The hen - egg problem of Autonomous vehicles
This Video illustrates the hen egg problem of autonomous driving when humans are involved. Granted this is not full autonomous but you get the point. There is myriads of situations like this on roads today. This even gets multiplied by differing road conditions, e.g. reflections on road when raining etc....
So either you prove to regulation that you have covered it all in training of your autonomous system or you let it run in beta mode to collect data which in turn will make the regulator nervous and may be even dangerous to people? On top, even if you prove that your autonomous system works in all instances the regulator may want to understand the explicable science behind it, which today in deep learning is only evolving.
Where does that lead to? Full mixed traffic scenarios with AV mastering all street types and all driving situations is still a good stretch down the road. And if not for technology, it may be for save and secure operational safety for all parties involved. Rolling it out too early clearly is a risk to people and technology advancement. Saying that the AV would rather protect passengers than other traffic passengers also does not help in particular. So it may be that instead of shooting for full mixed traffic scenarios right from the start full AVs will evolve as part of new transportation systems on highly controlled roads with good contextual understanding what may happen on that road. You may see variable street signs in future that allow full AV to be engaged if you still have a steering wheel or lanes that exclusively cater for AVs. As long as we have not taught AV 100% human intuition and add that to the superior driving skills of AV, AV will remain conditional.
To be fair, almost all announcements on AV coming soon are (by now) conditional or will work as highly controlled taxi systems. That and level 3 autonomous system which are now being launched, by the way, will already contribute massively to safer, more comfortable and efficient rides.
Founding Director of Aalberg, Bauer & Wilson Konsult - Strategy development, tactical implementation, leadership skills & team performance improvement. Identifying business & reputational risks and mitigating actions
6 年There are more stake-holders involved here, not just the Regulator, Passenger, Driver/owner but also the pedestrian, the Insurance companies, the traffic police etc. There is still no agreement as to on whom the responsibility will fall if their is an accident. Will the AV always be deemed to be not-guilty, therefore the victim, even if someone outside the vehicle dies? Will the AV always be programmed to protect its passengers rather than the group of children on the pavement (US=sidewalk)?? That might be okay for the insurance industry but not for the police and road safety regulators. Still too much to be agreed for there to be 'safety' within the AV L4/L5 environment.
Automotive Electronics Professional
7 年The question of how to validate deep learning systems is justified and not trivial. A simple answer would be statistics. Proof that it works well for 100 m miles and you have reason to start believing in it. For me that would be enough proof to deem it usable.
Resulting vs Reporting "A nail is not discussed into the wall"
7 年How could we (human people) ever lesen driving a car with about 30-40 driving Lessons on the road and a few more hrs in theory for the rules......??? And my wife can not only adapt to traffic, she also adapts to the car. The more hp under the hood, the faster shooting along the left lane.
Agricultural journalist specialising in farm machinery
7 年Or you just let people get on with their lives rather than trying to screw them for more money by pretending that AV's are the answer to anything in particular.