Doing everything in real-time is difficult with current sensor technology and processing algorithms, and leaves a lot of room for error. The way that they're doing it currently is fairly clever for minimizing the amount of calculation/reaction time. I'm not so sure it absolutely needs to be real-time, so long as the system works. That said, I'm no stranger to the 3D CAD universe, and I can tell you that housing all of the spatial data, regardless of format, is a LOT of disk space. A one inch resolution for 2,000 miles of mapped roadway is going to cost you. The best solution might be to operate in "blocks" of spatial data, similar to Minecraft. If the car fails to load the next block, it alerts the occupant, who will be forced to take the reins, or else the car will pull itself off of the road and park near the horizon of its current block until a connection with the server is restored. Naturally, for your native city, you'd just have the whole thing stored locally. For road trips, input a destination and route to download into the car before you left. You might only need the blocks solution if you're driving aimlessly, or didn't have time to input destination/route. You're always going to require a GPS connection or the next-generation equivalent sometime anyway. Sorry, just playin' devil's advocate, like I do. :)
No, that's exactly my point: a self-driving car should be a human replacement, not your GPS that happens to also control the wheel. GPS'es are loaded with data, too, whether it is pre-loaded or daily updated. The way this system seems to work is that it requires more detailed geographical data, which becomes ever more feasible in the near future. It's just a box, like a GPS, that uses a database of locational data to know where it is and make assumptions based on that data. Only now the assumptions aren't 'where should I go' but 'where do I steer this thing'. I understand that the way they do it now is a good stepping stone, but Google being Google, I am afraid that when as soon as they can they will mass-produce these cars and just make money off that. It's a reasonable business case. It is not, however, a real human-AI-like self-driving car. Which is exactly the problem. Normal humans don't depend on a colossal GIS/CAD database to find their way. A normal driver should be able to find its way reading signs and having a general idea of the route to take. I don't see why a self-driving car shouldn't do the same. If self-driving cars can operate without needing preloaded data, you don't need all that disk space at all: just lots of realtime calculations. All the car needs to know for more than a couple of seconds is where to go and the implied road speed. The rest should be handled by the car's AI. That's how I envision self-driving cars. Of course it is difficult, and I'm not saying it should be available now, but I don't see Google going down that path if this trick will put them far ahead of competitors and make them tons of money.You're always going to require a GPS connection or the next-generation equivalent sometime anyway.
and I can tell you that housing all of the spatial data, regardless of format, is a LOT of disk space.
I think this is something to strive for too, but our AI just isn't there yet. Yes, this is essentially a method of compensation... for now. So let's say that we do eventually develop a human-like perception and AI (or better) for this application. What incentive would you have to remove the CAD library functionality? You can argue that it's a crutch right now, but in the future it will be touted as a feature. Indeed, instead of having teams of people with surveying equipment romping around, the sensory data from the cars is inputted into the CAD database. Example: A car changes lanes because 700 ft. ahead (too far for sensor perception) the database knows there's some potholes in the previous lane, as perceived by the car that traveled through the area earlier, right after a heavy downpour. You're trying to implement human perception and processing in a computer (a good idea), but strip it of all of the benefits computers have over us. The two can work together for a system far superior to human-like perception and processing alone. We don't have a perfect track record, after all. And the cost of disk space? Well, we all know what direction that's headed. And... we kinda do depend on a database to find our way, it's just not 3D CAD. And of course, Google already owns the most extensive database (Google Maps). Whatever it takes to make auto-automobiles (heh) a reality as soon as possible... well, I'm for it, as long as you can maintain a failure rate lower than human error. I can understand your disappointment in the impurity/immaturity of the tech though. Appreciating the discussion. :)a self-driving car should be a human replacement
A normal driver should be able to find its way reading signs and having a general idea of the route to take. I don't see why a self-driving car shouldn't do the same. If self-driving cars can operate without needing preloaded data, you don't need all that disk space at all: just lots of realtime calculations
So am I! I'm actually sorta considering writing my thesis about this topic. They aren't mutually exclusive. The important distinction here is what the system is based on: geographical, stored data or sensory input. Google's car is based on stored data, but uses laser input to fill in the blatant gaps in its database, like other cars, people and traffic lights. What I'm mainly arguing is that I think a system based on sensory input, with additional database input, is what Google should be making, but it doesn't appear that they will do so (at least not based on what I've read and what I know about the company. This can change of course). Stored data needn't be abandoned, I just don't want my self-driving car to rely on Google's database. 'Real' self-driving cars (cough notruescotsman) are cars that can drive anywhere, on their own, just like a human being. The benefit here is that the AI can be massively improved by stored data. But I'm thinking more of Waze-like live-reporting of the road's errors, which your self-driving car can then adapt to.Appreciating the discussion. :)
You can argue that it's a crutch right now, but in the future it will be touted as a feature.
Niceeee, I've seen a few maps that you've made... excellent work, of course. And yeah, this would be a pretty fun thesis, are your other choices this cool? Well, then I think that we are in complete agreement. :) I know I've been led astray by Google Maps, and the thought of giving control of death on wheels to a similar database... Google claims impressive statistics right now, but we'll see what happens when this is slowly implemented. Most people won't want to be guinea piggin'. As you probably know, there is a large incentive for other companies (and maybe universities, I'm not sure) to develop the AI, sensory imaging, processing, code that Google then buy$. I dunno about the timescale there, but it can't come soon enough. I'm calling it now though, there will be some not-too-distant movie where someone "HAX A CAR" and commits murder via modifying database files and overriding sensory input or processing. Edit: You taught me the "no true scottman" fallacy today! Cool.I'm actually sorta considering writing my thesis about this topic.
What I'm mainly arguing is that I think a system based on sensory input, with additional database input, is what Google should be making...
It really depends if I can get a good teacher to support me on it. Usually I have to pick from a couple of themes and base my subject on that theme - this year had Sustainable Mobility as one, for example. That might be too complicated for the Michael Bay-fans! We'll probably see a Transformers rehash where the self-driving cars are the villains because they trap humans or whatever. Niceeee, I've seen a few maps that you've made... excellent work, of course. And yeah, this would be a pretty fun thesis, are your other choices this cool?
I'm calling it now though, there will be some not-too-distant movie where someone "HAX A CAR" and commits murder via modifying database files and overriding sensory input or processing.
Edit: You taught me the "no true scottman" fallacy today! Cool.
