Of course, I do understand the importance of getting infrastructure support for the future of "connected cars." But so many things have happened since the industry conjured up the blueprint of V2V and V2I. As some of the industry analysts point out in the story, it would be a big mistake for the U.S. mobile industry not to get involved in the V2X trials.
I guess I'm missing the point of the article, Junko. The title implies doubt as to the need for more than just local in-car sensors, and yet then you introduce the idea of leveraging off the cell telephone infrastructure, precisely to help provide that V2I link.
It seems rather obvious why you need information from the roadways and other vehicles. If for no other reason, when the self-driving car plans out a route, something needs to tell it if the route is viable or not. For that, you need real-time information. Also, you can't always rely on the visibility or perfection of lane marker and road edge stripes, for the car to steer itself. What about the intentions of adjacent cars?
Maybe the real question is, what form will these V2I and V2V links take? Are they mostly going to be optical, where the self-driving cars detect lights and reads signals from the road and from other cars, as we do with our eyes? Or are these optical cues going to be replaced by RF comms?
That discussion would make sense to me. Pretending a self-driving car doesn't need information transmitted in real-time, by other cars and by the roads, optical or RF, or a mix of the two, does not make sense.
V2X should take a lot of different "communication" forms. And yet, the mandate that's been discussed in the U.S. right now is abouthe use of Dedicated Short Range Communication (DSRC) tech operating at the 5.9 GHz frequency based on 802.11p.
I do understand that DSRC is useful and effective; but look, the auto industry sat on that spectrum for more than a decade now?
More importantly, LTE should definitelly become a part of the mix of V2X discussions. And yet, that hasn't exactly happened. US cellular operators are NOT part of the ongoing V2X trials and sitting on the sideline.
I am not sure even the short range communication will be reliable enough? It will be beyond acceptable that due to lost in signal or noisy channel car was not able to right decision leading to an accident...
DSRC devices for vehicular networks will be used for V2V communication and V2I communication (with roadside stations). DSRC works in 5.9 GHz band with bandwidth of 75 MHz and approximate range of 1000m
What happens if a newspaper blows on the windshield of your car, while you're driving, obscuring your vision completely? What happens if someone from a passing car sprays paint all over your windshield? What happens if you turn your head because your child is throwing things around in the back seat? Same thing. You need info from the environment to be able to drive.
@Bert22306 I do agree that driving is never risk free. However there is human element here who is smart enough to react to the situation. Also there are incident where the situation occurs and accidents do happen. I am just to point out that its just adding risk to it. And its just like any other new technology always benifits comes with some additional risks. The point I was making is that any such failure can lead to mass failures like if some beacon goes out of working order, then car looses its eyes and it wil happen to every car which is passing by there. There will be many such new scenarios which we are not able to think of and need to be figured out once this system will be deployed.
Also even one bad person can lead to accident, however there is always a smartmind behind wheel which can adapt to situation and reacts. In automation world we will be under controll of machines and they are not adaptive and smart enough.
I am not against system, just bit worried about learnings from its initial phases and how robust it will be once we establish it.
p_g, when you say, "Also even one bad person can lead to accident, however there is always a smartmind behind wheel which can adapt to situation and reacts. In automation world we will be under controll of machines and they are not adaptive and smart enough," is ther not perhaps an implied "yet" in there?
I've worked on or with automatic control systems for most of my career. It seems to me that the trend, for a very long time, has been that the more sophisticated a system becomes, the more the human operator MUST be taken out of the loop. It's usually a case where progress becomes impossible until some existing key human operator function is replaced with automation. Reason being, algorithms running on computers can be faster, more consistent, more precise, and more predictable, than that "smart" human.
Driving a car is a complex task, so althopugh a large number of manual controls have already been automated, the basic throttle, steering, and brakes are *mostly* still under human control. But even there, hard to deny that the trends to automation are obvious.
While the principles behind auto pilot in flying an airplane is based on "taking human decisions out of the equation," I think it will take a long time for drivers to get used to that "auto pilot" concept.
We would all like to belive that human beings are fully capable of making sound decisions at the moment's notice --- but let's face it, that's not always the case in many accidents we see on the road.
But at the same time, what's yet not clear to many of today's drivers like ourselves is how cars of the future could consistently make better decisions than what we do while driving.
The Google cars have driven over 500,000 miles on public highways safely. They seem to have dealt with lane markers, road edges, traffic signals, and the intentions of adjacent drivers. They've gotten to their destinations without real-time information abut their entire route.
"Connectedness" will be useful, but it's not necessary.
Sorry, I strongly disagree. Connectedness is essential for a real solution, one that works without limiting yourself only to carefully pre-planned routes and with a driver ready to take over instantly. The Google car is far from a complete solution:
How would the Google solution work when a road has been milled, or when it has just been repaved and lane markers not yet painted? How would the Google solution adjust the route when that bridge is blocked for repair or for an accident? The Google car assumes the speed limit on a road based on a database. How does it know when that speed limit has been changed, for some short-term reason? These, and the examples in the article above, prevent the existing Goggle solution from being complete. The human driver cannot be rerading the paper or taking a nap. To resolve the limitations, you need improved info from the environment, info that manual driving does provide to the driver.
The simple fact is, driving is never an "autonomous" experience. The driver is taking in all manner of inputs, all of the time, from the infrastructure and from other vehicles. The complete solution for driverless cars cannot ignore this. Instead, the solution has to replace or emulate these comms.
You forgot the most important design constraint on a self driving car, weather. It has to be able to operate year-round at all seasons in all weather conditions that a person can drive a car in, or its dead-on-arival for most areas where people live. So forget about standard speeds, it has to adjust its speed for the current conditions. Optically observing lines painted on the road can be useful, but not required. It must operate both properly and safely when the road is covered in snow and none of those lines are visible. Fog, rain, snow, wind, etc., whatever a person can drive in, it must be able to drive in too.
The Google car has a pretty precise 3D map of the area it's driving through. It knows within a few centimeters where it is within the roadway. I don't know if lane markers are part of its map, but it easily could be; it's not much additional data. If that data is there, the car can stay in its lane whether the lanes are marked or not. The google car might be able to handle this situation right now; if not, it would be a small, incremental improvement. So missing lane markers don't seem to make instructions from the roadway necessary.
As you mention, the Google car stores speed limits, but it can handle stop lights. It doesn't seem hard to me to teach it to read speed limit signs. That might seem like hand waving, but OCR has been around for a long time, and speed limit signs are simple. So another thing you present as a stopper seems to require a just small improvement.
Same for the bridge problem. Making sure that the map information that a self-driving car would follow on its trip is updated quickly when a section of road becomes unpassable is an incremental improvement.
I agree that the Google car is not a complete solution today. The drivers of Google cars do occasionally have to correct the car. The article you cite mentions two interventions over 140,000 miles. Each such correction is fuel for the improvement of the algorithm, and they're acquiring additional experience every day. Remember, the best research teams in the world couldn't get as far as eight miles in the desert without crashing, a mere nine years ago.
There are some things that I don't think the Google car can't handle right now: some weather conditions, construction, a collision blocking the road, and an intersection controlled by a traffic cop. I say it's only a matter of time. Not much, though, given how far they've come in so short a time.
There is information that the infrastructure could provide that would be useful to a driver, whether it's a person or a computer. It could tell me how long before the light that I'm approaching turns green. It could tell me if there's a car stopped in my lane around that curve. Providing that information would prevent a few accidents and move traffic along somewhat faster. But self-driving cars can work with no changes to the infrastructure and would still be a vast improvement.
"As you mention, the Google car stores speed limits, but it can handle stop lights. It doesn't seem hard to me to teach it to read speed limit signs."
We are in violent agreement! As I suggested previously, a better questionm to ask might be, will, or should, V2I and V2V comms be assumed to be RF only, or might they not be optical? If optical, then in principle, the EXISTING solutions, i.e. road signs and markings on the pavement, traffic lights, and the existing lighting and signaling schemes in other cars, could be drectly adapted for self-driving V2I and V2V comms. Just as a human determines the intentions of other drivers by the lighting cues, so can an algorithm in the slef-driving car.
The police can alreadu scan license plates electrnically. Surely, similar tech can be adopted to read all manner of road signs.
Here's a simple exercise. Take one of those manuals that teaches the meaning of road signs and markings. Then run through each one and ask how a self-driving vehicle would get that same info, or conversely, why a self driving vehicle would not need that information. For example, a self-driving vehicle may not need a "deer crossing" warning sign, presumably because it's radar and lidar are always alert. But a self driving vehicle would need lane merging indications, speed limits, traffic lights, lane markers, road edge markers, bridge height limita info, and a whole slew of other info.
Not sure where that came from, Barry, since I was agreeing with the post where you mentioned electronically reading the speed limit signs.
My disagreement with you, and with the title of the article, was the implied notion that self driving cars don't need V2I comms. Just because some regulators aren't coincentrating on V2I at the moment, or because the existing Google car doesn't use much V2I (it does use some, even if the information is potentially stale).
Electronically scanning road signs or pavement markings constitutes V2I. Perhaps you missed where I suggested that these V2I and V2V comms may not be RF-based at all.
You did reply to my comment about the speed limit signs, and I appreciate it, but you listed many things that you said made v2x necessary, to which I offered ways that self-driving cars might handle them without it, and I was hoping for your comments on all of them; instead you offered more. Whack-a-Mole's not much fun.
If even one of them prevents self-driving cars from operating safely, well, then v2i is necessary, and that would be a shame, because self-driving vehicles would take a lot longer to arrive and would cost a lot more. So I tried to respond to all of your points and wanted to know what you thought.
Regarding the simple exercise: all street signs are standard images that the car can store and match with the image it sees. That image can be associated in the car's database with whatever information you think a car should receive via a v2i system.
Okay, Barry, I guess it'sd because I thought I had covered all the points when I said that V2I doesn't have to be limited to RF. So here goes, your points about why you don't need V2I:
1. "Precise 3D maps." That's a form of V2I, although often too stale to be of any tactical use. The car can't get these maps with just internal sensors. A human creates those, and this is clearly infrastructure-related information sent to the vehicle one way or another.
2. "Stop lights." Reading stop lights optically is a form of V2I. The stop lights are human-generated information provided by the infrastructure.
3. "Read road signs." Same thing as stop lights. It's V2I. Whether those signs are permanent or temporary, they are infrastructure-related information that someone outside the car has to create and transfer to the vehicle.
4. "Maps updated quickly." That too is V2I. The infrastructure has to provide those maps, and btw, realistically, they have to be transferred via some RF link. How else would these real-time maps be made available?
Then you list examples where a car's own sensors can't handle the situation at this time, which we can more or less agree on. Although weather consitions should fall among those that a car's own sensors should be able to manage eventually, since after all, human drivers have to do that autonomously now. (E.g., cars can measure outside temp and humidity, and optically scan the road ahead, which will give a darned good indication of the risk of black ice, for instance.)
@BarrySweezey, thanks for commenting. While what you pointed out here is true, I'd have to say that we are hardly there yet -- at least at this point and time. Google car's demo has been more or less a "show." What we need is a much more robust testing and verifications.
Honestly, I think the auto companies are busy selling cars, know that real self-driving cars would require collaboration and endless series of meetings with a zillion private and government organizations, and consequently do NOT see any urgency in this matter. Car companies know that cup holders and LCD displays are where the real action is, not fully self driving cars. That's because they know their customers.
Whereas Google is looking for free advertizing from the press, and getting it.
Don't get me wrong. It would be more fun to be a Google engineer playing with these algorithms, than to be an auto company engineer who is told to stop spending time on that, and make a cheaper windshield wiper system instead. Realistically, as many, many articles have said, this self driving car thing is going to be a gradual evolution, where driver assistance will be the main thing for a very long time.
I don't see the huge urgency either, Junko. Driver assitance systems are infinitely easier to design. And RF-based V2X systems are also more credible for the near term, than would be V2I and V2V optical systems that will reliably read road signs and see traffic lights, brake lights, or turn signals from other cars. IMO.
For any real move to automated driving V2I has to be the backbone. V2V can help but IMO should always be secondary to V2I with target separation over 50ft or so.
Fully autonomous vehicles are a flash in the pan IMO. I would be extremely concerned to see multiple manufacturers stand alone implementation of autonomy driving on the same road at the same time. I've seen no comments on what the failure mode of Lidar/radar and the various sensors used to provide the data on which decisions are being made in each vehicle. Even if autonomy results in a reduction in accidents, there still will be some I assume, and the degradation of information fed into V2V network might be questionable.
V2I moves the responsibility of major sensor data to a fully redundant and reliable infrastructure with little chance of being compromised when accidents do occur. Scaling the sensor and communication network then becomes a public cost, hopefully with the correct maintenance and calibration included. Infrastructure based sensors should be a fraction of the cost of mobile sensors overall, and the computing power required is more easily planned as infrastructure.
Autonomous solutions for parking, multi-point turns and vehicles lane and separation verification would seem like good targets for the auto companies and close proximity low speed V2V. However autonomy of thousands of vehicles in a local region (a few blocks) IMO will be a big fail without V2I.
As a last comment on the need for V2I, this provides the Holy Grail for traffic route planning. With the proper data available, all streets and freeway lanes can be used and traffic dynamically allocated as conditions change. Route mapping now becomes an endpoint to endpoint decision with the infrastructure. You don't get to choose the route way points, and all roads become toll enabled
The way I see it, Junko, responding to your DSRC comment, this is indeed a viable and credible approach for V2V, which in some more futuristic reality could be replaced with optical systems in cars. There's no way that DSRC is replaceable by cellular schemes, though. You don't want to rely on a fixed cellular infrastructure for the immediate comms between adjacent vehicles. A WiFi-type of protocol is instead a good fit.
As to the cellcos, all of the existing telematics systems, like GM's OnStar, do use cellcos as their infrastructure. So at some level, they are playing with the automakers already. Maybe they aren't designing the road sensors and message transfer protocols required for V2I, but then again, I'm not sure I'd expect them to.
What will more likely transpire is, road information system designers will develop their schemes and then use existing cellcos as their transmission medium. Sort of the same as these telematics systems did. Verizon didn't create OnStar. It was just the choice of the OnStar system designers, when analog cellular went off the air. They could have picked any other cell company.
1) LTE is just an air interface, so if they are deciding to run LTE car-to-car, that may be great. But, I for one will not pay the extortionate fee the carriers want just to get LTE data for my car. I have grandfathered unlimited data on the smartphone, but I can't see people with these ridiculously low data caps the carriers favor these days wanting their car using up all their data either.
2) Privacy implications (not LTE speciifc but a problem with anything beyond communications with surrounding vehicles). There should be something like key rotation on car restart or the like, to make sure there's enough info to determine when there's traffic jams and so on, but not enough to track a specific car as it moves around. The authorities have shown their willingness to perform illegal and unconstitutional actions on a continuing basis, violating privacy rights. Because of this these systems will be a hard sell unless anonymity is assured.
Just saying, this is a big technical problem, but a working technical solution will still not be accepted if it doesn't keep track of some practical matters.
The challenges of autonomous car communications - and the very different reaction times of autonomous and human driven cars, make me predict that dedicated lanes will be used to achieve the greatest benefits of autonomous cars. I could imagine cars driving at high speeds and very close together if they were all following the same "rules of the road" and the lead vehicle could alert the trailing cars of hazards ahead. With the long cycle time to deploy new automotive technology (and the very long life of legacy cars), I suspect that the initial deployments will be in repurposed High Occupancy Vehicle (HOV) lanes. Use of a qualified vehicle will result in dramatic reductions in travel time.
Just to be clear, I do understand all the implications as to how both V2I and V2V are needed for the ultimate future of self-driving cars that may not be here for a long time -- yet.
But the purpose of this article is to point out a shift in focus in temrs of the market and regulators.
Since the early 2000s, the U.S. Dept. of Transportation has worked closely with the major automotive manufacturers and other state and private sector stakeholders on V2I and V2V.
While their effort initially focused on V2I, the department's more recent focus, through the NHTSA, has been on V2V safety communications, with a secondary focus on V2I and vehicle-to-mobile device applications.
That I see as a larger trend emerging in the segment.
I understand the point about the regulators' "shift in focus," but I guess my point is, this doesn't change reality. Or said another way, just because regulators at NHTSA are concentrating only on V2V at present, that doesn't mean that this is because V2V is sufficient for self-driving. It's just an artifact of what NHTSA might have determined to be manageable, short term. It's a lot easier to "stick" the expense of this vehicle evolution on the auto makers (V2V is their responsibility and cost), than to burden the DOT with the cost of upgrading the infrastructure. That's all. I wouldn't infer anything more from this shift.
Reality is that for the true self-driving car, one needs both types of comms. If we concentrate first on V2V, fine and good, and until we spend quality time with V2I, we'll not get that self driving car.
@Yoshida. The shift in focus in the article is understood, but V2V provides only a small subset of the communications required to support a mass of autonomous vehicles. As an incremental step it is worthwhile, but lets not kid ourselves that it leads to full automomy.
Given that there are many years ahead of a hybridised environment (both sensor data and vehicles), the chances of having successful per car autonomy is close to zero IMO. A great example is the amount of effort and computing power thrown at the DARPA challenges; do you really think that scales economically to a large population of vehicles?
Google have done a great job with a technology demonstrator, but the solution is a decade or more from prime time and is cost prohibative. Having the driver "on alert" just in case is dead in the water IMO. When eventually in a prime time state.....
1. Will the same automation controller be used in all vehicles ($20k-150k)?
2. Will there be a mandated standard for the automation responses or will different manufacturers have their own special sauce?
3. What will be the autonomous rule set when an accident is unavoidable?
4. Does the autonomous system allow speeding, swerving etc? Is there even a manual mode?
4. What about the legal nightmares after an accident involving an autonomous system (or worse still two)?
An Infrastructure based control system can more readily set speeds and spacing and has a better chance of not having a cascading accident front when things do go wrong. Once the infrastrucure sets the driving parameters there will be no speeding, running red lights or other traffic infringements, so no need for fast cars etc. (Though I doubt that the cars will actually be appealing then).
Islands of autonomy are in general a very bad idea when mixed in with human responders, and it won't be until automation is mandated for all that it can truly live up to expectations.
Thank you, Junko, for a brave piece of journalism. It's tough to swim against the do-good DOT current of V2V myopia. The best thing the DOT could do at this point is encourage Google et. al. in their private efforts to develop cars that do not collide with others cars!
#1 - The DOT has been spinning its wheels with nothing to show for 10 years sitting on valuable spectrum without saving a single life or delivering a single monetizable application.
#2 - The key objective of V2V, of course, is not autonomous vehicles but collision avoidance - autonomy is a by-product highlighted by the introduction of the Google car.
#3 - Google's arrival on the scene has changed the game, simultaneously demonstrating the power of private enterprise to stimulate technological progress and market adoption AND the shortcomings of leaving such initiatives - especially technical specifications/definitions/requirements - to governments.
#4 - Trying to specify telecom-related standards divorced from market demand is a strategy with a long history of failure.
#5 - The government should require (not I am not using the word "mandate") that class 6,7,8 trucks, emergency and service/construction vehicles be equipped with beacons to make their presence known to drivers and traffic controllers - yes, using DSRC technology.
#6 - All Bluetooth roadside implementations for measuring traffic flow should be equipped with DSRC as well.
#7 - In this way - and others - DSRC can find its way into a "demand driven" implementation scenario.
#8 - The DOT needs to facilitate, encourage and, fundamentally, get out of the way of progress - which includes keeping its hands and regulations off self-driving car development and progress.
Please don't get me wrong, I am a supporter of automation for vehicles, I just want to get there in the right way. I'd rather see the efforts initially in driver assist where huge progress could be made very quickly. Leave the automated vehicle as a pure research program for now.
With relatively small incremental changes to the auto environment society could make the roads much safer (using V2I).
1. Monitor every vehicle all the time (vehicle recognition and registration, vehicle location, driver recognition, driver monitoring, speed, lights, signals, car spacing etc). The roads are public and we have no right to break any rules while using them.
2. Brake assist, parking assist, route assist (tell the driver which lane to be in) etc
3. Infrastructure based speed control and vehicle start lock. (if the speed limit is 25mph, why is the vehicle ever capable of exceeding this zones speed limit? If the vehicle has the potential to be dangerous due to a fault, why allow it to start?)
With a small and incremental increase in infrastructure costs we could eliminate unregistered vehicles, unsafe vehicles, auto theft, speeding, DUI, and the potential for a huge number of minor infractions.
After a decade or so of reasonable penetration of the above, we'd almost be ready for the next step...more automation.
It is what it is. Ten years is too long to have valuable spectrum set aside with nothing to show for it. The current standards-setting regime has failed. Time to move in. The results speak for themselves. And, yes, I am deadly serious - 100 people killed per day. We have the technology and standards in hand today.
i think you have it backwards - the V2V standards setting activity has become a boil the ocean proposition. Commercial solutions exist today and are being blocked by regulators or by too-cautious commercial interests with too much at stake to take the necessary risks. Deploy DSRC technology NOW on commercial vehicles and at dangerous intersections. Leverage and encourage parallel wireless advances. Google is not boiling the ocean - just demonstrating what can be achieved when risks are taken. Let the market decide, not the scientists. Does anyone really trust the government to run this operation anyway? Does anyone WANT the government to run things? It almost guarantees failure and/or consumer rejection.
Some days ago the first Mercedes S-Class drove ~100 km autonomously on the streets from Mannheim to Pforzheim in Germany. The same way Bertha Benz and her sons drove with the Benz Patent-Motorwagen on 5 August 1888 in order to demonstrate the suitability of her husband's construction. So, I believe we are closer to autonomous driving than many people think.
In America we have about 130Million cars driving (in 2007) about 30miles per day. Say 1% of them changes to self-driver car, we are talkking about 40 Million miles of self driven per day. One testing of 100KM has not seen many situations which 40miles/day will see. It will certainly take many-many years and lots of improvement cycles before we can say good self-driven cars are ready.
Concept of electic cars was extremely simple, just put battery and a motor to drive the car, still take 15+ years of work (about 10 years from first electric car on road) to come to point where we had limited success and may be to hit a reasonable car like Tesla.
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...