Waymo operates a fleet of self-driving taxis in the East Valley area of Phoenix: they are Chrysler Pacifica vans with Level 4 autonomy, which do not require the use of a human operator tasked with taking over in case anything goes wrong. If that happens – and it does in the video – the car stops and an operator comes online to speak with the passenger, while a roadside assistance vehicle and team is dispatched to solve the issue.
In this particular case, the issue was traffic cones; more specifically, the computer did not know how to interpret the presence of cones in an extended construction area, so it kept trying to carve a path between them. In reality, this translated into three occasions in which the car stopped in the middle of the road, blocking traffic and stalling for minutes on end. While the computer “debated” on the best course of action, it turned on the hazard lights, so there was that at least.
Another strange occurrence here is that the Waymo fled the support vehicle twice, which would have otherwise ended a bumpy journey earlier. When the support team did manage to get close to it, they were able to turn it off and then conclude the journey manually.
The video is 35 minutes long, but it’s timestamped excellently, in case you want to get to the best parts. It’s worth a watch in full, to give you an idea of how much work is still to be done in terms of AVs. If such an advanced computer can become baffled by such a small thing as a traffic cone in a geo-fenced area, one can only imagine how it would fare against the unexpected challenges in a real-world scenario.
On the other hand, the video also highlights how far AVs have come. Despite the strange and repeated occurrence, the Waymo never does anything to put the passenger at serious risk. Sure, he could not have gotten out of the vehicle and, had he been in a hurry, would have lost precious time, but at least he was not in any immediate danger.
In a statement, Waymo says the strange behavior of the Waymo AV was due to “incorrect guidance” from the Fleet Response team. “Our team has already assessed the event and improved our operational process,” the company adds.