The logical outcome of this incident will be that the default response to ANY detected object will be a maximum-ABS stop in response to a piece of paper detected ahead of the vehicle.
My calculations above are based on a reasonable deceleration rate. If you are paying attention and following at a safe distance, you will have time to slow down. Robot stops will be annoying not dangerous.
Uber disabled the Volvo automatic braking system. The car had the lady in view for 6 seconds before impact. It did NOTHING except reclassify her for the next 4.7 seconds, and then realised it was in trouble, handed control back to the driver, and did not warn her.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
And it waited until the very last second to hand control over. At six seconds, it could have rung an alarm, and the driver would have had time to figure things out.
"My interpretation of the article is that the default behaviour is to not slow down. "
And that's is the BIG problem. Anyone else at that spot, in that situation, would have slowed down or changed lanes to the right to avoid the potential conflict. The fact that the car was programmed to ignore the uncertainty in the situation is problematic, particularly when it detected that the pedestrian was not completely stationary.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!
"According to data obtained from the self-driving system, the system first registered radar and LIDAR
observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph.
As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian
as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path.
At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver
was needed to mitigate a collision (see figure 2).
2 According to Uber, emergency braking maneuvers are
not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle
behavior. The vehicle operator is relied on to intervene and take action. The system is not designed to
alert the operator. "
So whether it is unknown or a vehicle or bicycle, it was never something that could have been ignored.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
Looks to me like the first car was running the red light, and the Waymo car appears to be going too fast facing a red light at an approaching intersection. It appears to have been only about 1.5 seconds from the start of the swerve to the impact. A more cautious and aware driver might have changed lanes to right to provide extra buffer, and given themselves about 2 seconds to avoid a potential collision with the light runner.
Supposedly, the Waymo car wasn't in control, but it would seem to me that the sensors probably "saw" the scene unfolding, i.e., a car clearly about to run the red light, obstacles moving into that car's path, and a high probability of a collision there or a swerve into the Waymo car's path. It seems to be something that could be programmed for, but obviously, it would require additional processor horsepower, and more processor, and more cost. This would seem to be an ideal case study for what an autonomous car could see and whether it should be programmed for it.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!
A scenario I get into all the time is turning left at a light that is turning yellow and red, and trying to decide what the oncoming car is doing. Will he run the yellow, or stop? This ought not to be a problem for a robot. The robot works out his acceleration to determine the driver's intention, then errs on the safe (no accident) side.
"The robot works out his acceleration to determine the driver's intention, then errs on the safe (no accident) side."
The safe thing to do in that situation is always to not make the turn, as long as you're not already in the intersection, which you're not supposed to be. An AV should be programmed to stay in the turn lane and not try to guess whether the oncoming car will stop.
You already are in the intersection, and you are turning. The question is whether or not you pull in front of the on-coming vehicle. Sometimes, they run the red light. The robot must allow for that.
Per the traffic laws, you are not supposed to pull out into the intersection until it is clear to make the turn. Obviously, most people just pull out and then wait for an opening, but that's not the correct way. Presumably, the AV would be programmed to do it the correct and legal way.
If I got caught behind you at a light, I would hate you.
Here in Ontario Canada, I failed my first driver test by turning left on a yellow light. Turns are to be done at low speed. You have to stop for the yellow if you are going slow.
You can pull into the intersection on the green light. If you pull right into the intersection, the people behind you pull into it too, and they all get through when the light changes. The police here would like us to do an "S" turn, pulling to the left as far as possible so that we can see, and keeping the wheels straight so that we are not knocked into the opposite lane when we are rear-ended.
I have gotten angry looks and gestures from other drivers on occasion, but usually not for staying behind the line at a left turn. I've seen more people get honks and angry gestures because they were still in the middle of the intersection after the light changed.
In any case, my comment was about AV programming, which I would assume would have to be in accordance with the law, even if it's not common practice for human drivers.
"It's not obvious that HotRod10 is one of the few..."
I am actually, not because of the law, but because it's the safer way.
When in freeway traffic in Denver, I flow with the speed of traffic, regardless of the posted speed limit. In that situation, speeding is the safer way. Not likely an AV would be programmed to exceed the speed limit, even if it was safer to do so. An AV on I-25 in Denver is going to drive 55mph while everyone else is going 75mph. Does that enhance safety for anyone?
I've always considered being on the spot prior to the turn was safer than staying behind the line; the issue is that in a large intersection that adds a second or so to the turn, and makes you more vulnerable to accidents, since you have to get to the point where the actual turn starts.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!
I've mentioned this before. Isn't part of the problem that the world view uses Kalman filtering to establish likely trajectories for the objects around it, which assumes steady state accelerations? Why not consider the full cloud of possible future trajectories each object can make? A pedestrian can accelerate by about 1g in almost any direction inside of 200 ms, for example.
Cheers
Greg Locock
New here? Try reading these, they might help FAQ731-376
A safe robot that has not reliably identified the object it is driving past must leave space. I passed a pedestrian just this evening, who was weaving quite a bit as he walked along the side of the road. I left extra space. We can work out a robot's ability to scan with LiDAR, radar or video and identify targets. We can then work out how close the robot can get to those targets.
"Why not consider the full cloud of possible future trajectories each object can make? A pedestrian can accelerate by about 1g in almost any direction inside of 200 ms, for example.
I'm not sure they are using Kalman filters; there are a number of tracking algorithms that do better than KFs, like particle filters or multi-hypothesis trackers. A KF is almost old-school these days.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!
![[evil]](/data/assets/smilies/evil.gif "[evil] [evil]")
