This video convinced me that self-driving cars are definitely going to beat humans at safety. Freaky how early the thing gets that there's going to be a crash and the distance at which it responds:
At 00:28.828, a human could tell that an accident is imminent -- the oncoming car will violate the median should it not apply immediate corrective action. However, other drivers are not obligated to wait until the aggressor vehicle crosses the median to take evasive action! Indeed, the silver vehicle (ahead of the dashcam'd vehicle) begins decisive and correct evasive action at 00:28.328. That vehicle maintains the same steering angle and at 00:29.295 crosses the solid white shoulder/"fog" line -- it's unambiguously an evasive action and absolutely not a normal lane change.
The autonomous vehicle only begins to brake/turn at 00:30.130 (the first frame where the car begins to pitch forwards due to the weight transfer). The autonomous vehicle has begun to take evasive action a full 1.8 seconds later than the silver non-autonomous vehicle. Awkwardly, the Tesla's reaction seems to be triggered not by the median-violating vehicle but by the white vehicle (who is attempting to take evasive action) cutting it off. Worse, the silver vehicle is already crossing the shoulder line a full second before the Tesla begins to brake/turn. This isn't even commensurate with human performance -- the two human drivers visible in the dashcam video react significantly earlier than the autonomous vehicle.
It's absolutely possible to make vehicles that respond quicker and more effectively to impending collisions than humans do. Sensing with radar and IR can give autonomous vehicles information that human eyes simply can't detect -- IR is useful for detecting human presence, for example, and radar can be used to detect positions of cars beyond-line-of-sight. Better sensing translates to earlier warning. Also, when the computer has full control authority over individual wheel torques there's a whole new range of operating regimes that get opened up -- and autonomous vehicles that exploit those have the potential to make cars much safer (see research like http://news.stanford.edu/2015/10/20/marty-autonomous-delorea...). This sort of tactic can use the tyre grip to the absolute limit can let the computers craft the optimal (given constraints such as the vehicle's position, surrounding vehicles/barriers, and tyre/road conditions) trajectory out of a collision and execute it with precise per-wheel torque control (via the ABS/ESC system) and high-speed steering input. Even the best drift drivers don't have that level of control authority over their cars!
There's lots of room to make autonomous vehicles safer than human drivers but this video is a demonstration that autonomous vehicles can be a lot worse than human drivers.
Welcome to driving.
And yes, the car was too close. But the automatic systems don't kick in until they're necessary to avoid an imminent collision. Hopefully autonomous cars won't make that mistake.
I've done exactly that on highways and on non-controlled-access roads. It's not hard to keep a mental model of nearby vehicles (look at mirrors, if you see a car disappear from them but not go in front of you they're in some blind spot. When I know I've been tip-top at keeping watch like that, I've done evasive manoeuvres into other lanes just based on that. If I've been shoddy with my situational awareness, I do a quick head turn while doing the swerving and am ready to bring myself back into my lane should there be a vehicle in the spot I want to occupy.
Part of defensive driving is being proficient at swerving; as sharp decisive steering input can get you out of collisions that would be unavoidable via braking. Being able to swerve as fast as possible without losing control, on an open-loop basis (without needing feedback by looking at the road) and recovering is a useful skill to be confident in. I've practised swerving and thus can controllably and swiftly shift my car sideways and realign with my previous direction of travel -- it's not something I need to think about, it's as ingrained in my muscle memory as operating the pedals or the gearbox!
A few weeks ago, I used that skill to avoid hitting a pedestrian who had ran onto the road (without warning) from behind a parked car on a rainy night when I was driving at 35mph. Given my reaction time and the time needed to take the foot off the throttle pedal and onto the brake, I'd still have hit them at ~25mph even with maximum braking effort -- there wasn't enough distance. Indeed, braking would have given them more time to get in front of me, as they didn't seem to be slowing down / stopping. "Possible hit at 35mph" vs "more probable hit at 25 mph" is one horrid dilemma!
Given my (legal, albeit not so prudent) speed and how little warning I had (it was night and my driver's side window was covered in rain), when I first saw the pedestrian (through my windshield!), my immediate reaction was to swerve and not to brake. My steering input added a meter or so of space between my vehicle and that pedestrian. I did not hit anyone that night.
Self-driving vehicles should be capable of that sort of performance as well! Braking is not always the best option.
In both scenarios whether it can or cannot, its a distaster.
In fact a failure where it slows until it confirms with another sensor (probably optical) is ok. Many a driver won't even see the object and will drive though it.
You rarely will find that baby without a stroller next to it, and without a human close by hurrying towards it.
So, if you avoid hitting that stroller and that human, and, next, avoid almost hitting that human, you probably are safe. A jury wouldn't easily convict you for accidentally hitting a baby that fell out of a stroller on the middle of the road and then was left alone without any adult supervision.
I would think the difference between a solid object such and a piece of concrete or steel girder and various low-density objects (plastic bags, empty carton boxes) is harder.
On the other hand, I guess it might be easier than laymen would think to use machine learning to build a "what's the density of the object on this photo" program that is in par with humans.