Your browser is not supported. Please update it.

29 March 2018

Tide turns against Uber – suppliers baffled, Intel-Mobileye digs in

As time passes, the circumstantial evidence and public comments are painting a bad picture for Uber, in the wake of its fatal pedestrian collision. Rivals are turning the screw, with both Google’s Waymo and Intel’s Mobileye saying that their platforms would not have allowed this to happen. Uber has been very quiet, but it looks like a case could be building against the company.

The fallout from this could set precedents for liability in this industry, before it has even launched a commercial product. Equipment vendors have already begun defending their own technology, and rivals are claiming that their approaches would have avoided the problem in the first place. As long as the industry can make this look like a problem with Uber alone, then it could collectively avoid harmful regulation. The National Transportation Safety Board (NTSB) is investigating, and Arizona has now suspended Uber’s test license.

So while somewhat cynical, one can see the sense in this approach. They want to make sure that Uber is stung, and not them. However, there might be a useful precedent that they could use – the ongoing saga of Takata and its faulty airbags.

The supplier was recently bankrupted after trying to deal with lawsuits and recalls, after it was found that its designs sometimes fired metal into the faces of occupants of vehicles that deployed airbags. While these recalls have been expensive for the automakers, the likes of Honda and GM have avoided being the subjects of these lawsuits, despite using Takata parts. So it seems that there might already be a useful legal avenue to use when it comes to suing suppliers of self-driving products that might be responsible for a death.

The Tempe police have also shifted their stance in the past week. Initially saying that it would have been nearly impossible not to have hit Herzberg, a later statement said “Tempe Police Department would like to reaffirm that fault has not been determined in this case.”

In addition, Herzberg’s daughter has hired a legal firm to pursue personal injury litigation. That firm, Bellah Perez PLLC, said “by encouraging businesses like Uber to set up shop in Arizona, the state has hoped to be at the forefront of emerging technology and the sharing economy. But the potentially drastic shift in accident liability associated with self-driving technology is causing many professionals to question the legal implications of the industry.”

According to the on-board camera, it would have been very hard for a human to see Elaine Herzberg in enough time to avoid hitting her directly. However, there may have been enough time to brake sufficiently to lower the chance of death, or swerve out of the way. The on-board camera is not clear enough to make an accurate judgment in this case.

The interior camera shows that the test driver, Raffaela Vasquez, frequently looking at something in her lap. It is not clear if this is some piece of equipment used to monitor the car, or perhaps a personal mobile phone, but it is clear that she is not looking at the road ahead. The regulations on what such test drivers need to do when operating these vehicles is unclear and apparently dependent on the state’s laws. A two-person crew would solve that issue.

Vasquez might actually go some way to proving why so many players in the self-driving industry have said that cars need to be fully autonomous or not at all. They fear that any middle ground, where a machine might want to hand off control to a human if it meets a situation it doesn’t understand, will lead to crashes because the human drivers won’t be paying attention – they will distract themselves, and not pay attention to their surroundings.

A report from the New York Times claims that Uber’s cars were struggling to meet their target of one driver intervention per 13-miles, back in March. Apparently, the vehicles struggled near construction sites and tall vehicles, where they would meet unfamiliar road markings and lose sight of landmarks, respectively. The potential bombshell here is the revelation that Uber had shifted from using pairs of test drivers to just single operators, some time in November, potentially to increase testing runs in the lead up to a visit from Uber leadership.

The NYT’s sources say that new CEO Dara Khosrowshahi had considered shutting down the self-driving project, when he took the helm. Apparently, his mind was changing, but the visit to Arizona was viewed by the team there as a make-or-break moment – which is why they had apparently changed the test driver policy.

LiDAR supplier Velodyne has been quick to get its leadership out in public and claim that its technology was capable of spotting the pedestrian – thanks to its 100-meter range. Similarly, radar should have been able to find Herzberg in the dark, as would the forward-facing camera – if it was using an infrared mode. There has been some speculation that Uber was only using one LiDAR sensor, instead of the total 5 (only the rooftop one, and not the corner units).

Velodyne president Marta Hall said “we are as baffled as anyone else. Certainly, our LiDAR is capable of clearly imaging Elaine and her bicycle in this situation. However, our LiDAR doesn’t make the decision to put on the brakes or get out of her way. It is up to the rest of the system to interpret and use the data to make decisions. We don’t know what sensors were on the Uber car that evening, if they were working, or how they were being used.”

In similar fashion, Aptiv (formerly Delphi, less the drivetrain division) has said that Uber had disabled the safety system software that it supplies to Volvo. The XC90 uses a radar and camera system from Aptiv, and the company is adamant that Uber’s software had cut them out of the loop.

Elsewhere, Intel’s head of its self-driving car program, Amon Shashua (formerly CTO and co-founder of Intel’s Mobileye wing), has said that it is now time to have important substantive conversations about autonomous vehicle safety. He argues that recent development in AI have led some to believe that the process is easier than it actually is, and that computer vision should be central – and of course, that’s conveniently Mobileye’s area of expertise.

Shashua adds that cars should treat each sensor system independently, when processing driving information, rather than combining the separate streams into one view – as is currently done in ‘sensor fusion’ implementations. Mobileye says it has built such systems, and Shashua argues that this architecture provides better redundancy.

On that alone, one could read Shashua’s comments as a criticism of Uber’s approach. However, when Intel begins running its own machine-vision algorithms on the police-released video recording to show how the Mobileye system spotted Herzberg, it becomes very clear that Mobileye is looking to score points.

Similarly, Waymo CEO John Krafcik said to Forbes that “we’re very confident that our car could have handled that situation. We know that for a lot of different reasons. It’s what we have designed this system to do in situations just like that,” in front of an audience at the National Automobile Dealers Association conference.