I completely agree. If prediction algorithms properly implemented, the Uber crash could be avoided.
I think the next step after proper visual prediction is interpreting the noise around. For example, a human can hear the ambulance siren approaching and slow down in a junction without seeing the vehicle. I think sound interpretation is very important especially in city driving.
If Uber had bothered to follow a basic safety protocol that person would still be alive today.
Staying alert on watch is something that nearly every solider, security guard, sailor, will have to do. Where you basically stare into the dark in case someone comes.
There are a number of ways to mitigate the inevitable failure of attention:
* Having a parter;
* Doing short shifts (change with partner);
* Very simple non-distracting tasks - like reading out to your parter a figure at a glance.
You certainly should not:
* Be on your own;
* Have a mobile phone on you (all screens should be locked away while moving).
It's so fucking arrogant that Uber didn't care to implement the most obvious risk mitigation. The cost of an additional person on watch is negligible compared to the cost of their self driving care program. My gut feeling is Uber simply misled regulators with how developed their self driving car program is, so it could continue to mislead investors.
I don’t think accurate prediction is the right long term strategy. It’s edge cases that cause accidents, so a more adversarial approach is safer.
Taken to the extreme this causes problems, but gently slowing down in the case of extreme tailgating is a response to slow reaction times not direct path prediction.
Adversarial approach to driving results in staying at home, because you're likely to be rammed by a bad driver from behind, and you should almost never brake and drive always as fast as possible. Alternatively always stop when you see any pedestrian on the walkway, because they will drive in front of your car.
Driving is cooperative not adversarial, even certain classes of illegal behavior are predictable.
I think the next step after proper visual prediction is interpreting the noise around. For example, a human can hear the ambulance siren approaching and slow down in a junction without seeing the vehicle. I think sound interpretation is very important especially in city driving.