Kilkrazy wrote:It made the point that self-driving cars will almost never be programmed to swerve because the best option in nearly any accident scenario is to brake in a straight line to avoid losing control or rolling.
The main reason to program it not to swerve or not do a controlled breaking in a towards the side of the road etc is due to not knowing how to account who has better reaction times. Let's say it could have broke to the right into oncoming or side lane, would other drivers be able to suddenly respond in an appropriate time and behavior. These issues go away to a degree once all vehicles are required to be part of the same network. When Car A swerves to avoid Pedestrian because it knows Car B and C who it could hit by swerving know how to auto correct, then you the better solution. Until you get to that point, you have to program for what you consider the least damaging or controlled area which is in front of the car.
The earlier accidents by self-driving cars, at least a good majority of them, were proven to be the driver's fault. When the car when to break or correct, the driver took the wheel, accelerated or broke instead causing the accident instead of relying on the car to do the job. It is kind of the same thing, except in this aspect everyone around it isn't part of the AI network. Once all vehicles and you could in theory even program bicycles to brake, then the network as a whole works better and is easier to program for accidents.