[NA]Driverless 18-wheelers

I don't think they need to do as well as humans. I think they just need to operate for long enough without killing a bunch of people for people to get used to then. After that, people will blame something else rather than the autonomous trucks, because by that time we're already used to them. It'll be the fire departments fault that they were run over when at an accident.

By example, when cars were first introduced, as I understand it, people were highly upset that they couldn't stop fast enough to avoid hitting people who were in the road. Brakes didn't improve, and cars didn't slow down, people just learned to not hang out in the street and look and listen before crossing a road. Well, except lately in NYC and parts of CT and MA, where somehow people have lost the ability to stay out of the street so they're lowering speed limits to horse and buggy days.
 
Lindberg said:
They only need to do as well as humans.
Click to expand...
That should be easy. Humans know they shouldn't drive at normal speeds in blizzard conditions, but they do it anyway. Robotic vehicles will not feel the need to drive faster than conditions allow and will not tailgate as much or as closely.
 
Albany Tom said:
… Well, except lately in NYC and parts of CT and MA, where somehow people have lost the ability to stay out of the street ….
Click to expand...
Welcome to San Antonio, where people walking in the street is a cultural past time.

2023, the last year data is available, had 841 pedestrian accidents with 91 fatalities. That’s a normal year here. And no, the city’s not doing much about it.
 
TCABM said:
Welcome to San Antonio, where people walking in the street is a cultural past time.

2023, the last year data is available, had 841 pedestrian accidents with 91 fatalities. That’s a normal year here. And no, the city’s not doing much about it.
Click to expand...
Self correcting problem.
 
SoonerAviator said:
It could if the data were coming in far enough in advance. However, getting enough data to decipher it is going to likely take at least a minute with lag and time for AI to realize there's a pileup. I doubt that's going to help the first dozen semi trucks.
Click to expand...
Nah, it wouldn't take a minute. I wrote a tech demo for Toyota almost 20 years ago to show how cars could share information. This was pre-smart phone. The idea was to have major highways have, basically, wifi and have the Toyotas share information in realtime about the road conditions, including stoppages.

Since then the smart phone revolution has happened and networks are pervasive. This will happen. The only question is whether it will happen on a proprietary or open standards basis.

SoonerAviator said:
Robots can only see a few cars ahead, without telemetry from cars 1/4 mile up the road the robots are still flying blind.
Click to expand...
Robots with lidar can see far better than humans. This is why it was so disappointing to see Elon get rid of lidar on Teslas. It's a real advantage that made the robot vision better than human vision. The cars still have the advantage of looking everywhere all at once. But lidar is game changing.

Bob Noel said:
and, of course, being able to validate the data (and protect it from hacking and corruption)
Click to expand...
If this weren't a solvable problem then commerce wouldn't be possible on the internet.
 
rwellner98 said:
Nah, it wouldn't take a minute. I wrote a tech demo for Toyota almost 20 years ago to show how cars could share information. This was pre-smart phone. The idea was to have major highways have, basically, wifi and have the Toyotas share information in realtime about the road conditions, including stoppages.

Since then the smart phone revolution has happened and networks are pervasive. This will happen. The only question is whether it will happen on a proprietary or open standards basis.


Robots with lidar can see far better than humans. This is why it was so disappointing to see Elon get rid of lidar on Teslas. It's a real advantage that made the robot vision better than human vision. The cars still have the advantage of looking everywhere all at once. But lidar is game changing.


If this weren't a solvable problem then commerce wouldn't be possible on the internet.
Click to expand...
I am talking about capability currently, not with the addition of an extra data protocol. Sure, we have cell phones and it's pretty easy to see traffic slow downs on Google Maps with that aggregate data. However, that data does not occur instantaneously when a wreck happens. It takes a minute for enough data points to signal a wreck or other hang-up. An autonomous truck isn't going to see that data from 1/4 mile back. Once there's an established highway data system that allows data to be mostly-instantaneous it would be fine. I'm not saying it's not possible right now, I'm saying the systems are not in place to allow real-time actions to be taken by autonomous vehicles when a patch of black ice causes a pile up 100yds ahead. If that semi is behind other semis, the cameras/lidar will see nothing but the back of the semi ahead of them (same as a human). If there's a curve in the road, humans are better at seeing up ahead with enough resolution to spot a wreck or brake lights being activated. A camera and lidar probably won't be able to make that distinction at this point because it's not in direct line of sight.
 
SoonerAviator said:
I am talking about capability currently, not with the addition of an extra data protocol. Sure, we have cell phones and it's pretty easy to see traffic slow downs on Google Maps with that aggregate data. However, that data does not occur instantaneously when a wreck happens.
Click to expand...
You said, "is going to". Future tense. Not current capability.

Future tense could be next week with the hardware on the road today. There is no technological problem stopping it. Just a matter of vendors deciding to address the need and/or consumers demanding it. It's not only possible, it's not even hard. As proof of that I submit that *I* already did it. 20 years ago. Without AI. And now we live in a world with pervasive networking, ACN, fall detection in watches, lidar, etc...

I should stop typing. The more I write, the more frustrating it is that this isn't shipping already. The takeaway is that this is an easy problem, not something that will require a lot of lag time or super creative problem solving.
 
Albany Tom said:
Brakes didn't improve, and cars didn't slow down,
Click to expand...
So purely mechanical brakes were just as good as hydraulic brakes? Sorry, that is just wrong.

Albany Tom said:
Well, except lately in NYC and parts of CT and MA, where somehow people have lost the ability to stay out of the street so they're lowering speed limits to horse and buggy days.
Click to expand...

And now they are no longer required to stay out of the streets! NYC has repealed the “no jaywalking” law. Now you can cross the street anywhere you like.

-Slip
 
rwellner98 said:
You said, "is going to". Future tense. Not current capability.

Future tense could be next week with the hardware on the road today. There is no technological problem stopping it. Just a matter of vendors deciding to address the need and/or consumers demanding it. It's not only possible, it's not even hard. As proof of that I submit that *I* already did it. 20 years ago. Without AI. And now we live in a world with pervasive networking, ACN, fall detection in watches, lidar, etc...

I should stop typing. The more I write, the more frustrating it is that this isn't shipping already. The takeaway is that this is an easy problem, not something that will require a lot of lag time or super creative problem solving.
Click to expand...
Unless you mandate that all vehicles/cell phones are to provide that data, in a specified format, and have hardware to store/distribute that data across every mile of highway . . . it's not a current capability or even one that will be feasible for years. Lots of government red tape around it I'm sure. We have the technical capability to do all sorts of stuff, but putting that into actual practice is another animal. This is basically just ADS-B for road vehicles, only there is no ADS-B mandate for every car, truck, or semi to play along. The best that is available now is taking cell phone signals/app data and trying to model traffic patterns out of it (which you mentioned you had already done long ago). Eventually enough new vehicles will adopt that data/telemetry sharing system and the majority of the cars on the road will be able to utilize P2P data which might prevent a lot of wrecks from attempted lane changes and improve traffic congestion as autonomous driving helps smooth out the flow. We're not there yet since there's no mandate and no mass adoption of the tech, not to mention the sheer number of legacy autos on the road that will never be in compliance. However, the cameras/lidar can be used to fill in those gaps in data.
 
omg - I hope we don't use ADS-B as a gold standard for information sharing. Multiple protocols, complete lack of validation/integrity. Please let's learn from ADS-B and do it right.
 
SoonerAviator said:
Unless you mandate that all vehicles/cell phones are to provide that data, in a specified format, and have hardware to store/distribute that data across every mile of highway
Click to expand...
I agree open standards and universal implementation would be better. And over time we'll get there. But this is definitely one of those things where something is better than nothing. If only one in ten vehicles had this capability, the ability to sense would still be nearly instantaneous and benefit everyone. Even the 90% without it. The 10% getting a near instantaneous report of a crash half a mile ahead would begin slowing down and the 90% they are traveling with would do the same as a result.

You mention ADSB. I wonder if that could be the carrot the government uses to encourage adoption someday. ADSB regs limit the areas I can fly my plane. And since I was flying under a Bravo at the time, made it functionally impossible for me to avoid putting it in my plane. I wonder if there will be some future world where the feds say "you can only use urban interstates if you have a smart car" or "you can only use the express lanes if you have a smart car", making it not functionally impossible to get around in a 1990 Grand Caravan, but giving consumers real day to day benefits to having a more capable vehicle in any case.
 
Constantly logged in to Waze, monitoring speed of Wazers miles ahead. The data is there, and small enough size that it would not overcome an iPad size computer.
 
geezer said:
Constantly logged in to Waze, monitoring speed of Wazers miles ahead. The data is there, and small enough size that it would not overcome an iPad size computer.
Click to expand...
Until something happens fairly close to you that hasn't been reported yet. But yes, crowdsourced data is generally pretty good.
 
Cap'n Jack said:
Until something happens fairly close to you that hasn't been reported yet. But yes, crowdsourced data is generally pretty good.
Click to expand...
Yeah, this network stuff won't stop you from hitting the cement block that falls off the truck, that's the problem cameras and lidar solve. But it will stop the guy behind you from hitting you after you hit the cement block that fell off the truck.
 
Lindberg said:
Is a human truck driver?
Click to expand...
The human eyes are able to focus/discern things at a distance a lot better than a typical consumer/commercial camera. An attentive driver is able to pick up a brief glimpse of a brake light or puff of smoke (say from locked up tires or such) that a camera would not likely see. Same with a vehicle going off the road and into the median 15 cars up ahead. I can also imagine a driver picking up on a deer coming out of the woods up ahead (and thus being prepared to take evasive action) that the autonomous vehicle won't notice until it comes into camera view or lidar/radar sensor. It's just those small nuances that are likely limitations at this point unless you have a seriously powerful software engine utilizing multiple long distance, side, and short distance cameras and lidar/radar sensors. Maybe they've got it figured out where it can detect those small nuances in movement outside of the main focal point ahead, but I kind of doubt it's available outside of the military-use case for threat detection.
 
SoonerAviator said:
An attentive driver is able to pick up a brief glimpse of a brake light or puff of smoke (say from locked up tires or such) that a camera would not likely see.
Click to expand...
That "attentive driver" is a big qualification. I can't drive for 3 minutes without seeing 3 cars drifting out of their lane, be it on a suburban neighborhood street or a six-lane freeway. I'm not sure attentive drivers are the norm.

Specifically to your example, though, Lidar and/or mesh-networked-data-exchange could give an automated car more data about the change in speed/direction of the affected car than your attentive driver could deduce from a puff of smoke.
 
SoonerAviator said:
The human eyes are able to focus/discern things at a distance a lot better than a typical consumer/commercial camera.
Click to expand...
No way.
SoonerAviator said:
An attentive driver is able to pick up a brief glimpse of a brake light or puff of smoke (say from locked up tires or such) that a camera would not likely see.
Click to expand...
Even an attentive driver will only see what they're paying attention to. A computer see everything at once.
 
Lindberg said:
Is it 1975?

And if you think humans don't only look for what they're programmed to look for, google inattentional blindness, priming, pareidolia, and all manner of hallucination.
Click to expand...

doesn't matter if it's 1975 or 2075. The computer won't see things it isn't programmed to see. The computer isn't a magic box.
 
rwellner98 said:
Yeah, this network stuff won't stop you from hitting the cement block that falls off the truck, that's the problem cameras and lidar solve. But it will stop the guy behind you from hitting you after you hit the cement block that fell off the truck.
Click to expand...
I saw almost that scenario yesterday- a large panel of plastic fell off a truck hauling junk, it almost fell on a car. I don't think a properly programmed robot could have avoided it because of momentum and other traffic. I was thinking if something a few hundred to a thousand feet ahead but, as you note, camera and lidar sensors should notice the traffic stopping.
 
Lindberg said:
No way.

Even an attentive driver will only see what they're paying attention to. A computer see everything at once.
Click to expand...
No way what? Have you ever seen the video output from an automotive video camera feed? They aren't good about focus/resolution in multiple focal planes. Even your cell phone has trouble knowing where to auto-focus when taking a picture without a reference point. The human eye/brain is generally better about that, especially given the wide field of view. Now, for autonomous uses there will need to be multiple cameras all with different intended focal lengths, both forward/side/rear-facing as well as the engine to compile and analyze those multiple data streams real-time in order to reliably detect what a human eye/brain can do. I'm not saying it can't be done, but it's not like you're doing it with what's on a Tesla Model S.
 
Cap'n Jack said:
I saw almost that scenario yesterday- a large panel of plastic fell off a truck hauling junk, it almost fell on a car. I don't think a properly programmed robot could have avoided it because of momentum and other traffic. I was thinking if something a few hundred to a thousand feet ahead but, as you note, camera and lidar sensors should notice the traffic stopping.
Click to expand...
I was talking about fact that trafficMesh™ would allow the trailing car to be alerted by the forward car. So the forward car panic brakes (either the human of the FSD). It doesn't manage to avoid collision with the panel and further brakes to a complete stop. The trailing car gets all this information in realtime and has more data on which to make better judgements. But, yes, camera and lidar are still useful.
 
SoonerAviator said:
No way what? Have you ever seen the video output from an automotive video camera feed? They aren't good about focus/resolution in multiple focal planes. Even your cell phone has trouble knowing where to auto-focus when taking a picture without a reference point. The human eye/brain is generally better about that, especially given the wide field of view. Now, for autonomous uses there will need to be multiple cameras all with different intended focal lengths, both forward/side/rear-facing as well as the engine to compile and analyze those multiple data streams real-time in order to reliably detect what a human eye/brain can do. I'm not saying it can't be done, but it's not like you're doing it with what's on a Tesla Model S.
Click to expand...
This seems like better oversight than the human eye can muster to me. You saying this isn't possible?

1746558951791.png
 
SoonerAviator said:
No way what? Have you ever seen the video output from an automotive video camera feed? They aren't good about focus/resolution in multiple focal planes. Even your cell phone has trouble knowing where to auto-focus when taking a picture without a reference point. The human eye/brain is generally better about that, especially given the wide field of view. Now, for autonomous uses there will need to be multiple cameras all with different intended focal lengths, both forward/side/rear-facing as well as the engine to compile and analyze those multiple data streams real-time in order to reliably detect what a human eye/brain can do. I'm not saying it can't be done, but it's not like you're doing it with what's on a Tesla Model S.
Click to expand...
Human eyes are not really that great at vision and have many limiting factors, including age and structural anomalies. Most of what we think we see is constructed in the brain. And the brain is "preprogrammed" with certain priorities and heuristics that don't always reflect the "real world" in front of us.

1746560132435.png
 
Lindberg said:
They only need to do as well as humans.
Click to expand...
No. To be accepted they must do much better. We humans do have a deep seated Frankenstein complex AND we feel much more comfortable in control (however wrong we may be) of our fate than trusting a machine.
 
Bob Noel said:
The computer won't see things it isn't programmed to see.
Click to expand...
The better systems aren't programmed anymore, they are trained. Neural nets watch countless hours of video and train themselves on how to drive, identify threats, and avoid them.
 
Larry in TN said:
The better systems aren't programmed anymore, they are trained. Neural nets watch countless hours of video and train themselves on how to drive, identify threats, and avoid them.
Click to expand...

and how to anyone ensure that a safety-critical system is adequately "trained"?
 
You must log in or register to reply here.
Back
Top