Requesting your input--is this idea good or crazy?

The Big Thought behind your idea ist dont force computer to "view" like humans with cameras and object recognition ai is a big point. The part with guidewires ist are they "smart" do they are connected to a masterbrain or are the dumb steel cables? In my view the self driving car needs to get a bigger picture like a maps in a computer game. I thought about a car to car mesh network in a circle of 100m- 1km in different of the speed. So each car gets the speed, heading and lane of each other car in this 1km circle with the dopplear effect the direction the signal came from and some math the car could get a pretty good picture of its surrounding even in a snowstorm as the magnetic waves dont bother. So at 120km/h you car knows each car in a kilometer radius and if a car gets nearer it slows down faster as a human can see it. in Germany we have sticks <a href="https://de.wikipedia.org/wiki/Leitpfosten" rel="nofollow noopener" target="_blank">https://de.wikipedia.org/wiki/Leitpfosten</a> on each side of the motorway why not put radio reflectros on them an tell the car to stay in a certain angle and distance of them which would be in the middle of the street and for road work they have portable sticks which would direct the car to the other lane. With only a dumb steel cable the only way to determine if a car is in front of a car would still be radar or cameras ways where the ai has to guess. With a mesh car to car network the car gets the overall picture and if the computer gets the bigger digital picture it beats the human in lenghs in accident, traffic jam etc- .

D-Snake91

This is a terrible idea, you should get some sleep. Not to say It’s not an idea interesting, however they already developed and tested this technology in the early 2000’s and before, by embedding magnets in the road. The challenge is not following a suitable path (as that’s fairly easy) rather it’s reacting and adapting to unplanned circumstances.

D Money

The idea is good but this would require trusting any municipalities you drive in to have safe paths, and some can't even handle potholes or reaching minimal safety level for pottable water. It's worth exploring and figuring those limitations out.

I like this idea, but it will never come into fruition. It will work great for highways, but for streets it will have big problems. What to do if there is a dump truck blocking half of the lane? Is there an accident ahead, and traffic is being rerouted manually? "AI" self driving on highways is already pretty good. But in any case where things differ from the normal, your system would have no idea what to do. The system would work if it is just a driver assist system, and falls back on the driver if it fails, but no government would want to put in that kind of funding into that system, especially when's Waymo/Tesla/Volvo/Hyundai already handle that use case pretty well. In public transport, systems like the one you're describing could have massive implications. There are a few bus rapid transit systems out there, and some of those systems could probably phase out drivers without much hassle, since they have a separated path, and pedestrians aren't allowed in that path. But they might very well still have a bus attendant that could take action (drive manually very slow) if something happens, kind of like the DLR in London. The idea itself is great, but it has a lot of limitations that means we will never see a widespread general-purpose version of this system

I think you could make a good video with this. Highways are easy, and Tesla (and another company or two) has made a car that can do it on limited access roadways. I think Tesla's system leads to complacency on the part of the driver, and will probably lead to accidents. With Disney parade vehicles, they use pucks below the street, and all the vehicles have drivers, but the pucks are used to know when the lead vehicle reaches certain points on the route, so they can start the audio program in the next section at the right time. They do go at faster speeds in the event of rain, so that the performers don't have to be in the rain getting costumes wet for as long. I like you better when you're well rested though. Please always have good audio on your videos. I'll forgive so much as long as there is good audio. I also noticed that the focus was wandering between the background and you.

William Gray

Also, one of the key things missed in most discussions of autonomous vehicle technologies is that they don't have to "safe", they have to be "safer". Many people assume it's going to impossible to roll out autonomous vehicles until they're demonstrably accident/risk/etc. free for a lengthy period of time, but they really just need to be less dangerous than human beings are right now. A 1-in-a-million chance your car will drive you off a cliff is way better than a 1-in-ten-thousand chance someone who's intoxicated/texting/etc. drifts out of the oncoming lane and kills you.

Even on pedestrian-free interstates, you're going to want some really solid sensing system to handle road debris, stuck vehicles, wildlife, etc.

I can understand that tiredness for sure. I usually wake for work at 6:00-something for 7:30 work time and lately I've randomly been awake HOURS earlier a couple times this week for no apparent reason. But I gotta stay up and make it through the day and end up running on fumes, much to the dismay of my cognitive ability, all dang day. I like the idea! I know you've magically scooped so much information about obscure (in depth) topics into each video.. and it sounds like you're fully capable of making this subject/video a great, congruent release.

Alec Jahn

My vote is that they are going the right way with autonomous vehicles. The problem with autonomous vehicles isn't the vehicle itself, but the fact that it has to share the road with other non-autonomous vehicles. Even taking the pedestrians, bicycles, and other slow moving vehicles out of the equation, for example let's say highway driving only, there are still plenty of scenarios where more intelligence is desired. The best example I can think of is what should happen in the case of somebody cutting off the autonomous vehicle on the highway? If we have a "dumber" AI, would it simply try to follow the guide wires, in which case, there would probably be a collision… It might be the other driver's fault, but a collision is a collision, and even people not at fault (i.e. passengers in the autonomous vehicle) could still get hurt... A smarter autonomous vehicle should try to swerve out of the way, but to do that safely, it needs to have very good situational awareness, the kind you get if you have a very good AI with its own sensors (light, sound, radar/LIDAR)... If you get to that point where the vehicle can do decent defensive driving, then it should be able to perform very well for normal driving without any additional external help (i.e. guide wires). Accidents are often caused by one of the parties not following the rules. It's fairly easy to design a system if the rules are known (and known not to be broken), such as in the case of the airport shuttle or automated train systems, but out on the roads in the real world, we know that this isn't the case. A lot of road accidents are caused because one of the parties involved was not following the rules... I think having more intelligence behind the wheel is the best way to go to handle these unexpected situations... The "cost" of adding more sensors and computing power is very low compared to the "cost" of the alternative (i.e. more accidents/injuries). FYI, do a search on youtube for "tesla saved my life". One of them is a guy showing dashcam footage of how the autopilot automatically took evasive measure to avoid a collision. Pretty impressive... As for your idea of using guide wires or RFID puck, I'd say bring it on... In addition to helping autonomous vehicles, that technology could also be used in non-autonomous vehicles to provide another input to the driver... I would see this as especially useful in very bad visibility conditions. Anyway, I just wanted to add that I really enjoy your videos... I can tell that you put a lot of effort into each of your videos! (well, most of them anyway :-) ). Keep up the good work!

BrianL

I think we are 10+ years away from making an autonomous vehicles a reality on a large scale, except for trains, in trains we are probably under 5 years away. Public buses will have similar issues to private cars. You need this processing for kids, the blind, the smartphone/music addict, etc. Disney technology works in a controlled environment. Public roads are not a controlled environment. In the next 10+ years a lot can change. Your videos are fantastic and well researched so I think putting out an opinion peice on something you don't know much about is a bad idea, but that's just my opinion.

Also starting with lifts (elevators) ties in super well to your recent up / down indicator bells video, and might draw in more viewers to think about them as autonomous vehicles, and thus follow you onto the rest of the idea. Ewen

Ewen McNeill

Lifts (elevators) are also trusted, autonomous vehicles, isolated and on tracks. I think people are trying to do a “Big AI” equivalent with autonomous vehicles, and replace humans, because “wouldn’t it be cool” and perhaps more importantly, it solves the “chicken and egg” infrastructure issues without needing coordination. But the result is “boil the ocean” hard, or at least “fly to the moon” hard. It sounds to me like your talking about *busways^ for autonomous vehicles — defined route, no other types of vehicles, no same level cross traffic. And I think that’s a good point that’s being overlooked. The “autonomous vehicles are all around you, everyday” is also a good point. Especially combined with looking at the history of lifts/elevators, which *were* routinely staffed, for safety (and scheduling) in their early years. Possibly make the “autonomous vehicles are all around you” video first, stand alone. Then do a follow up video suggesting the bus ways / guide magnetic route thing as a second part, when people have seen the context? At a guess most people have been on one of those airport busways at least once, and everyone will have been on lifts (elevators) a lot without thinking about it. Ewen

Ewen McNeill

I can see how this will be very useful in public transportation. For private cars--well--it really depends on the manufacturers' goal. I feel that they want to achieve the reality of self-driving cars with the "smart" AI thing going on. It's a feeling, so take it as a grain of salt.

A guide wire autonomous lane on interstates would probably be the most feasible "dumb" autonomous car option. Perhaps instead of investing in millions of RFID chips or whatever, some kind of machine could just "paint" a line along the lane, with some kind of cheap material that could function as the guide wire but still be resilient to snow/weathering/etc. Maybe something the car itself can wirelessly "power" the wire to light it up for the upcoming however many meters. However I would imagine it would need some kind of predefined entry and exit stations, with logic built into the autonomous system that would allow you to pick which station you want the car to drive itself to. It probably couldn't be a lane you just go in and out of as you please, because I would imagine falling asleep or perhaps passing out through some kind of medical emergency at the time of exiting the lane would be a problem. If every car exiting would exit at some kind of station, where attendants could make sure there is a responsive driver able to take up manual control, whom they could then "buzz through", then that could work. Would also need to figure out what to do if, like, debris falls off a vehicle in the lane. Would probably need at least two lanes so traffic can go around in such a scenario, which could also be used as a shoulder in case a vehicle runs into mechanical problems. I'm sure this is a solved problem for trains. If a central system can guide all the vehicles, you can do whatever train infrastructure does.

I am amazed that autonomous vehicles exist. It seems the real world is too unpredictable. One example, what happens when a vehicle hasn't been properly maintained by it's owner and suffers random unforseen glitches?

Wim

Guide wires could be used in some places for extra guidance and reliability, but it would be infeasible to put them in every road in America. I'm not convinced that they would make a good enough subject to make a video about.

Wrenly Bewick

I think it's interesting. I was pretty skeptical until you mentioned interstates. I still think the biggest problem with the guide wires is that you have to worry about interacting with other cars. But, for large stretches of interstate (especially ones used heavily for shipping) I could actually imagine some states putting in barriers to carve out one lane (or more) for these kinds of guidewire vehicles. Shipping companies would probably be the first to use such a setup due to the labor savings, but as someone who's taken cross-country trips before the idea of being able to drive into one of these lanes and let the car take over for a few-hundred miles is pretty appealing. As you mentioned with the o'hare train: people trust these for some pretty specific reasons. The fact that they are separated from the rest of traffic by barriers and that they have a guide rail that they can't deviate from would mean: 1. other users of the interstate won't feel uncomfortable because the robots are in their own lane 2. users of the robot cars can feel safe knowing their car is simply following a wire, and possibly communicating with nearby cars to ensure proper spacing I definitely think such a video is interesting if you look into it some more. That said, there has to be some prior art here. I'm noticing from the comments that I'm not the only one to notice this "separated lane for robots" solution, and it might be worth looking into any kind of history around that.

Not long ago there was a discussion about autonomous vehicles in general and if a guide-wire system or a more intelligent system could be more easily "weaponized". This discussion took place in the background of a terror attack where attackers used transporters or trucks steered directly into a crowd. It was believed that guide-wire systems could be more easily modified to act as remote controlled attack vehicles.

Arragon

I agree 100% and I've been thinking the same thing. I envision a guide wire system on major highways (in only one lane) between depots outside major cities. For the use of driverless trucks. Driver bring trailers to the depot, auto-trucks are attached to the trailer and take the trailer to the depot of it's destination city. Where a human driver takes it the last few miles. In addition these trucks could have remote control via cellular network from a central control center. So if the a truck has something strange happen, or the truck stops and pulls over because of an issue, central control can deal with it, including driving the truck remotely (at a very reduced speed on the shoulder) to get it off the road. Central control could also take care of stopping trucks in bad weather, having them all pull off and wait.

James Sutherland

Self driving cars will have to handle pedestrians, other traffic, signs etc. so they cannot be dumb. A dumb car would require special infrastructure like dedicated fenced lanes and would still need to have transmitted info like speed limit and other information that is currently on signs. It would not be simple. The "dumb" selv driving car would also only be able to drive where there is support. Sorry: You cannot visit Uncle Tom. When self driving cars becomes popular, you could imagine special lanes on "ultra high ways" where they could go 180 or 200 km/h safely - with help of some extra guidance. A guidance system would also improve their ability to drive in snow. Because currently they cannot drive when they cannot see the lines on the road. Adding such a help would be obvious for self driving buses, as you cannot just drive them when the system cannot see.

Simon Mikkelsen

Jason would be an excellent resource here. Take him up on his demo offer if you can. And make a video of it for us!

I have extensive experience with autonomous vehicles. I've owned every generation of Tesla vehicle with "autopilot". I've done extensive modifications to those systems over the years as well (google me, wk057, for those details). Overall, I definitely agree. Autonomous vehicles would definitely be best done with some sort of integration with the roads they'll operate on. Adoption for anything like this is the biggest issue. Believe it or not, finding a path is actually pretty much the easiest thing done by autonomous vehicles. The issue are generally humans. If there were only autonomous vehicles on existing roads, they would work great in the vast majority of situations. In my personal vehicle with my augments I can drive hundreds of highway miles without touching a single control. So, your guide wire idea probably wouldn't actually solve much, unfortunately. Most of the issues remaining are basically the same issues that would face a vehicle following a wire. For closed access roads, like interstates I think we're basically already there. I'd be happy to demo my own vehicle for you if that's something you'd like to work out. My opinion would be that you need to do some more research into the current state of affairs in the autonomous vehicle space before making such suggestions. Happy to chat about it. Feel free to reach out!

Jason Hughes

Personally, the biggest problem is that it's impossible to stay in any lane (or even set of lanes) because [as Ashley Grant has already said, I now see] there will always be unexpected hazards; even if all cars were required to be self-driving there will be outliers (pedestrians, malfunctioning vehicles, sinkholes, etc.) that would throw a simple system into chaos. And even though it would happen almost exclusively in TV crime shows, there will be people who try to sabotage the systems (for terrorism or lulz). That being said, I do think you're right that a more-gradual approach to what kinds of driving cars can do may be a better way to develop the technology. (Yes, there are pros and cons to that kind of thinking. Hitting a dead end after decads and dollars would be a huge bummer.) We already have HOV/carpool lanes, so creating a guide system like you're thinking in those first may be worthwhile. I mean, every company is trying to make the leap to fully-autonomous vehicles, but most advancement comes from continuing basic research and slow-and-steady improvement... but that doesn't get mentioned much because it's not media-sexy. As for being "crazy" - this isn't anything close to the development idea I'm half-tempted to write up for the IIHS… (Though I'm half-certain it's been thought of before and another half-certain that some companies have already started doing this without proper disclosure, leading to a lot of bad press and a small fine.)

I vaguely recall reading, many years back, about a proof of concept autonomous vehicle that did something similar to what you suggested. As I recall, they buried magents of some description in the road surface every so often, and the vehicle had sensors underneath that could follow these tracks. Neat idea, but as a couple of other mentioned in the video's comments, it'd require each country/state/county/city to actually go ahead and bury millions of magnets in the road surfaces at predefined intervals and to within a specific tolerance which, when you look at the bigger picture, is a much bigger ask to get this right in a way that vehicles can be compatible with. As a result, the whole concept is/was impractical and the project was binned. Cursory Google search seems to suggest Volvo was involved, and even they realised it was impractical, calculating the cost of implementation at over $22k/km. I think we're heading down the right track (pun intended) with the way autonomous vehicles are being developed, but you're right, there's a lot of risk involved with the way it has to be built.

BTW, "private paths" for transportation is, basically, the bottom line of The Boring Company. (OK, you can tell that I agree with Elon Musk in many things :-) )

Andres Plaza

I think the idea is good, but for public transportation. For regular cars, the problem is that each city/state/country can have its own infrastructure that can vary a lot, so it make it unviable (because the cars will have to support all of them, or only one and be limited to a geographical sector). It is better to make them use the current infrastructure.

Andres Plaza

My random tired thoughts: * Isn't lane striping pretty much guide wires? * The color of lane striping is irrelevant in snow. * Good luck passing licenses-to-kill-for-robot-cars law. "Purge: Cars" is a great crossover movie idea and explains why there are no humans in the cars universe. * Weren't horses the best autonomous vehicles?

I don't particularly agree with your thesis, as I think the problem with self-driving isn't the steady-state (as it were) driving down a highway. The problem is dealing with all the stuff guide-wires wouldn't solve: the unexpected stuff like trash, people, animals, in the road. It would definitely help with the snow/less than ideal vision situation, but it won't help deal with all the other drivers on the highway that are driving themselves. Any infrastructure solution that requires special lanes or a separate highway would be prohibitively expensive to implement. Think about how it took 35 years to build out the original Interstate highway system. How would we ever convince taxpayers to build out a completely separate highway system just for AVs? So guidewires might be a help, but not a panacea.

Ashley Grant

Yes, good idea, but for most of the time (i.e. when it's not snowing), current image processing techniques looking at road lines are good enough. Plus, it's cheaper to redirect autonomous vehicles with white lines. People may call for it when a lot of people are using autonomous vehicles and they find they can't use it in the snow / fog. You would need an additional sensor in every car to make this work, plus need current autonomous sensors, so cars would cost more. As you can see from this Wikipedia article, it's been thought about since the 1930's. <a href="https://en.wikipedia.org/wiki/History_of_self-driving_cars" rel="nofollow noopener" target="_blank">https://en.wikipedia.org/wiki/History_of_self-driving_cars</a> As others have said, you could modify raised pavement markers (<a href="https://en.wikipedia.org/wiki/Raised_pavement_marker)" rel="nofollow noopener" target="_blank">https://en.wikipedia.org/wiki/Raised_pavement_marker)</a> to do this. Cheap to create and install, and something traffic highway authorities are used to installing. Even with this idea, self-driving in snow will be tough. Imagine I'm in the fast lane and want to get off at an intersection. How many lanes do I need to cross? The car may know this, but the highway may have changed since. Current technologies are trying to cater for the world how it is, not how it's described on their storage as this will never be perfect. 802.11p (<a href="https://en.wikipedia.org/wiki/IEEE_802.11p)" rel="nofollow noopener" target="_blank">https://en.wikipedia.org/wiki/IEEE_802.11p)</a> is the V2V stuff VW and others are championing, but it won't guide you down the road in the snow, even if all other cars have it. I feel all autonomous tech will start on the freeway. I'm happy to buy a car that will autonomously drive me on the freeway and I'll get there, as most of my driving is on the freeway anyway. I'm skeptical autonomous driving can be solved for 100% of scenarios (let me know if you want some scenarios I don't think computers can solve), but it could get to 98%.

Big Car

1. Get some coffee, 2. I also agree that it is bizarre that auto makers are trying to make cars that can auto navigate through our current day roads that are at the mercy of weather, construction, just being old, etc... 3. I'd love to see you explore this topic, it's a fascinating one that will become all of our reality in the near future

In the sense of Interstates, it would probably work. I truly agree on adding technological aids for truly autonomous vehicles. Until we can make all streets a true no-go for pedestrians, AVs will need to monitor their surroundings. When it comes to dramatically changing the paradigms of life versus making AVs exist in the world, changing AVs will win every time. Given how far we are down the path of true autonomy, we might as well keep going down that path instead of figuring out how to isolate pedestrians entirely. As much as it makes sense on a hypothetical level. I think you might be underestimating the costs of conversion to pedestrian-free roads. If you take Chicago, for example, how do you reconfigure every street so that they're pedestrian-free? Both in terms of finding space for conversion in dense urban areas and sheer scope of converting low-density residential areas. If pedestrians still exist, AVs will still have to track them and be aware of their surroundings. In terms of only looking forward, I don't have a feeling for how much more complex a 360 tracking system would be. My guess is that problem isn't hugely more difficult compared to alternatives. At the end of the day, there's definitely merit in the idea of dumb-tech guide systems as an accompaniment to the true AI systems. The costs of changing the world to fit dumb-tech guidance, rather than developing smarter tech is probably too high to be worth it in the long-run. I think putting guide wires in Interstates as a form of autopilot is probably a good starting point for getting wider acceptance of AVs. They'll still need to be able to detect objects, though. [Unless we fully fence off all freeways.]

Kevin Kostka

I don't necessarily agree that guidewires are the way to go, but I'd really like to hear some more points from you. I find your ramble very interesting and I think it'd be valuable if you shed some light on this side on the argument. So, I give it a yay.

Nordfuchs

I love your opinion series. I think your style of analysis would be an interesting viewpoint, even if has been covered by people before. I think infrastructure would still be a major hurdle, even with the simplicity of guide wires. In my opinion, the current tech of AV's will win out every time, even if it is more complicated. But when it comes to people movers, I think this guide wire technique is a no-brainer. For smaller applications like parking lot trams, it would be a lot easier/cheaper to embed a wire than to have to build an elevated platform or tracks. I say, do the video. If anything, its a tech people might not know about.

Lucas Bennetts

It'd be interesting to research into this. I seem to recall "Popular Mechanics" type magazines proposed solutions like this. Not guide wires, if I recall correctly, but beacons. The idea is basically the same. It makes it easier for a car to follow, but requires all things, not just cars, on the road to understand this new technology, so your initial comment about separation is key. And you still need a way for cars to sense obstacles, since you'll need to make sure you don't plow into a car in front of you that's decided to stop working. This doesn't provide a path that current autonomous cars provides of slowly upgrading and adding cars to our roads, it's all or nothing. And that means you no longer have a cheap solution you can put on some roads, but a solution that requires new roads or taking existing roads making them "new cars only" roads, which won't sell. Also: Anytime you talk about adding new features to roads for cars to sense, and only cars could sense, you have to ask yourself: How are you going to protect it?

Scott Seligman

I don't know, the way AI is evolving is really impressing, and I think that in a few years, we won't need anything but cameras and sensors to achieve an -almost- failproof car, at least way safer than anything driven by humans.

lululombard

Yeah I should have clarified--the guide wire idea in no way solves those problems, it's more the separation aspect that I think might be worth discussing. Has anyone ever considered requiring some sort of V2V protocol that could be easily added to a non-AV? I've always kind of felt that a simple device with an accelerometer and a radio could serve as a beacon for AVs to know what other vehicles are doing without relying on visual input. I would imagine a similar protocol could work with police vehicles. Since you are an expert, mind if I send some questions to you?

Technology Connections

My thought: by using wire guided systems, you take away the "car" aspect of the vehicle and make it a train - which nobody wants or asks for. The goal is to make the vehicle smart enough to drive on the existing infrastructure and use its full potential (e.g. even leaving the road if considered "safe"). A wire does not help to mitigate the human error aspect - people show up in all kinds of places "where they are not supposed to be". A wire would not help a bit.

Markus Schumacher

This is certainly a subject worth pursuing. Driving is complex due to the unpredictability of surrounding events. I see Nathan's post just came in -- spot on!

Michael Steeves

Hey Alec, first my background is in this space, I work on the self driving team at Toyota Research Institute and previously at Google X. I think you miss-attribute what the difficulty is. Getting a car to stay safely in the lane is easy, the hard part is dealing with humans and unexpected situations.... For example, how do you know program the car to stop if a police officer is directing traffic? If you want some research papers I would be happy to dig them up for you.

Like he says at the end, it's not necessarily tearing up infrastructure nor a physical wire. It could just be a series of RFIDs laced on top of a road or in some pattern.

avfusion

Do you know how much it would cost to tear up every road in America and bury a wire? And what about rural areas? Would you bury a wire in EVERY street in the country? The tech being developed now would use existing infrastructure and would play nice with current cars as they age out of the national fleet.

I'm sure it's somewhere. I'll need to research some arguments for and against (and I will have some time to kill shortly). Of course I don't have figures, but I wonder if Google, Waymo, etc. could actually save money just by funding guide wires or similar while reducing their AI research and development costs.

Technology Connections

You know those reflectors they use on roads so you can see the lines in the dark? There's been talk of embedding RFIDs into them that cars can talk to. They can be several tens or hundreds of meters apart, and they're fairly cheap to add into the cost of each reflector. They just become a normal maintenance job rather than new infrastructure. The car just needs to triangulate itself between them, and every RFID can identify itself uniquely so the car can fairly intelligently understand where it is in relation. Works in snow, doesn't need every one working, and is processor un-intensive.

avfusion

I am 99% certain you’ll find this idea has been discussed when you start your research. I seem to recall it but can’t remember enough to tell you where to look. I think it’s a good idea as well but I want to say the reason google is trying so hard with their approach to as that every municipality balked at any infrastructure cost. I.e. state costs and local costs would be a voting nightmare