Work in Progress: Waymo Van Stakeout Reveals the Challenges of Self-driving Tech

Steph Willems
by Steph Willems

The self-driving Chrysler Pacifica vans operated by Alphabet-owned Waymo didn’t know they were under surveillance, but indeed they were. Reporters from the Arizona Republic were on their tail, watching as the autonomous vans — safety driver behind the wheel — tooled around the streets of the Phoenix, Arizona metro area for a period in October. In total, the rolling stakeout covered 170 miles of sun-drenched roadway.

Earlier this week, Waymo announced it had become the first company to offer a commercial ride-hailing service (“Waymo One”) using autonomous vehicles, even though there’s still a live human being behind the wheel. That employee’s job is to monitor the vehicle and take over if needed, as self-driving tech is still in its early days. There’s bugs to be worked out.

What the newspaper’s surveillance showed was that vehicles operating “by the book” — ie, with a strict adherence to the rules of the road and an abundance of caution — sometimes don’t mix well with humans. Go figure.

We knew this already. Last summer, a Waymo Pacifica was filmed attempting to merge onto a Phoenix-area freeway, to no avail. The cushion of safety built into the self-driving vehicle’s actions sometimes means a maneuver you or I would accomplish with ease (or maybe with a little bit of forcing, in the case of merging) proves impossible within the parameters dictated by the vehicle’s computer overlords.

One interesting interaction observed by the Arizona Republic crew came as multiple lanes of traffic were slowed approaching the site of a serious car crash. A police officer was routing everyone around the accident site.

“The self-driving Waymo awkwardly, slowly, rolled towards the scene, even as dozens of other vehicles merged into the turn lanes far sooner,” the paper wrote. “A human driver in this situation might try to make eye contact with the drivers already in the crowded turn lane, or even wave, to try to cut in.”

After some hesitation and a failed attempt, the Waymo van finally got in line.

Lane changes apparently give the vans the most trouble, and merging falls into this category. The Republic noted numerous instances of failed lane changes after the vans chickened out of a maneuver they had already initiated. Sometimes, just the turn signal comes on, then shuts off, with no changing of lanes. This can happen multiple times before the lane change is accomplished. These vans like their buffer zone; the lane change will not occur until the vehicle is damn sure it won’t come into close proximity of another vehicle.

One lane change took a minute and 23 seconds to pull off in moderate, flowing traffic, the publication stated. This repeated, half-hearted attempt to change lanes “appeared to confuse [other] drivers.”

An Early Rider participant, Reid Beer (solid name, by the way), said his van sometimes missed an exit on the highway “because it was being too cautious.”

While missed exits and longer trips are an annoyance for customers, the alternative would be to take on additional risk. Imagine the bad PR resulting from news stories about the brash self-driving van that caused a pile-up on the freeway or bashed another vehicle out of its lane. There’s trade-offs to erring on the side of caution, and Uber aptly demonstrates the other side of the coin.

Waymo officials say they’ve spent months working on making the vehicles more “assertive,” hoping to cut down on the number of instances where a vehicle fails to merge.

Continuing on the topic of safety, the Republic reporters did witness moments when Waymo’s caution spared the vans from disaster. One incident took place at an intersection a Waymo van was about to cross. The Waymo faced a green light, but a Honda Element approaching from the right apparently hadn’t noticed their light turn red. The van stopped short of the intersection while the Element skidded to a stop in the center of the intersection. In another incident, a Waymo van entering a freeway aborted a merge to avoid an aggressive driver attempting to pass.

Does the credit for these interventions go to robots or humans? Waymo won’t say who or what tapped the brakes or grabbed the wheel in those incidents.

The specially selected members of Waymo’s Early Rider program, which offered up free rides earlier this year, won’t be travelling far afield via Waymo One. The vans operate in a specific area that’s been extensively mapped (“geofenced”), comprising areas of Chandler, Mesa, Tempe, and Gilbert, Arizona — though the company says this area will expand over time. Eventually, new paying riders will be able to sign up and the Pacificas will start shedding their safety drivers.

While Waymo succeeded in being the first company to tap a potentially lucrative market, it desperately wants to avoid the kind of black eyes suffered by rival Uber. As such, it’s starting small.

[Image: Waymo]

Steph Willems
Steph Willems

More by Steph Willems

Comments
Join the conversation
2 of 12 comments
  • Russycle Russycle on Dec 07, 2018

    "Does the credit for these interventions go to robots or humans? Waymo won’t say who or what tapped the brakes or grabbed the wheel in those incidents." They're testing these on public roadways, they damn sure should be telling us how well they're performing.

  • Jagboi Jagboi on Dec 08, 2018

    I wonder how it would react to a policeman standing in an intersection directing traffic to go through a red light?

  • W Conrad I'm not afraid of them, but they aren't needed for everyone or everywhere. Long haul and highway driving sure, but in the city, nope.
  • Jalop1991 In a manner similar to PHEV being the correct answer, I declare RPVs to be the correct answer here.We're doing it with certain aircraft; why not with cars on the ground, using hardware and tools like Telsa's "FSD" or GM's "SuperCruise" as the base?Take the local Uber driver out of the car, and put him in a professional centralized environment from where he drives me around. The system and the individual car can have awareness as well as gates, but he's responsible for the driving.Put the tech into my car, and let me buy it as needed. I need someone else to drive me home; hit the button and voila, I've hired a driver for the moment. I don't want to drive 11 hours to my vacation spot; hire the remote pilot for that. When I get there, I have my car and he's still at his normal location, piloting cars for other people.The system would allow for driver rest period, like what's required for truckers, so I might end up with multiple people driving me to the coast. I don't care. And they don't have to be physically with me, therefore they can be way cheaper.Charge taxi-type per-mile rates. For long drives, offer per-trip rates. Offer subscriptions, including miles/hours. Whatever.(And for grins, dress the remote pilots all as Johnnie.)Start this out with big rigs. Take the trucker away from the long haul driving, and let him be there for emergencies and the short haul parts of the trip.And in a manner similar to PHEVs being discredited, I fully expect to be razzed for this brilliant idea (not unlike how Alan Kay wasn't recognized until many many years later for his Dynabook vision).
  • B-BodyBuick84 Not afraid of AV's as I highly doubt they will ever be %100 viable for our roads. Stop-and-go downtown city or rush hour highway traffic? I can see that, but otherwise there's simply too many variables. Bad weather conditions, faded road lines or markings, reflective surfaces with glare, etc. There's also the issue of cultural norms. About a decade ago there was actually an online test called 'The Morality Machine' one could do online where you were in control of an AV and choose what action to take when a crash was inevitable. I think something like 2.5 million people across the world participated? For example, do you hit and most likely kill the elderly couple strolling across the crosswalk or crash the vehicle into a cement barrier and almost certainly cause the death of the vehicle occupants? What if it's a parent and child? In N. America 98% of people choose to hit the elderly couple and save themselves while in Asia, the exact opposite happened where 98% choose to hit the parent and child. Why? Cultural differences. Asia puts a lot of emphasis on respecting their elderly while N. America has a culture of 'save/ protect the children'. Are these AV's going to respect that culture? Is a VW Jetta or Buick Envision AV going to have different programming depending on whether it's sold in Canada or Taiwan? how's that going to effect legislation and legal battles when a crash inevitibly does happen? These are the true barriers to mass AV adoption, and in the 10 years since that test came out, there has been zero answers or progress on this matter. So no, I'm not afraid of AV's simply because with the exception of a few specific situations, most avenues are going to prove to be a dead-end for automakers.
  • Mike Bradley Autonomous cars were developed in Silicon Valley. For new products there, the standard business plan is to put a barely-functioning product on the market right away and wait for the early-adopter customers to find the flaws. That's exactly what's happened. Detroit's plan is pretty much the opposite, but Detroit isn't developing this product. That's why dealers, for instance, haven't been trained in the cars.
  • Dartman https://apnews.com/article/artificial-intelligence-fighter-jets-air-force-6a1100c96a73ca9b7f41cbd6a2753fdaAutonomous/Ai is here now. The question is implementation and acceptance.
Next