Not being able to change lanes as traffic creeps along is strange. There are many cities where you are creeping along and you have to change lanes. Zipper merges in construction zones. Having to go from one Interstate to another through a City in heavy traffic.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
willow_hiller
Well-Known Member
It's all about limiting liability. Mercedes must have decided that changing lanes creates too many scenarios where their system could be at-fault for a collision. Meanwhile, if it stays in a single lane, there are only 3 scenarios I can think of that would create potentially expensive liability for Mercedes:
1. Striking the vehicle in front
2. Striking a VRU
3. Striking an object or obstacle in such a way that the occupants are injured
And even those scenarios are largely mitigated by requiring a tracked lead vehicle, requiring lower speeds, and requiring operation on a closed highway. Any other scenario where the only damages are to the Mercedes itself would probably be pretty inexpensive for them to settle with the driver.
Driver hits woman in S.F., then Cruise driverless car runs her over
A hit and run on a pedestrian that knocked the woman in front of a Cruise. The bulk of the article is about the Cruise vehicle and the dangers of autonomy, not that people can hit and run pedestrians. No mention of how the pedestrian fared. Just cracker jack reporting.
diplomat33
Average guy who loves autonomous vehicles
It is already being discussed in the Cruise thread. No need to discuss the same thing twice in two different threads.
diplomat33
Average guy who loves autonomous vehicles
willow_hiller
Well-Known Member
This is a rough piece of work for them to be showing off at this point. Look at some of the weird artifacts in the generated imagery.
Pedestrian shapes merging into background vehicles and other obstacles:
Garbled road markings and lane guidance signs:
This reminds me of the state of image generating neural networks from several years ago. It may have a world model for internal consistency in the movement of ego and other objects in the scene, but I don't think it understands the nuances.
diplomat33
Average guy who loves autonomous vehicles
You bring up a good point. Wayve is trying to build the simulator that will train their end-to-end. That is what GAIA-1 is. But if the simulator has flaws, as it will, then it could corrupt the training. So I think wayve could have problems if they rely too much on GAIA-1 for training their e2e. GAIA-1 is pretty cool but I think there will be challenges. But I think I know why Wayve is focusing so much on GAIA-1. Wayve has a small fleet. They don't have enough real data to do e2e training at scale so they are hoping to compensate with synthetic data.
willow_hiller
Well-Known Member
Chuck Cook getting some Waymo testing in:
It never occurred to me that being outwardly recognizable as an AV would be an advantage in terms of managing other driver expectations. In this short clip, the Waymo did not have the right of way, but other drivers still yielded to it because they could recognize it as an AV that they didn't want to engage with.
It never occurred to me that being outwardly recognizable as an AV would be an advantage in terms of managing other driver expectations. In this short clip, the Waymo did not have the right of way, but other drivers still yielded to it because they could recognize it as an AV that they didn't want to engage with.
diplomat33
Average guy who loves autonomous vehicles
Yeah. This is an interesting question that experts are still debating: should AVs have a distinct look that makes them easily identifiable as an AV or should AVs blend in and look like any other normal vehicle? As you point out, if the AV looks different then it will change the behavior of other road users. This could be a good thing in some cases if it leads human driven vehicles to drive more cautiously. Or could it cause more hostility towards AVs like we see in SF with the coning? I think one argument for making AVs look normal is that you want all vehicles to behave normally and not drive differently. Plus, for consumer cars, there is the aesthetic argument that people want a "nice looking" car, so they don't want to see big sensors on the car. It is one reason why Tesla went vision-only. Personally, I think I learn towards making AVs look different. I think it is important for other road users to know that a car is an AV since AVs do drive differently than humans. If AVs look like normal cars, people might assume it is driven by a human and expect certain behavior and then get surprised when the car drives differently. This could cause humans to make prediction mistakes about how the AV will drive, which could cause accidents.
A related question is should the AV car have status lights to indicate when it is driving in AV mode or not? Sometimes these cars are manually driven. Obviously this does not apply however to AVs with no manual controls.
diplomat33
Average guy who loves autonomous vehicles
Personally, I would say "yes", AVs should have status lights to indicate when it is in autonomous mode. I think it is important for other road users to know when a car is in autonomous mode or being driven manually since that affects how it drives. And this would be a good solution for AVs that look like regular cars.
willow_hiller
Well-Known Member
diplomat33
Average guy who loves autonomous vehicles
You will note that the protest was not just the teamsters but also involved other community groups opposed to AVs. So it was an assortment of various groups, who found common cause in opposing AVs, who joined forced and picked on Waymo because they are an obvious target.
Strike while the metal's hot. The UAW is running amok with the big three, so now is the time to get the union members moving on issues.
Well, to me the obvious target is Cruise, because they clearly suck, while Waymo is the bigger name.
willow_hiller
Well-Known Member
They picked it because the slogan "Waymo Hell No" was easier to chant than "Refuse Cruise."
diplomat33
Average guy who loves autonomous vehicles
Yeah. I was thinking that since Waymo is the bigger name, hence the one that might get targeted more.
Doggydogworld
Active Member
Waymo is the most competent, and thus the biggest threat.
I wonder if they can do the opposite of typical Generative-Adversarial networks. Usually the adversarial part is trying to distinguish real from synthetic and the generative part trying to minimize that.
But for training drive policy they would want to make it as insensitive as possible to the difference between the generated and physically observed data---make it invariant.
voyager
Active Member
"There has been some grumbling that some of the negative media attention being given to Cruise has been spurred on
by corporate warfare. Cruise’s main rival is Waymo, a subsidiary of Alphabet which likewise happens to own Google.
Interestingly, Waymo doesn’t seem to be getting the same negative attention in San Francisco that Cruise has despite
the former having been chided for similar antics in and around Phoenix, Arizona."
www.thetruthaboutcars.com
by corporate warfare. Cruise’s main rival is Waymo, a subsidiary of Alphabet which likewise happens to own Google.
Interestingly, Waymo doesn’t seem to be getting the same negative attention in San Francisco that Cruise has despite
the former having been chided for similar antics in and around Phoenix, Arizona."
NHTSA Launches Safety Investigation Into GM's Cruise
General Motors’ Cruise may be one of the first companies in the world to field driverless vehicles as part of a commercial endeavor. But this doesn’t appear to have endeared the brand to everyday people. The business appears to be loathed by San Franciscan residents and emergency response crews...
www.thetruthaboutcars.com
Similar threads
- Article
