They recognize much more objects on the road but major problem is to do not cause a lot of false alarms. e.g. You don't really want to drive car which decided that plastic bag on the highway is reason for emergency braking, on the other hand is that really plastic bag or some drunk person who is sleeping on the road...

I remember when Volvo introduced IntelliSafe technology few years ago that car was totally "paranoid" and it was beeping all the time -> most of drivers deactivated that system.

If it's confused between a plastic bag and a human, it hasn't recognised it.

It didn't recognize a plastic bag as a human, it recognized a plastic bag as an unknown object. A solid object the size of a plastic bag could cause a major accident.

I own a 3 and my opinion is while people are raving over the new summon features they should instead realize that it also shows how far off autonomous driving truly is. Just go watch all the videos, all we truly have is a 60k RC car that moves slower than a RC car and still requires permission to move.

Now the current software does very well on limited access roads; highways and interstates; and has for over a year that I have owned the car. It does fine on two lane roads but the use cases there are far more limited and mostly to showing off for friends. Well that and doing stuff you normally would wait till your stopped to do.

Having actually test-driven many of these cars this year, in practice Tesla is much better than the competition. There might still be problems in different conditions, and it is not fully autonomous of course, but the common HN myth that other cars basically have the same lane assist that tesla has is just a lore. Tesla autopilot is light years ahead of the rest of them.

Look at the IIHS ratings though. The Tesla failed to detect a crossing child in both the 12 and 25mph tests. In both, it still hit the child with only a 5mph speed reduction.

Seems unlike Tesla to be out-classed like this. Six other models had significantly better speed reduction or didn't hit the pedestrian at all.

https://www.iihs.org/news/detail/performance-of-pedestrian-c...

This is definitely interesting, and it is completely possible that in some situations other manufacturers might have the upper hand.

They did test it on autopilot though, right? It says "standard" on Model 3. AFAIK autopilot features are not standard, it's an option you have to buy on Tesla. I also can't find anywhere in the report whether the car was actually driving on autopilot or not. All of these details I think make a big difference. A standard non-autopilot-equipped model 3 might just be subpar to other cars, it would not surprise me.

>I don’t know how they can claim autonomous driving “coming soon.”

How is easy: it doesn't cost anything, and makes them profit.

It's like how the had the gal to nam their system "Autopilot" when it's not, and full awareness is required from the driver in the small print...

> they’re all failing in low light conditions

Isn't this an issue with lights ?

No, because car lights only illuminate some of the surrounding area. You can't illuminate much more without causing issues for other road users (which is why there are laws around car head lamp & bulb design & usage).

Easy - just remove windows on cars so other car users are not impacted.

How good are humans detecting pedestrians in low light conditions?

It's an important test case, but I wonder how many of those would be undetectable by a human as well.

Ideally some 'low light camera' should warn you that in darkness far ahead there's 'something' unusual on the road and you will have time to react. Recently there was accident where truck with palm oil overturned on highway and first cars crashed into the truck because they were travelling too fast and only with low beam lights : https://www.youtube.com/watch?v=igX765Vw1fY

For example this camera module will be used soon in cars and it's really more sensitive than driver's eye : http://www.camera-module.com/download/sony-imx390-sensor-sel...

Adding thermal imaging to the sensor suite would probably help with a lot of edge cases, but the hardware is still very expensive. It has come down some in the last 5-10 years but there is a long way to go.

The necessary sensors are really expensive, and that's why the production cars from Tesla and GM and such just aren't that great at it.