Waymo

[Daniel in SD]

Specifically in this case perhaps it was possible to see the cyclist come out of the bike lane (if it existed; have not looked at the map) and enter the middle of the traffic lane behind the truck, well before the collision and before anyone stopped.

That would suggest a left turn by the cyclist. Would that not result in some caution?

I wonder what happened and what was visible 15-20 seconds before the collision.

There are bike lanes. It seems plausible that the Waymo never saw the cyclist because they were behind the truck the entire time. Maybe the cyclist even arrived at the intersection before the Waymo. But I wouldn’t think that reaching the intersection in the bike line gives one right of way to turn left.

 

[enemji]

Ain’t no one got time for that.

If the CEO of Waymo is of this thought, I would like him to be in an accident and have his family learn it the hard way.


I have all the time. I am not going to miss winning a million dollars because I took 1 more second to ensure safety.

 

[AlanSubie4Life]

It’s just not part of my defensive driving to delay passage through an intersection because there’s a tiny chance that someone will do something illegal.

For a VRU which you hypothetically saw moving to make a left turn, I suspect you might exhibit caution, because you would reasonably assume they were moving behind the truck (they might have known you were not turning, having observed no signal as they saw you approaching the intersection, before settling in to hide behind the truck). The only unknown would be whether or not they were going to veer unexpectedly into your path. You would even know they were in the intersection since they weren’t anywhere else (hypothetically). I would guess at least a little bit of caution, even if you were not expecting an illegal move.

We’ll never know of course whether this was the situation.

Maybe the cyclist even arrived at the intersection before the Waymo. But I wouldn’t think that reaching the intersection in the bike lane gives one right of way to turn left.

No but if the Waymo saw the cyclist approaching and NOT in the bike lane (we don’t know whether it did - it’s very possible the cyclist was hidden the entire time but we’d need video of the lead up to know - either is possible), it seems like the only conclusion was that the bike was going to turn left. No other reason to be in the traffic lane! One of the approaches to the intersection does not have a bike lane, but not sure if applicable (probably not).

It’s interesting that the Waymo only applied heavy braking when the bicyclist was fully visible. That seems like a bit of a major failure that Waymo is confessing to! I had not noticed that earlier.

For example bicycles would be readily visible behind both this cement mixer and this jeep when fully behind such vehicles, even (especially, actually) at a considerable distance. It just seems a bit hard to believe the bicycle was not visible at any time (though it is definitely possible, it seems unusual).

In any case applying hard braking only when the VRU is fully visible seems like a bit of an admitted failure. Sounds like the cyclist was visible but not properly tracked - otherwise much different wording would have been used in the carefully vetted press release. Something to work on!

 

[Tam]

Here's an article with the following wording:



That scenario makes a lot more sense than the bike following the truck through a left turn. In this case, the bike can be obstructed by the truck for a much longer time. So the biker doesn't make sure the intersection is clear, make his turn and marries the Waymo, which never had a chance to see the biker.

Viewed that way, it seems like much ado about nothing, with a biker not using much sense. My only hesitancy in condemning the biker with this theory is that it requires a lot of irresponsibility on his part.

Thanks for the link. In that scenario, the bicyclist was at fault for making an unsafe left turn.

1) It should have waited for its turn and not bundled the turn together with the truck. There should be one vehicle at each stop sign, not a truck and bike as one vehicle. That's a failure to stop at a stop sign.

If the bicyclist let the truck go first and it would stop after the truck moved, it would be visible.

2) The turning-left bicyclist should yield the right of way to the straight-going Waymo.

Legally, the bicyclist is at fault in this article, but as a human driver, I would not want to hit someone and then have to prove myself innocent.

As a robot, Waymo should be cautious and avoid accidents, regardless of who is at fault.

 

[kabin]

Without any details of the incident I'll bet the cyclist tried to run the stop sign. All sorts of possible scenarios but I could picture the biker trying to maintain speed and followed the truck into the intersection without stopping.

 

[Daniel in SD]

it seems like the only conclusion was that the bike was going to turn left

Not true. Sometimes I’ll ride in the traffic lane if I’m planning on turning left after the intersection because I know it will be harder to get to the left later.
Though I’m having trouble visualizing how the cyclist would be seen moving out of the bike lane to behind the truck but then become not visible as the truck went through the intersection.

 

[kabin]

"Misleading"?... An experienced driver would have anticipated this and waited until he/she got a clear sight of what's behind the truck.
But I guess this will be learning material for Waymo.

The right of way is based on a protocol for the vehicles presented at the 4 way stop not invisible moving vehicles waiting in line.

 

[AlanSubie4Life]

Though I’m having trouble visualizing how the cyclist would be seen moving out of the bike lane to behind the truck but then become not visible as the truck went through the intersection.

Yeah it depends on the situation. Could have been 20-30 seconds before the collision - hard to know.

I agree if you did not see the movement from bike lane to traffic lane it would be harder to predict what the bike was going to do.

Note that Waymo did NOT say the bicyclist was not visible. They said it was occluded: “The cyclist was occluded by the truck” leaves a bit of ambiguity, since occluded does not necessarily mean completely invisible. A bit of a semantic argument revolving around word definitions, but I wonder about what was visible and when…

Need the video which we won’t get, due to woefully insufficiently strict regulations.

Without any details of the incident I'll bet the cyclist tried to run the stop sign. All sorts of possible scenarios but I could picture the biker trying to maintain speed and followed the truck into the intersection without stopping.

Seems like they certainly did run the stop sign. Seems to be little doubt that the cyclist was at fault here, whether they stopped and then followed the truck or sailed through, not realizing a car was coming.

My only question is whether the cyclist was visible. That seems unclear from the press release. It is very possible that the cyclist was invisible (a long truck would make this quite possible), and the only failure of Waymo is that the cyclist had to be fully visible before hard braking was applied (a strange limitation!).

 

[flutas]

Here is statement from Waymo:

"On February 6th at 17th Street and Mississippi Street in San Francisco, one of our vehicles was involved in a collision with a bicyclist. The Waymo vehicle was at a complete stop at a four-way intersection. An oncoming large truck progressed through the intersection in our direction and then at our turn to proceed, we moved into the intersection. The cyclist was occluded by the truck and quickly followed behind it, turning left and crossing into the Waymo vehicle’s path. When they became fully visible, our vehicle applied heavy braking but was not able to avoid the collision. Waymo called police to the scene and the cyclist left on their own, to our knowledge reporting only minor scratches."

So do we actually know the scenario anything beyond this? The way it reads I can see at least 2 scenarios.

Scenario 1: The cyclist was directly behind the truck and turned into the path of the Waymo vehicle out of turn.

Scenario 2: The truck was going straight across the intersection, and the bicycle was turning to be on the same road as the Waymo vehicle.

If it's Scenario 1 it's the bikes fault.

Scenario 2 would come up to who had the right of way. Waymo tries to assert that they had the right of way in their statement, but if they never saw the bicycle until it became "fully visible" who's to say the bike wasn't stopped first, with the truck going slowly across the intersection, Waymo approaches and decides to proceed thinking there's nobody else waiting and then hits the bicycle, who would have had the right of way.

I guess until we see video or a more conclusive accident report we won't know.

 

[diplomat33]

So do we actually know the scenario anything beyond this? The way it reads I can see at least 2 scenarios.

Scenario 1: The cyclist was directly behind the truck and turned into the path of the Waymo vehicle out of turn.
View attachment 1016925

Scenario 2: The truck was going straight across the intersection, and the bicycle was turning to be on the same road as the Waymo vehicle.
View attachment 1016926

If it's Scenario 1 it's the bikes fault.

Scenario 2 would come up to who had the right of way. Waymo tries to assert that they had the right of way in their statement, but if they never saw the bicycle until it became "fully visible" who's to say the bike wasn't stopped first, with the truck going slowly across the intersection, Waymo approaches and decides to proceed thinking there's nobody else waiting and then hits the bicycle, who would have had the right of way.

I guess until we see video or a more conclusive accident report we won't know.

Without a video, there is really no way to know for sure.

Personally, I lean towards scenario 1 because I think it is the one where the cyclist would be occluded the most and therefore seems to fit how the Waymo would not have seen the cyclist until the last second. In scenario 2, it seems the cyclist would have been visible to the Waymo before the truck passed. I don't really see how the cyclist would have been occluded.

 

[AlanSubie4Life]

I guess until we see video or a more conclusive accident report we won't know.

“The cyclist was occluded by the truck and quickly followed behind it, turning left and crossing into the Waymo vehicle’s path.”

Seems like scenario 1. I agree we can’t be 100% certain but I can’t read this and make it fit scenario 2. Following (behind!) has specific meaning (it is too sloppy to refer to following in sequence/temporally) - as does “fully visible.”

Waymo tries to assert that they had the right of way in their statement, but if they never saw the bicycle until it became "fully visible" who's to say the bike wasn't stopped first, with the truck going slowly across the intersection,

They were assertive here in their statement (“our turn to proceed”). I suspect they know with certainty they had right of way because they started moving before the truck had cleared the stop line on the other side of the intersection (in other words they were stopped before the truck had cleared the stop line).

 

[JB47394]

Scenario 2 would come up to who had the right of way.

The Waymo has right of way. Knowing that the truck was going straight, and not having any cars to its right, the Waymo was safe to proceed. Even if anyone was to the Waymo's left, the Waymo would have right of way. The biker, on the other hand, had to yield to traffic to his right. He couldn't see to his right, so he shouldn't have moved.

 

[diplomat33]

Driving in rain with some interesting cases:

0:00 Continued from previous episode
0:16 Reversing to pull out
2:12 High speed nudge for vehicle with open door
2:42 Shooting narrow gap with stalled vehicle
4:25 No reaction to active parking enforcement vehicles
5:20 Pull over & pull out
5:57 Puddle with spray
6:01 Passing stalled vehicle at high speed
6:23 Uncomfortable trajectory with misclassified garbage bin
8:43 Lane change with pedestrian in street
10:04 Discussion about speed oscillation
12:25 Missed light cycle due to following another Waymo
17:55 Right on red
18:12 Entering parking lot

 

[kabin]

My only question is whether the cyclist was visible. That seems unclear from the press release. It is very possible that the cyclist was invisible (a long truck would make this quite possible), and the only failure of Waymo is that the cyclist had to be fully visible before hard braking was applied (a strange limitation!).

It is curious. Maybe semantics for increased dwell time and/or being fully visible is a requirement for high object confidence to reduce false alarms, label the object, as well as determine its speed, range, and path before having enough input to apply the brake.

 

[diplomat33]

It is curious. Maybe semantics for increased dwell time and/or being fully visible is a requirement for high object confidence to reduce false alarms, label the object, as well as determine its speed, range, and path before having enough input to apply the brake.

I do wonder how long did it take for the cyclist to go from partially occluded to fully visible because I think that matters too. Was it a case where the cyclist was fully occluded, became partially visible and it took a couple seconds to become fully visible and the Waymo did not react until the cyclist was fully visible? If so, then yeah, I would say Waymo needs to work to improve that. But if it was a case, like scenario 1, where the cyclist was fully occluded and suddenly turned became fully visible in a split second, then that is different.

 

[AlanSubie4Life]

where the cyclist was fully occluded

Note that Waymo did not indicate the cyclist was fully occluded.

Certainly possible the cyclist was fully occluded - not excluded by Waymo’s statement.

And the transition could have taken less than a second.

Would have to have video for that (and that would not even answer it since the lidars have different views which might differ from the video).

 

[DanCar]

How many milliseconds / seconds between partially visible when people may have reacted and fully visible when Waymo reacted? Just a curiosity that would be answered by the video. I think a law should be created to make this kind of info available from driverless companies. I'm confident Waymo was not at fault.

 

[AlanSubie4Life]

How many milliseconds / seconds between partially visible when people may have reacted and fully visible when Waymo reacted? Just a curiosity that would be answered by the video. I think a law should be created to make this kind of info available from driverless companies. I'm confident Waymo was not at fault.

Since we’ll never know, I will guess 15-20 seconds.

It’s annoying that we won’t know whether it was 500ms (which is probably more likely).

 

[ThomasD]

It may or may not be Waymo's fault but the Waymo didn't use situational awareness to make sure the path was clear. It was trained to turn after a vehicle has gone past it. Not to see if another vehicle may be following too closely behind that vehicle and may or may not be recognized.

 

[Daniel in SD]

It may or may not be Waymo's fault but the Waymo didn't use situational awareness to make sure the path was clear. It was trained to turn after a vehicle has gone past it. Not to see if another vehicle may be following too closely behind that vehicle and may or may not be recognized.

The Waymo did not turn.