Administration (NHTSA) to “do its job and recall these vehicles … A vehicle that enables a driver to not pay attention, or fall asleep, while accelerating into a parked fire truck is defective and dangerous.” Engadget has reached out to Tesla for comment.

Tesla’s Autopilot was engaged in at least three fatal US crashes, two of which are still under investigation by the NTSB and NHTSA. Oddly enough, the January 2018 crash wasn’t the only time an admittedly distracted driver crashed into the back of a fire truck while using Autopilot. The system does issue “hands on warnings” and Tesla advises drivers to keep their hands on the wheel. But as others have pointed out, calling the driver assist features “Autopilot” may be a bit misleading.

Both Google and Apple agreed to update their navigation apps with railroad information, but haven’t given a reason for the delay. The search giant initially partnered with the Federal Railroad Administration back in 2015 to integrate its railroad data with Google Maps, but didn’t commit to a timeline. A Google spokesperson told Politico that the company remains “aware” of the recommendation and will continue to work on new safety features.

The FRA has its own app for railroad crossings, but reviews have largely been negative. Users complain about that the app is slow to load, out-of-date and glitchy.

Since the NTSB’s request, Google has unveiled new features to help navigate drivers away from natural disasters like hurricanes and floods, warn them of speed cameras and AR navigation. But railroad crossings don’t appear to be in Google Map’s plans for the immediate future. Apple Maps revamped its entire Maps app earlier this year — but it still lacks railroad information.

Other apps are doing slightly better. The Waze app will occasionally register a stuck train as traffic and route drivers in a different direction. The app partnered with the New York Metropolitan Transportation Authority to introduce railroad crossing alerts around Long Island. OpenStreetMaps is also building its own railroad mapping system, OpenRailwayMap. Given that 2,214 vehicle-train collisions happened in 2018 alone according to FRA data, it’s clear that more action is needed.

Apparently, the truck driver thought the shuttle would stop at a “reasonable” distance from the truck. Although the shuttle did start slowing down when it was 98.4 feet away, it’s not programmed to stop until it’s only 9.8 feet away from obstacles. The attendant hit the emergency stop button when the vehicle was 10.2 feet away from the truck, but it clearly wasn’t enough to prevent the incident.

In an interview with the investigators, the attendant said they considered switching to manual mode to move the shuttle out of the way, but they didn’t have easy access to its handheld controller. See, Las Vegas’ autonomous shuttles operate on a predetermined route, but attendants could use a controller to activate the horn and switch to manual operation. When the accident happened, the controller was stored in an enclosed space at one end of the passenger compartment. After the incident, the operator started requiring attendants to take the controller out of the storage space at the start of the trip and to keep it accessible at all times.

The NTSB clarified that it doesn’t normally investigate minor collisions, but the involvement of an automated vehicle warranted a closer look. From the sound of things, the agency will continue keeping a close eye on automated vehicle testing, as well. It wrote in the incident report:

“Pilot testing of highly automated vehicles on public roads is under way at various locations in the United States. The NTSB will monitor the development of those vehicles to better understand their potential safety impacts and any unintended consequences.”