Prius Autonomous Driving

I believe that the real problem is too many people on the planet. We seem to have already exceeded the carrying capacity of Earth and this underlying problem seems to be almost entirely ignored.

 

well, they'd better die, and decrease the surplus population

 

Retroactive birth control

 

volunteers gladly accepted. in oregon.

 

DARPA_Grand_Challenge_(2005) was not a technology. It was a competition to achieve a defined objective using any technology.
The 2005 winner was not part of any "actual street" test. That was the subject of the DARPA_Grand_Challenge_(2007).

 

going to give a whole new meaning to "blue screen of death".

Really though, building a safety critical software system that passes all the multitude of certification requirements is going to be tough to do. Basically impossible for a home-brew person (you have to test every conceivable code path in your system, and using open-source libraries is almost impossible to do this with since you will have had no experience with the source and need to know how to force it to go down exception paths in the code). you're choice for operating system is severely restricted... Linux is out for right (not life-safety qualified) so you're down to a purpose built os. Then you also have all the image-processing libraries...can't use any open-source since, again, they haven't followed the safety critical development standards processes.

you can certainly home brew up you're own solution, but anyone selling it or giving it away, let alone using it is taking a huge liability risk. Safety critical software development is a huge pain in the nice person! I've done it, and don't like it!

 

I don't currently give a rip about 'possible'. Multiple bigger projects already demonstrated the self-driving POC on public streets and roads.

The essential POC target now is to demonstrate that they can get the fatality rate down to the required level, and they are not there yet. Waymo currently has 12 million miles (extrapolating from 10M benchmark in October & 1M/mo rate) with no fatalities. Uber has 3M with 1 fatality. After reviews (and problems found) following that fatality, Uber just got clearance to restart testing.

Until either the industry, or a major player, can demonstrate the required safety level, I'm not giving anyone any regulatory slack. Small hackers trying to make aftermarket COTS kits must meet the same safety standards and undergo the same reviews and oversight as the bigger guys.

Only after somebody meets the safety target, will I entertain the exploration of lower cost, lesser regulated paths to that same target.

Put another way, I want to see any guerilla hacker 'cheaters' that produce any traffic deaths, incarcerated on intentional homicide charges.

 

I have 20k miles with comma.ai doing 98%+ of my long distance freeway driving. It is just a lateral extension of cruise control. With a cruise control that can handle both longitudinal and lateral control of my car, it takes much of the workload of driving off of me. This makes driving long distances more enjoyable and relaxed, but my brain cannot be disengaged from driving the car. It is only an extension of cruise control technology.

I no longer care about fully autonomous. Semi-autonomous is working well enough right now. Being physically more separated from driving the car makes it much easier for me to hypermile behind a semi in the slow lane, because I always allocate more time for travel than I need, so I am not suffering from an artificially induced deadline to reach my destination.

Since I have altered my behavior due to the application of this technology, it would tend to appear that technology has hacked me.

 

for most of the driving public, fully autonomous would be much safer. most people are asleep at the wheel these days, without any computer assistance

 

the issue is that these things work 99.999% of the time, but that one time, say when a little kid darts out from behind a parked car, and your only choice is to run him over or to purposefully crash into the parked car to stop your forward momentum, what do you think these kits will do? i'm not going to be fully happy until they're built using ai techniques (both statistical and neural) and trained up on various situations in much the same way a human would be.

Now... that's not saying that people are any better. You have the typical Fight, Flight, and ARGGGG response possibilities. I've seen more than enough people who just freeze up and let their brain turn off and take whatever comes. But that doesn't can't insist on better from the machines we build.

 

humans do it with only two camera inputs... that's theoretically all that should be needed for a vehicle as well.

Until cars can recognize that the 3 year old walking on the sidewalk with their 90 year old grandparent (example of someone with very slow reaction times) may suddenly dart in front of the vehicle with no warning, and have the vehicle make the informed decision to veer hard in the other direction, even though this may result in a head on collision with another vehicle, I wouldn't trust full autonomy.

It's relatively easy to write code to cover almost all of the normal driving conditions (e.g. lane keeping, navigation, intersection management, etc.). You can load up a car with a crapload of sensors to cover the entire area to allow the vehicle to acquire a fairly precise state of the world around it from a pure object/kinematic point of view and from there extrapolate a safe path through. It will get much harder, however, when you get non-typical situations (e.g. an accident scene where you have a bunch of rescue vehicles parked in all different directions, with various people directing traffic into non-ordinary situations). There will be times when it simply can't drive itself and require human intervention.

But my biggest fear, however, is that the systems will need to understand psychology of what other drivers may do that is improper and what should be done to avoid accidents in anticipation of those events. We've all had situations where we've thought to ourselves: "that idiot is going to...") and have taken preemptive action to avoid an issue because we understand human nature. An autonomous vehicle can't do that without some serious, serious artificial intelligence that is way beyond what we're capable of today (and will be for a while).

And if you have something that's good for 99% of the time you'll get complacent and allow yourself to be distracted to the point where you won't be able to react in those situations and deaths will occur.

Now, it certainly may well be that an autonomous vehicle will cut deaths down and while corner case deaths will go up, normal driving deaths will go down so that the overall benefit will be there. So it may be worthwhile to do. I'm not saying it wouldn't be, I just hope people keep an open mind about what these systems can and can't do.