Highway merging rant
I guess I’ll never stop being amazed how most people merge onto highways from on-ramps. It practically never fails – here I am almost exclusively in the right hand lane, where I’m legally supposed to...
Forum Discussion
ok, time to get down to brass tax on this computer driven car subject.
As someone who has used and written AI programs, here is my insight (FWIW)
There are a couple of very real and very serious limitations to this technology.
Of course, programming, hardware, and OS reliability are a must. In today's world the quality mostly stinks. Resolvable, but requires real effort and *real* programmers, not just OO scriptors ;)
Interraction with other objects, which may change their location, speed, and other parameters is the biggest challenge. It isn't like a UAV where you usually have 'lots' of space available around you.
Assuming you can have a perfect prediction and resolution routine (no small challenge!), you still have the problem of asynchronous event and action cycles.
Simply put, unless you establish a fixed timing (think pulse) for a given path, even the best systems will fail when Vehicle A makes an adjustment in the middle of Vehicle C's Decision-Action process. If they are both operating on the same timing, a conflict can be resolved.
Think of it as mid-cycle processing. If you have a rogue process that makes a change on the data you are currently processing, you cannot stop until you complete one cycle, whatever that may be.
Inside a computer, these items can be handled in a number of ways, including interrupts (that interrupt a process/cycle for some reason)
Simply 'Navigating' by computer is easy. Sit in your car and sleep. But actual 'Driving' where the vehicle must anticipate and adjust to weather, traffic, other drivers, other automated cars, system failures, timing failures, logic failures, and many more items is a very different story.
Without an electronic roadway with OOB hard limits (like a CNC), we are a REALLY LONG way from having a reliable driverless car on standard roadways, intermingled with standard traffic and drivers.
This is a very brief comment that barely touches on a few issues I wanted to point out. There are a lot of others.
And if you are thinking of adaptive programming, I've had my share of experience with authoring and using those, too. They can be very good, but aren't the holy grail, either.
Besides, aren't we talking about California? I seem to recall reading about efforts they were making to allow flying cars, back when the tri-engine version was announced by...was it Mueller? It's late and I don't feel like looking it up.
Hype isn't always reality ;)
As someone who has used and written AI programs, here is my insight (FWIW)
There are a couple of very real and very serious limitations to this technology.
Of course, programming, hardware, and OS reliability are a must. In today's world the quality mostly stinks. Resolvable, but requires real effort and *real* programmers, not just OO scriptors ;)
Interraction with other objects, which may change their location, speed, and other parameters is the biggest challenge. It isn't like a UAV where you usually have 'lots' of space available around you.
Assuming you can have a perfect prediction and resolution routine (no small challenge!), you still have the problem of asynchronous event and action cycles.
Simply put, unless you establish a fixed timing (think pulse) for a given path, even the best systems will fail when Vehicle A makes an adjustment in the middle of Vehicle C's Decision-Action process. If they are both operating on the same timing, a conflict can be resolved.
Think of it as mid-cycle processing. If you have a rogue process that makes a change on the data you are currently processing, you cannot stop until you complete one cycle, whatever that may be.
Inside a computer, these items can be handled in a number of ways, including interrupts (that interrupt a process/cycle for some reason)
Simply 'Navigating' by computer is easy. Sit in your car and sleep. But actual 'Driving' where the vehicle must anticipate and adjust to weather, traffic, other drivers, other automated cars, system failures, timing failures, logic failures, and many more items is a very different story.
Without an electronic roadway with OOB hard limits (like a CNC), we are a REALLY LONG way from having a reliable driverless car on standard roadways, intermingled with standard traffic and drivers.
This is a very brief comment that barely touches on a few issues I wanted to point out. There are a lot of others.
And if you are thinking of adaptive programming, I've had my share of experience with authoring and using those, too. They can be very good, but aren't the holy grail, either.
Besides, aren't we talking about California? I seem to recall reading about efforts they were making to allow flying cars, back when the tri-engine version was announced by...was it Mueller? It's late and I don't feel like looking it up.
Hype isn't always reality ;)
