What if a Waymo Robotaxi Causes a Fatality?

Article By : Junko Yoshida

"Autonomous" taxis still have human backups watching out. Some AV company just bought watch-the-watcher technology. Was it Waymo?

What if Waymo’s robotaxi — currently wheeling willing early Waymo riders around Phoenix — kills an unwilling pedestrian?

“That will be a game over,” said Colin Barnden, lead analyst at Semicast Research, “for the autonomous vehicle (AV) industry.”

The standard assurance is that this is unlikely. But never say never.

Waymo is known for the most advanced autonomous vehicle (AV) technology and the most AV miles racked up in real-world driving.

Moreover, Waymo One, its robotaxi service, has not taken safety operators out of the car yet, although that was what Waymo CEO John Krafcik promised at the 2017 Web Summit in Lisbon. Even though technically the AV technology drives the car, Waymo’s robotaxi has a backup human behind the wheel to take over in case the machine goes haywire.

But think about it.

If so much of the AV industry’s future is riding on Waymo One, isn’t it imperative for Waymo to make sure its safety operators are not overworked and stay alert to properly function as a backup driver?

Recommended
Bringing Robocar Fans Back Down to Earth

Seeing Machines targets AV testing vehicles on public roads

Seeing Machines (Canberra, Australia) last week announced an agreement to supply its Guardian Backup-driver Monitoring System to a leading North American self-driving car company.

Seeing Machines is not disclosing who that company is.

Seeing Machines said, “Based on an initial purchase of A$2.2 million (US$1.5 million), Seeing Machine’s Guardian Backup-driver Monitoring System is expected to be installed into a significant portion of the self-driving car company’s fleet over the coming months, with an option for follow-on purchases under the agreement.”

Nobody should be surprised if the customer happens to be Waymo, according to Barnden. Waymo, among all AV companies, has the most to lose if another fatality like Elaine Herzberg (Uber ’s accident victim last year) occurs.

Seeing Machines, in the business of driver monitoring technology since 2010, is a globally recognized company for advanced computer vision technology. While it launched last fall a new hardware called Guardian Backup-driver Monitoring System, Barden suspects that Seeing Machines has spent recent months completely reprogramming its software specifically for a backup driver doing on-road tests of autonomous vehicles.

“There are other companies such as Lytx and Nauto that specialize in fleet management and driver monitoring,” said Barnden. “But so far, Seeing Machines is the only company which has custom designed a driver monitoring system for AV testing vehicles.”

Why reprogram software for a safety driver in a test vehicle? Barnden said that a safety driver behaves differently from a normal driver in a conventional car.

Instead of constantly keeping an eye on the road ahead, safety operators in a test AV tend to look all over the place, for example, for lane markers, free space on both sides of the road and other matters machines must heed. “You don’t want to use a driver monitoring system used for conventional cars, because it could unnecessarily punish safety drivers for doing their jobs and annoy them,” explained Barnden.

According to Seeing Machines, the primary function of the camera-based Backup Driver Monitoring System is to track the backup driver’s face and eyes during on-road automated or semi-automated vehicle testing. It reports driver state information (e.g. on-road, or off-road attention state), and identifies distractions of increasing severity (e.g. insufficient driver attention to the road scene).

“Driver state and video data are made available via an Ethernet connection to a central data collection client computer provided by the customer for possible synchronization with other vehicle/sensor data inputs,” Seeing Machines explained. “The system also provides configurable escalating real-time alerts (haptic, audible, visual) meant to train and alert the backup-driver to maintain appropriate attention to the road scene and (back-up) driving task, whether the automated driving system is turned on or off.”

Seeing Machines’ Guardian Backup Driver Monitoring System will not be used to punish safety operators. It will, rather, inform robotaxi operators when to call test vehicles back to home base when the monitoring system detects signs of fatigue in backup humans, said Barnden.

Do you have the humility to use DMS?

In the world of AVs, where key stakeholders are known to play one-upmanship over compute power, advanced sensors and AI algorithms, the adoption of a driver monitoring system (DMS) has always taken a backseat.

But if any of AV companies “have the humility to use DMS,” testing of R&D vehicles on public roads would become much safer, observed Barnden. “And Elaine Herzberg would not have been killed by Uber.”

Without question, the Uber accident last year prompted many players in the AV industry to seriously reassess the self-driving car market.

A growing consensus centers on the thesis that an AV that is just as safe as a human driver is not good enough. It should be ten times better.

Seeing Machines believes that eventually the world can entrust its driving to computers, but “the journey will be gradual, complex, and potentially dangerous.”

As automated driving technology is tested and refined, Seeing Machines argues that a common and prudent practice is to “have a human backup-driver who does just that — backs up the yet-imperfect automated driving system to help assure safe operation of the vehicle at all times.”

Specifications

The Guardian Backup Driver System is designed for retrofit into current automated driving test vehicles/fleets. It operates in real world lighting and environmental conditions.

The system tracks the face and eyes over a very wide angular range and recovers tracking immediately when face or eyes become visible.

Seeing Machine also noted that the system can be integrated into existing user back-end data collection server/cloud (via ethernet), if provided with SDK API — which exposes all real-time event/alert data, video, and system diagnostic/health information. The system offers 1Mbps CAN message output option (via USB3 to CAN adaptor) with Driver Attention State, Driver Distraction Event, Alert and system health information.