Yes, people crowding around the AV definitely makes the stall worse because AVs are programmed not to move when there are pedestrians so close to it or even touching it. I think many don't know this about AVs because they don't know how AVs work. But I believe some people also do it on purpose because they want to make the stalls look worse in order to push an anti-AV agenda.
I think it is very clear now that Cruise is not ready to scale as fast as they wanted to. I cringed when I heard Vogt talk about doubling driverless miles every month or scaling to like 7-8 cities in a year. That seemed way too fast for a company that is having so many issues.
I wonder if Cruise was lulled into a false sense of over-confidence because of their high disengagement rate. Last year, they reported a disengagement rate to the CA DMV of almost 1 disengagement per 100,000 miles. That might have made them think they were ready to scale. And they assumed that that they could start with driverless at night with less traffic, and risk would be low, so it would be ok. But as we see, as they scaled, they started encountering a lot of situations that caused problems. It is a cautionary tale that just because your disengagement rate looks good, does not necessarily mean you are ready to scale everywhere. I would also note that I think they only did like 800,000 miles so it was a relatively small sample. So I think that might have made their high disengagement rate not a statistically accurate representation of their true capability.
Yes, that is Cruise's programming, just stop when confidence is "too low". I think the idea is that stopping is generally less risky because you can't hit anything when you are stopped. If the AV lacks confidence but it acts on it anyway, it could do something really unsafe and hit something, but if it just stops, it might block traffic and cause some annoyance but it won't hit anything. So stopping is the less risky choice. And of course, if the AV is stopped and another vehicle hits it, then technically Cruise can claim they were not at-fault. So while stopping might not actually be safe, it might protect you from liability.
This strategy might be ok if they were rare. I don't expect AVs to always handle everything perfectly. Heck, even humans can sometimes encounter a situation we are not 100% sure what to do. It's happened to me. The problem with Cruise is that they are happening too often and in unsafe situations like in the middle of intersections. Either Cruise software is really bad because it is unsure a lot or Cruise turned up the threshold too high so the car will stop at even the slightest drop in confidence. I wonder if Cruise is using it a crutch, figuring they can go driverless and just have the car stop when it is not sure because like you said, it won't count as an accident.
IMO, there are two instances where an AV should never stop in the middle of the road because of lack of confidence, unless it is for a safety critical reason. One is highways and two is intersections, especially big busy ones. On other roads, if the AV wants to pull over when it is not sure, that's fine, but it should not just stop in the middle of an intersection just because it is not sure what do. And we've seen with Cruise that apparently most road closures will cause the car to just stop and freeze. That's not good. So yeah, I think Cruise needs to put the safety driver back in and make their system much more reliable and more robust before they remove the safety driver. But I am sure GM won't like that. IMO, the AV has to reliably handle intersections and road closures before it can go driverless. There are many other things it needs to do reliably as well, but those are two big ones.
