[throwawayzRUU6f]

There's a large and perhaps unsolvable problem of the state of the art solutions that doesn't get nearly as much attention as it should - even if the car can safely handle 99% or 99.9% of the situations, the algorithms are not really able to detect if they're in the last 0.1%. The car will slam into stationary trucks, suddenly brake for no reason, consider the tollbooth to be a bus - all with utmost confidence it's doing the right thing.

In other words - the value of proposition of autonomous driving would be completely fine if the car gives up on unarguably difficult situations and alerts the driver to take over. As of now, however, the car will boldly do something stupid and dangerous.

[johndubchak]

That "brake for no reason" is a big one. I have a 2015 GMC SUV that has "adaptive cruise control" which I (used to) love.

Since my last service, it seems they've updated it and it's far more sensitive than it used to and seems to break in many situations where it previously did not. What sucks about that is the vehicles behind me are caught off-guard with no obvious signs of the need to break and then the jackass in the GMC SUV slams on his breaks.

[woleium]

I had a similar issue, turned out the tech had moved the sensor and it needed recalibration

[aledalgrande]

There are some models that provide the confidence that they have in the area of data being explored. I forget the name, but I think it's Bayesian Deep Learning. Something like: if it's something they've never seen, they would know they can't take a choice with confidence.

[darepublic]

If the car alerts the driver very seldomly, like once or twice in the lifetime of the car, then it could be a dangerous situation as the driver would be somewhat shocked. That or the driver would need to struggle to remain vigiliant as, day after day, the car did the right thing.