Definitely a big concern, but given the number of times in my lifespan that I've seen pictures or video of human-driven vehicles that have got stuck on railroad crossings (or just straight up drunk people trying to drive linearly down a railroad track)...
I would be curious to compare stats of 100,000 hours of human drivers getting stuck on grade crossings or doing something dumb, such as trying to drive around crossing barrier arms, vs 100,000 hours of automated driving. I would bet the automated driver does a lot better.
I recently saw a video from (I think not Phoenix) of 3 waymos that were next to each other blocking traffic in an intersection, refusing to move, because they were facing a traffic signal intersection where the signals had reverted to blinking red mode. Humans who paid attention when learning to drive will understand this means the intersection has reverted to a 4-way-stop due to the traffic signal failure.
The problem is that multiple red lights were blinking in view of the waymos not in sequence with each other, so the waymos interpreted it as a alternating-blinking red railroad signal crossing, and all of them refused to proceed, even when it was their "turn" in a 4-way-stop arrangement.
> The problem is that multiple red lights were blinking in view of the waymos not in sequence with each other, so the waymos interpreted it as a alternating-blinking red railroad signal crossing, and all of them refused to proceed, even when it was their "turn" in a 4-way-stop arrangement.
What's the hot fix for this? Are they just stuck until a tech can physically go out and reset and move them? Or can someone in a office somewhere remotely get alerted, look at the video feed/data, and override it with instruction on how to proceed?
Silly stuff like this happens all the time even with human drivers, I feel like the important piece when hearing that the technology encountered an issue is how long did it take to resolve?
This is the right way to look at it. For autonomous fleets, there are typically tiers of intervention, starting with a simple remote check - "can I drive through this?" type confirmation, to much more detailed remote instructions that are slower to give, to getting someone from operations out (or in an emergency first responders) to manually move the car. One reason why you might want to keep traditional controls in the vehicle for the near term.
It's a big operations challenge, and hope Waymo (and everyone else TBH) get it smoother and smoother.
> Or can someone in a office somewhere remotely get alerted, look at the video feed/data, and override it with instruction on how to proceed?
For Waymo, it's this one.
> or just straight up drunk people
Waymo. Slightly better than an irresponsible alcoholic. As long as the maps are up to date.