I think instrumenting the roads in one way or another will go a long way to enable real autonomous vehicles. Sure it's expensive, but on the upside it might actually work in this century. We just need to define a standard and let the free market go to work.
I'm typically an autonomous skeptic, but I disagree. I think autonomous cars are decades away UNLESS we start designing roads specifically for vehicular autonomy. design lanes that are extremely easy for cameras to see. Traffic lights that are easy for computers to parse. Street signs like QR codes etc.
Obviously the cost of this is prohibitive. But at some point we have to get it through our heads that for a computer to excel at a task, the task needs to be redesigned.
For years I've thought full autonomous driving is never going to be practical without an equivalent engineering effort on the road infrastructure side of things. Standardized markings, drive by wire, special AI compatible lanes, etc. Why are we still doing this the hard way?
We need smart roads. Not just smart cars. Until that happens, autonomous is impractical. You don’t want to build an autonomous system to match human error rate. That basically got us nowhere in the past seven years since Elon bs’d his way to selling vaporware.
So maybe we need high definition maps, more than just cameras for sensors, and tight clear government regulation in order to get to full autonomous driving?
Ha-le-freaking-lujah! I've been saying the same for years, ever since I completed the Udacity self-driving car AI course (taught by Sebastian Thrun, then head guru of autonomous vehicle research at Google). 70% of the problem was solved in mid-90's, by Ernst Dickmanns et al, Daimler Benz had autonomous cars on the autobahn. After billions of dollars and two decades, we solved another 20% of the problem. The remaining 10% is not going to be conclusively solved in the next 20 years at least, not under the current constraints.
Now if roads are outfitted with instrumentation for autonomous vehicles, _and_ human drivers are prohibited on such roads, _then_ we _might_ see full autonomy. But not before.
If it were up to me, I'd focus on this instrumentation instead: RF guide wires/tags for car localization on the pavement, machine readable signage (even in fog, snow, and heavy rain -- conditions with which cameras and LIDARs cannot deal in principle), inter-car coordination mesh networks and security thereof, autonomous vehicle readable road work signage, police gear to direct traffic of autonomous vehicles, and so on and so forth. There. Hundred billion dollars worth of startups in one paragraph.
As things stand, you can only be fully autonomous at 25mph in California where it never rains, as long as there's no fog, no road work, and no one has messed up the markings on the pavement.
The capability we have today is working on fully mapped, well lit, dry roads. No one has an automatic car that could navigate in heavy snow/rain or unmapped terrain. There's very few ideas how to tackle it, unless we equip every road in some sort of trackers. I fully believe that in short time we will have all sorts of very clever cruise control systems which are essentially like Tesla Autopilot, maybe a bit more versatile. But fully autonomous vehicles? I stand by my statement that they are very far away for commercial distribution, unless you plan on selling them in Florida only, or only for use on private land.
I think I am going to side with VW on this. I have always been skeptical of fully autonomous vehicles, and I do not believe they will _ever_ exist on the roads that currently stand. Driving safely in all conditions without aid from a human is simply too complex a task for code that can be audited and verified. If some AI model that's been trained on a billion years of driving experience shows promise, but it is some incomprehensible black box of weights, I won't be getting in that car.
Autonomous vehicles will only ever truly exist upon infrastructure literally designed to aid them, greatly simplifying how they need to interact with the environment, thus making the problem tractable with code we can prove works. I really think it will take more than putting markings on existing roads. It is going to take new roads full stop, probably with various wireless checkpoints built into them.
Always felt like autonomous driving without smart roads isn't going to happen on a large scale, just computer vision isn't enough.
What would a smart road be? Either some infrastructure that is designed for this or a digital overlay attached to it that every car nearby continuously updates in real time. Or maybe both.
That would mean an open standard and shared access to such an overlay, plus technology for that would have to be mandatory for new cars regardless of whether they plan to be autonomous or not. Right now it's just a capitalist rat race that no one is winning.
I feel like we are going to need smart roads before we can have autonomous vehicles. If the road could talk to the car, the car would be harder to fool or confuse. Maybe.
Perhaps an easier and quicker fix (although extremely costly) is to outfit all roadways with aids for autonomous driving. Otherwise, we will have to wait decades for the technology to be mature enough.
The AI part of autonomous driving could have easily been handled by creating a network of roads (that is, converting existing roads) that has the appropriate markings and that is easy navigable by machines.
But that requires cooperation between governments and/with the private sector, and in 2019, God forbid that.
That sounds great if we want to rebuild our entire road infrastructure. So using this for self driving will never happen since it would require every country to implement a very costly standard.
Sounds great in practice but isn’t realistic in the slightest
If you added up all the R&D money spent on these self-driving features, we probably could have already covered 80% of American roads with simple guidance systems that would assist in piloting cars safely. Then nav systems wouldn't need to be as complex, and wouldn't depend so much on trying to guess road factors.
If self-driving really is the future, the government's going to need to be involved, the way it is in building and maintaining roads. Many roads already have cameras and sensors built into them just for measuring traffic, so it's not a big leap to say it should also have tech to improve automated driving.
The thing I've always wondered about is: if governments can put street signs for humans on every road, why can't they do the same for autonomous vehicles? Sure, you could vandalize the signs and wreak havoc, but you can also remove a stop sign or hack a traffic light and do the same now. If a standards committee was formed to develop a spec for autonomous vehicle guides it seems like we could get to full autonomy far faster than waiting for AGI. Maybe you still have to drive on backcountry dirt roads, but wouldn't automating 90% of traffic be an enormous win for society?
I share that sentiment and in light of that, do you think the better and more economic way forward is possibly to not improve self driving cars as much but the infrastructure around them? Geofencing, urban and street design conductive to autonomous vehicles and so forth?
While we may get to the point where sensors and a form of AI can reliably drive a car, we're so far from it that we don't even know if we're heading towards the goal post.
What we need, if we want to solve the partial self driving problem right now, is not fully autonomous systems. We could start by agreeing on a way for cars to sense the road - think in the way robotic lawn mowers work where you have to bury cables. Then implement that in highways, ring roads and other long stretches of road.
Then we can agree on a way for cars to communicate and exchange relevant information.
Then we can equip cars with hardware like the Tesla have, and then use it to collect data and compare the "auto pilot" with the known working system of following a signal in the road. We still need LIDAR to reliably stop for other cars, but that is a mostly solved problem today.
Agreed. We won't ever get autonomous systems. I don't think that should even be a goal for production cars. We'll get roads/areas retrofit with signals to help cars and I think some type of coordination signals/beacon locally between cars.
I'm sure people are thinking/working on this but it seems that nobody is talking about the infrastructure improvements that would likely be necessary to support fully autonomous vehicles (smart roadways, etc). To me, this should really be a two-pronged approach; the vehicle and the environment. Trying to engineer a vehicle to successfully navigate an environment soley designed for human cognition seems like a losing battle. But, if roadways were fitted with a myriad of sensors and communication networks which couldn't be thwarted by a can of spray paint, I think the success rate of the actual autonomous vehicle would be much better.
I think it would be hard to argue that we are a long way from fully autonomous vehicles, since millions of miles have already been driven by autonomous vehicles. What remains is mostly regulatory hurdles, psychological hurdles, and fine tuning.
reply