Autonomous Truck Success

[Overmod]

Let's see how clear I can make this.

Longitudinal position on the track isn't a particular issue.  GPS, even single-precision, will give you 'good enough' speed and location information for typical block separation; HA-NDGPS with beacons will give you sufficient information for effective CBTC if you have good communications between the trains and wayside equipment. 

The issue is distinguishing trains on different track centers.

 

[Euclid]

Operating the train in relation to its exact location on the track is something that requires constant judgement by a human engineer in relation to various guide markers.  So this location awareness has to be accomplished with autonomous operation, and done so without human judgement. 

 

I looked at a video yesterday that showed passenger trains in Britain that had the entire track hard wired with a system that constantly read the exact location of the train on the line of track at incremental points spaced maybe 100 feet apart.  As I recall, it does not use GPS to track location, but it does count wheel rotations.  

 

I will have to see if I can find this link again, as I did not save it. Apparently the fixed electronic marker readings count the whole distance between their placement locations, and the wheel rotations count the distance between the markers in a range finer increments.  Generally, this system turns the track into an electronic tape measure, or a perfectly accurate odometer that cannot accumulate error.

 

It would be interesting to know how Rio Tinto accomplishes this goal with their heavy haul autonomous ore trains.     

[tree68]

I suspect the ultimate solution would be "all of the above."

GPS would provide the position of the front and rear of the train to a certain level of accuracy.  Radar verifies the speed (and would be compared to the GPS), and trackside or center of track transponders would provide site specific information, verifying location and identifying the specific track, among other items.  

Transponders already are used to track individual cars - look for the little grey device on the sides of the cars.  It would not be out of the question to use the information from track transponders in combination with the GPS to confirm specifics and to provide information to on-board PTC systems.  

Anyone who has an E-Z-Pay transponder (or other similar toll payment system) knows this can work, even at highway speeds.

Many reefers (and some other cars) are already being tracked via satellite.  It may even be possible to parse some of the information provided by those cars into such a system.

[Erik_Mag]

Hmmm, getting an idea...

The reason for using GPS is to be able to know within a few feet where he train is along the line, i.e. how far has the train traveled? This is important to know wehere the train is in respect to fouling points. Due to wheel slippage and wear, counting the number of turns of an axle cannot be relied on to give an exact number for distance traveled. One thought on improving distance measurements is using radar to measure speed and thus distance would drived from integrating the measured speed. To improve accuracy, one could place a series of radar reflectors where the relative spacing between reflectors would be a code for position, e.g. milepost and track number. These could be placed in areas where GPS reception is impossible.

[Euclid]

I would think that modern autonomous train operation would have the train and track in constant communication with each other.  And routes would simply be lined up by controlling switches, train speeds, stopping, starting, and the effects of grades and curves.  So what would be the need for GPS? 

 

The train will be following a program, and even though there is no on-board crew, there will be plenty of humans working as dispatchers to intervene and make changes in the programmed route as the plan changes.  I don’t see it as requiring AI to think like a human and reason out how to drive the train through relationships with other trains.  For the time being, humans will still be making those decisions. 

 

Eventually, it may turn into a completely automatic rail transportation vending machine, but a lot of the railroad infrastructure will have to be revised in order to make way for those most dramatic changes.  

[Overmod]

If railroads were easy to automate trouble-free we'd have started with the Times Square Shuttle lo! these many years ago, and NAJPTC wouldn't have been such a colossal laugher.

One major issue is precisely what Erik mentioned: the 'one degree of lateral freedom' often leading to very close track-to-track centers and some uncertainty with fouling points of 'listing' or damaged equipment.  As mudchicken and others will confirm, typical GPS location has difficulty distinguishing what track a train is on, which is a reason the QNS&L-style radio transponder separation has trouble on railroads with multiple tracks in service.  It's much the same issue as using drones for navigation aids; it makes a good show when shiny and new, but at rainy zero-dark-thirty under heavy cloud, with a few years of typical railroad negligence-posing-as-maintenance on the clock, "things can start to happen" that the AI won't correctly anticipate, or deal with.

[Erik_Mag]

charlie hebdo

Not sure about V-1s landing without damage.

That kind of sums up some of my misgivings of the pilotless cargo aircraft... The folks launching the V-1s didn't care if less than 100% made it to the target as long as a significant fraction did.

Aircraft, OTOH, operating in three degrees of freedom space (up/down, left/right, faster/slower) are much less likely to collide than ships operating two degrees of freedom space  (port/starboard, faster/slower), an even less likely than cars/trucks with two degrees of freedom (left/right, faster/slower), where one of those degrees (left/right by width of road/lane) is tightly constrained. Trains have one degree of freedom and this require much stricter rules of operation especially considering that stopping distances can be further than line of sight. The one degree of freedom for trains makes automatic train control a much easier proposition where the ROW can be protected.

One objection for pilotless aircraft and it's much stronger for autonomous road vehicles is impairment to the sensors. The 77GHz radars used on vehicles are rendered useless when covered with slush or mud and LIDARs are even more sensitive to slush/dirt. Yet another is that a lot more work needs to be done on the pattern recocognotion routines. An example of the latter is that a consientious human driver will recoginize that a bunch of very young children requires a lot more attention than a similar sized bunch of bushes.

OM: I was living in San Diego when that happened and my oldest sister was 6 blocks away from where the planes hit the ground. Number one problem was that there were two aircraft in roughly the same airspace that were being directed by two diferent facilities, ATC and the Lindergh Field tower. Also note that in-flight collisions are more likely around airports.

Looks like pilotless aircraft should have some sort of balloon detection, thouh a more serious issue for delivery drones is detecting wire - power lines, guy wires, antennas.

Speaking of ATC - the rules with respect to airways defined by LF/MF ranges or VORs seem to have been derived from RR dispatching practice.

[BaltACD]

Euclid

 

charlie hebdo

Railroad mainlines would be the most logical and easiest to safely automate if it weren't for labor contracts. 

It is true that railroads would be the easiest form or transportation to automate, but for the resistance of Labor.  However, management views automation as a way to overcome Labor’s resistance to reducing crew size. That is why they like automation.  Automation just sidesteps that issue by eliminating crews. 

 

Railroad management abhors some improvements such as ECP brakes, but they seem to be very fond of autonomous freight trains.  Management loves autonomous freight trains as much as much as they love monster trains, and for the same reason. 

 

Automation will be easy to expedite because manufacturers and vendors of automation see a golden opportunity in marketing these sorts of revolutions to such a big customer as the U.S. railroad system, which is all standardized with the same universal needs.  This fantastic marketing push will sweep the industry off its feet with this marketing revolution.  It will be the biggest thing since dieselization.

 

But it won’t all be easy.  On one hand, railroads don’t suffer the problem of trains needing to sense and react to endless obstacles on their right of way in the same manner that cars and trucks have to on public roads do.  However it will be very challenging for autonomous freight trains to navigate through all of operational contingencies just by following a program of operation the way Rio Tinto trains do with the consistency and predictability of their operations. 

 

So autonomous trains are likely to require a lot of changes to simplify their operation where it is most congested if they are going to follow a program of automatic operation. Otherwise, if they are just programed, the program will require frequent modification.  This will tend to require human intervention, so it will be like the trains are not actually autonomous,  but more like remote control.

Better start checking the quality of the hemp rope you all have been smoking.

[Euclid]

charlie hebdo

Railroad mainlines would be the most logical and easiest to safely automate if it weren't for labor contracts.

 

It is true that railroads would be the easiest form or transportation to automate, but for the resistance of Labor.  However, management views automation as a way to overcome Labor’s resistance to reducing crew size. That is why they like automation.  Automation just sidesteps that issue by eliminating crews. 

 

Railroad management abhors some improvements such as ECP brakes, but they seem to be very fond of autonomous freight trains.  Management loves autonomous freight trains as much as much as they love monster trains, and for the same reason. 

 

Automation will be easy to expedite because manufacturers and vendors of automation see a golden opportunity in marketing these sorts of revolutions to such a big customer as the U.S. railroad system, which is all standardized with the same universal needs.  This fantastic marketing push will sweep the industry off its feet with this marketing revolution.  It will be the biggest thing since dieselization.

 

But it won’t all be easy.  On one hand, railroads don’t suffer the problem of trains needing to sense and react to endless obstacles on their right of way in the same manner that cars and trucks have to on public roads do.  However it will be very challenging for autonomous freight trains to navigate through all of operational contingencies just by following a program of operation the way Rio Tinto trains do with the consistency and predictability of their operations. 

 

So autonomous trains are likely to require a lot of changes to simplify their operation where it is most congested if they are going to follow a program of automatic operation. Otherwise, if they are just programed, the program will require frequent modification.  This will tend to require human intervention, so it will be like the trains are not actually autonomous,  but more like remote control.

[tree68]

charlie hebdo

Not sure about V-1s landing without damage.

True, but they landed without a pilot, if you can consider falling out of the sky "landing..."

[charlie hebdo]

Railroad mainlines would be the most logical and easiest to safely automate if it weren't for labor contracts.

[charlie hebdo]

Erik_Mag

the V-1 and many airliners are capable of taking off, flying and landing without the pilots needing to touch their controls.

Not sure about V-1s landing without damage.

[Overmod]

Erik_Mag

I also wonder how many are familiar with the events of Sep 25, 1978?

As I said, mandatory longer-range intelligent TCAS, combined with autonomous oversight of all traffic.

(I made the mistake of listening to the PSA flight audio.  If you haven't... don't.)

[Erik_Mag]

Based on their headquarters, I wonder how many of the people there are former GA employees? I also wonder how many are familiar with the events of Sep 25, 1978?

FWIW, autonomous aircraft go back a long way, with the V-1 and many airliners are capable of taking off, flying and landing without the pilots needing to touch their controls. Pilots are still required for taking over when problems occur (flying into a flock of birds) and avoiding other traffic and obstructions.

[Overmod]

charlie hebdo

Small pilotless cargo planes to be based in Dallas...

Not that small; 3.8-ton nominal payload, three engines.  The company says they have a 460+ backlog with 6.8B committed.

Here's the contact information if interested:

https://natilus.co/contact/

I for one would like to hear Backshop's brother's take on this, in detail.

[Overmod]

Mr. Klepper, it's not as bad as that makes it sound.

Many of the 'true' advantages of level 4 autonomy that are at all cost-effective are currently achievable -- platooned operation terminal-to-terminal over interstates; automated parking assistance at docks; yard unloading and loading on flats with tractors fitting under trailer noses.  (See Shadow's owner's comments from early last year; those activities are well within reasonable programming and don't suffer from extreme or emergent traffic or weather difficulties.)

And you will see aviation use of autonomous operation long before you see it on roads -- or on railroads, for that matter.  One best use of "AI" is in predictive extension of TCAS out to greater range, with following of a large number of relatively labile tracks.  Integrate steerable BRS foils with GPS/GIS for 'optimized touchdown' in emergent conditions.  It has been my opinion for some time that this, not 'autonomous buses' or widespread dependence on level 4 "owned vehicles", is the most practical solution for regional transport that lacks the pure volume for HrSR.

[charlie hebdo]

Small pilotless cargo planes to be based in Dallas 

https://www.fox7austin.com/news/ups-fedex-partner-airline-agrees-to-buy-20-pilotless-cargo-planes-in-134m-deal

 

[daveklepper]

Comments appreciated.

Apologies for duplications.  In this case I seem to lack the necessary skill to remove duplications without some needed text removed as well.

t

[daveklepper]

Related discussion (MIT Senior-House graduates forum

John r	o)
tu

 

 

 

Jon, Azamat, and everybody else,

 .

 

Following are some excerpts from the article by GM.

 

John

 

--------------------------------------

 

Turns out Tesla staged their famous 2016 driverless car demo, with the
famous tagline “The person in the driver’s seat is only there for
legal reasons.  He is not doing anything.  The car is driving
itself.”.  What they showed, it seems, was aspirational, not real, not
a single, unedited run taken by a single car.  Roughly $100 billion in
investment went in, partly on the strength of that demo and the
excitement it generated, in part on Musk’s say so; in the subsequent
six years, no car has yet achieved what Elon Musk promised would soon
arrive.

 

OpenAI turns out to have been using sweatshops behind the scenes.  You
might think that the ChatGPT is just a regular old massive neural
network that soaks up a massive amount of training data from the web,
but you’d only be partly correct.  There is in fact a massive,
massively trained large model behind the scenes, but it’s accompanied
by a massive amount of human labor, built to filter our bad stuff.  A
bunch of that work was done by poorly paid labor in Kenya, paid less
than $2/hour to evaluate (e.g.) graphic descriptions of sexual
situations involving children and animals that I prefer not to
describe in detail.

 

Riley Goodside, one of the people who best knows what large language
models can and can’t do, put Claude, the latest large model to the
test; the focus on this model is on alignment.  You can read his
detailed comparison for yourself, but one of the things that popped
out to me is that the system still quickly lands in the land of
hallucination that has so haunted ChatGPT.

 

CNET became the first casualty of the recently fashionable tendency to
put too much faith in ChatGPT.  Without making a big deal of it, they
started posting ChatGPT-written stories.  Mistakes were made.  A lot
of them.  Oops.

 

The coup de grace?  The musician Nick Cave got a listen to ChatGPT
riffs on his music.  If you can believe it, he was even more scathing
than I am.

_________________________________________________

 

 

 

Nona	7:05 AM (5 hours ago)
to [e-mail], [e-mail]

john, good read, thanks

so lucky to have you all share with us these gems informed by your

expertise.

John F Sowa via mitalumprod.onmicrosoft.com	6:52 AM (5 hours ago)
to Ontolog, [e-mail]

Jon, Azamat, and everybody else,

 

Jon, Thanks for that reference:  

\https://open.substack.com/pub/garymarcus/p/24-seriously-embarrassing-hours-for

 

That article by Gary Marcus digs into the sordid underbelly of some of the

most hyped AI technology. 

I have been working on AI-related projects for over 40 years. 

There has been a solid amount of good R & D. 

But there has also been a huge amount of hyping impressive,

but half-baked AI projects. 

That hype has created a frustrating series of boom & bust funding

for AI projects for the past 60 years.

 

Azamat, I suggest that you read more of Gary M's bubble-piercing

reports instead of that nonsense about

a singularity bubble that will magically pop up in 2030.

 

Following are some excerpts from the article by GM.

 

John

 

--------------------------------------

 

Turns out Tesla staged their famous 2016 driverless car demo, with the
famous tagline “The person in the driver’s seat is only there for
legal reasons.  He is not doing anything.  The car is driving
itself.”.  What they showed, it seems, was aspirational, not real, not
a single, unedited run taken by a single car.  Roughly $100 billion in
investment went in, partly on the strength of that demo and the
excitement it generated, in part on Musk’s say so; in the subsequent
six years, no car has yet achieved what Elon Musk promised would soon
arrive.

 

OpenAI turns out to have been using sweatshops behind the scenes.  You
might think that the ChatGPT is just a regular old massive neural
network that soaks up a massive amount of training data from the web,
but you’d only be partly correct.  There is in fact a massive,
massively trained large model behind the scenes, but it’s accompanied
by a massive amount of human labor, built to filter our bad stuff.  A
bunch of that work was done by poorly paid labor in Kenya, paid less
than $2/hour to evaluate (e.g.) graphic descriptions of sexual
situations involving children and animals that I prefer not to
describe in detail.

 

Riley Goodside, one of the people who best knows what large language
models can and can’t do, put Claude, the latest large model to the
test; the focus on this model is on alignment.  You can read his
detailed comparison for yourself, but one of the things that popped
out to me is that the system still quickly lands in the land of
hallucination that has so haunted ChatGPT.

 

CNET became the first casualty of the recently fashionable tendency to
put too much faith in ChatGPT.  Without making a big deal of it, they
started posting ChatGPT-written stories.  Mistakes were made.  A lot
of them.  Oops.

 

The coup de grace?  The musician Nick Cave got a listen to ChatGPT
riffs on his music.  If you can believe it, he was even more scathing
than I am.

__________________________________________________


 

 

 

 

 

Nona	o)

john, good read, thanks

 

so lucky to have you all share with us these gems informed by your expertise

 

[blue streak 1]

tree68

Absolutely!!!

 

 

blue streak 1

Visual clues help in these conditions both for engineers and pilots. 

 

 

Don't forget good old seat of the pants...

 

[tree68]

blue streak 1

Visual clues help in these conditions both for engineers and pilots. 

Don't forget good old seat of the pants...

[blue streak 1]

Engineers have many variables to deal with.  Every trip is different.  Not only must he understand the dynamics of the present train but has to be aware of outside happenings.  I wonder if the airplane hit Would the robot even know it happened.  Or what if a small piece of iron on track might derail the next train?  Or his train goes over a cracked rail and passes it but reports same., etc,etc,etc

A pilot uses the auto pilot but has to constantly update the autopilot. Too many changes happen when enroute or sudden weather or even a vvolano? No auto pilot has ever been able to handle gusty weather especially in cross wind conditions. 

Visual clues help in these conditions both for engineers and pilots.  Robots do not see that well.

[MJ4562]

Tee hee!  Maybe in 30+ years or so. Trains and airliners will become automated long before public road going vehicles.  Too many variables involved, too much liability.  

[CSSHEGEWISCH]

tree68

 

 

NittanyLion

I'm referring to pilot error induced by sensory illusions. 

 

 

Case in point (it's been surmised)  - JFK Jr.

 

I won't comment about JFK Jr but that is what killed Buddy Holly, Ritchie Valens and the Big Bopper.  The pilot was barely qualified to fly in CAVU but hoped to make a buck flying in conditions that would challenge experienced pilots.

[Backshop]

tree68

 

 

NittanyLion

I'm referring to pilot error induced by sensory illusions. 

 

 

Case in point (it's been surmised)  - JFK Jr.

 

That was the airplane equivelant of a 16 year old getting a Ferrari for his birthday. He thought that because the plane was capable of something, that he was too.  Nope, it doesn't work that way.

[tree68]

NittanyLion

I'm referring to pilot error induced by sensory illusions. 

Case in point (it's been surmised)  - JFK Jr.

[NittanyLion]

Backshop

 

 

NittanyLion

   Turned out the robot is way better than humans at the flying game too.

 

 

 

Wrong.  Autopilots take the strain off the pilots during cruise.  When something goes wrong, the humans do a better job. The pilots aren't just sitting up there twiddling their thumbs when autopilot is on, either.  They are constantly monitoring radio communications, fuel burn, alternate airfields for emergencies, weather reports, etc., etc.

 

 

I'm fully aware of all of that.  I'm referring to pilot error induced by sensory illusions.  Humans aren't built to fly.  I (as a passenger) suffer greatly from the leans.  A gryoscope doesn't.

[rdamon]

Backshop

 

 

NittanyLion

   Turned out the robot is way better than humans at the flying game too.

 

 

 

Wrong.  Autopilots take the strain off the pilots during cruise.  When something goes wrong, the humans do a better job. The pilots aren't just sitting up there twiddling their thumbs when autopilot is on, either.  They are constantly monitoring radio communications, fuel burn, alternate airfields for emergencies, weather reports, etc., etc.

 

 

But when they can't  ....

 

https://www.youtube.com/watch?v=IyYxbiZ1FCQ

[Backshop]

NittanyLion

   Turned out the robot is way better than humans at the flying game too.

Wrong.  Autopilots take the strain off the pilots during cruise.  When something goes wrong, the humans do a better job. The pilots aren't just sitting up there twiddling their thumbs when autopilot is on, either.  They are constantly monitoring radio communications, fuel burn, alternate airfields for emergencies, weather reports, etc., etc.

[SD60MAC9500]

Hyundai's betting big on CAV's. However the "drone" is some years off. Some of you may not know. Hyundai builds domestic 53' trailers for our market under the label Hyundai Translead. They're manufactured in Tijuana, Baja California. One application I can see happening for AV's are supplier to plant shorthaul. In my area we have multiple tier 1-3 automotive suppliers that are only on average 15 miles from final assembly plants. There's definitely a market there for this to be the first market AV's penetrate.