Outbreak of sanity:
Fully autonomous cars may never reach public roads, according to the chairman of the US National Transportation Safety Board.
Speaking in an interview with MIT Technology Review, Christopher Hart said: “I'm not confident that we will ever reach that point. I don’t see the ideal of complete automation coming any time soon.”
...
“Some people just like to drive. Some people don't trust the automation so they're going to want to drive. [And] there’s no software designer in the world that's ever going to be smart enough to anticipate all the potential circumstances this software is going to encounter,” Hart said.
It's almost like the NTSB reads
Borepatch.
Something nobody else seems to be saying is the feeding frenzy this will be for lawyers.
ReplyDeleteTake your rev 1 autonomous car on the road and it makes a mistake and hits someone. Can you see the lawyers lining up because of the deep pockets involved? Are there any deeper pockets than Google's? I can see the lawyers suing the car maker, software companies, everybody they can think of.
Finally!!! Geez... Unless you're going to put a RAID setup in the car, it's going to get the blue screen of death, and GPS is only accurate to about 100 feet without differential antennas, regardless of what the mooks say.
ReplyDelete[And] there’s no software designer in the world that's ever going to be smart enough to anticipate all the potential circumstances this software is going to encounter,” Hart said.
ReplyDeleteIt's not even about "smarts". You can code only for what you can imagine or what you have experienced, and it's a safe bet that no software designer in the world has the driving experience to foresee every possible scenario, in every possible weather/road condition. Not even a professional DRIVER has the driving experience to anticipate every possibility. It's almost literally an infinite spectrum.
Combine that with technical limitations on computer sensors (say, the kind that cause the computer to interpret a big white truck as a distant cloud and disregard it), and we're facing a no-win situation. There's literally no way to do this and meet, let alone exceed, human judgment on the road.
Writing this, it occurs to me that possibly the BEST thing the software designers could code into self-driving cars is a fail-safe such that when the computer encounters too many variables beyond its programming, it will pull over to the shoulder, turn on the hazard lights, engage the parking brake, and sound an alarm to wake up the driver.
Auto pilots for aircraft took years to develop and came around after instrumentation landing systems, and there is MUCH less to run into in the sky than on the ground. Major aircraft makers have been working on autonomous aircraft 'see and avoid' systems for years now and they haven't come up with something good enough to satisfy regulators - Goodly, Apple, and Tesla, despite all of the hype, are VERY new to the unmanned game and are companies who do NOT have a history of rigor in their main lines of business; why would anybody expect them to have rigor in a line of experimentation?
ReplyDeleteCarnegie Mellon has had a self driving car on the roads since 1991 and doesn't think enough of it to sell it to the public and they DO have a reputation for rigor!
I agree - it will be YEARS before self driving cars are common, and the simplistic, short cutting, attempts of Google, etc. will ADD years to the process!
Personally I suspect that we will begin to see autonomous only car lanes which will likely derive from HOV lanes. As more and more people use autonomous lanes, more lanes will be built/ converted until we see entire toll and regular roads that become autonomous only. This will just be for long distance known route, seperate lane driving to begin with, but eventually most surface streets will be autonomous only.
ReplyDeleteThe real problem with self driving cars, is trying to get them to fit into society as a whole. You try to fit autonomous cars into human driven traffic and you will always have issues.
ReplyDeleteTackle the problem from another route, standard autonomous car to car communication, ranging, and data interfacing. Dedicated road space, IE lanes for auto driving cars, perhaps even dedicated roads. Telemetry information built into the roads. The issue really is that everyone is trying to solve the problem the wrong way