Autonomous Car Progress

[daniel]

Tesla still calls my EAP "beta," but used as intended it is damn good and makes the car safer than if the car did not have it. That's the threshold for me: When a feature makes the car safer than it would be without it, it's ready for public release.

I fear I may never be able to buy a car that lets me sit in the back seat and take a nap while it drives me where I want to go, but I sure am glad Tesla lets me use the "beta" EAP.

 

[Daniel in SD]

Tesla still calls my EAP "beta," but used as intended it is damn good and makes the car safer than if the car did not have it. That's the threshold for me: When a feature makes the car safer than it would be without it, it's ready for public release.

I fear I may never be able to buy a car that lets me sit in the back seat and take a nap while it drives me where I want to go, but I sure am glad Tesla lets me use the "beta" EAP.

Just curious, do you think it's safer than lane departure avoidance, forward collision warning, and automatic emergency braking? Why?

 

[powertoold]

Mobileye only releases stuff when it's ready, Tesla throws out alpha software for customers to test.

lolllll are you kidding? Have you seen how horrible the lane keeping is with current mobileye equipped cars? It’s way worse than tesla.

edit: sorry, I can’t confirm if the 2020 Honda sensing is using mobileye or not

edit 2: it’s actually difficult to find any videos of confirmed mobileye lane keeping tests. My guess is that it sucks so bad no one cares to test it (considering they have 50 bajillion mobileye cars out there). If you disagree find me a video of a consumer production mobileye equipped car running a mountain road like AP current can. Or any lane keeping stress test.

 

[banned-66611]

Lane Keep Assist is just to stop you going over the lines.

Lane Follow Assist actually keeps the car in the middle of the lane. Two different things.

I've tried the latter in a Honda, it works great.

 

[powertoold]

New Lane Keep Assist is just to stop you going over the lines.

Lane Follow Assist actually keeps the car in the middle of the lane. Two different things.

I've tried the latter in a Honda, it works great.

Everything in this response is wrong, except for your opinion, which isn't supported by a video.

What is Honda Sensing® Suite? Features & More | Honda

 

[daniel]

Just curious, do you think it's safer than lane departure avoidance, forward collision warning, and automatic emergency braking? Why?

EAP includes all those things. So yes, it's better than just those things alone. The same way a pizza and a carton of triple-chocolate lactose-free ice cream is better than just a pizza alone.

Note: I don't know how well the Mobileye system works, but I cannot buy a car with the Mobileye system. It will surprise me if another company does not beat Tesla to true driverless autonomy by using Mobileye technology, but until then, Tesla is selling the best driver-assist systems we as ordinary consumers can buy.

Lane Keep Assist is just to stop you going over the lines.

Lane Follow Assist actually keeps the car in the middle of the lane. Two different things.

I've tried the latter in a Honda, it works great.

That's good to hear. I've never driven the Honda so I cannot compare the two. I will note that if it keeps the car in the middle of the lane then it has the same shortcoming that Tesla has: There are times when the middle of the lane is not where you want to be. A driverless car needs to know when to hug one side of the lane or the other, or even depart from the lane. Tesla has not even begun to address this, AFAIK.

 

[daniel]

What is Honda Sensing® Suite? Features & More | Honda

This looks like the Honda does not steer the car. It looks like it just notifies the driver if he departs from the lane. And it looks like it only operates between 45 and 90 mph. And doesn't operate with the windshield wipers going.

So nothing at all like Tesla's AP/EAP.

 

[powertoold]

This looks like the Honda does not steer the car. It looks like it just notifies the driver if he departs from the lane.

It does autosteer, just horribly. Do some more research into the Honda LKAS

 

[hgmichna]

And it looks like it only operates between 45 and 90 mph. And doesn't operate with the windshield wipers going.

So nothing at all like Tesla's AP/EAP.

And Honda requires lane markings, which the Tesla autopilot does not (if the road is wide enough and the implicit lane is free of obstacles like parked cars). No comparison, really.

 

[hgmichna]

I will note that if it keeps the car in the middle of the lane then it has the same shortcoming that Tesla has: There are times when the middle of the lane is not where you want to be.

I agree. In fact, the middle of the lane is rarely where I want to be. When I steer manually, I constantly consider distances to other cars and obstacles and keep optimizing the lateral position.

In this respect, the Tesla autopilot is no better than the others. But if you are willing to forgo this optimization, the Tesla autopilot steering is often less than optimal, but still good enough. Occasionally I have to take over and steer away from other traffic, the most obvious example being an oncoming truck driving on the dividing line on a narrow two-lane road.

 

[diplomat33]

And Honda requires lane markings, which the Tesla autopilot does not (if the road is wide enough and the implicit lane is free of obstacles like parked cars). No comparison, really.

AP does require lane markings to initiate lane keeping. So if it starts with lane markings, it can continue driving even when lane markings disappear. But AP cannot initiate on a road with no lane markings. There are plenty of residential streets and campus streets where I cannot use AP.

 

[hgmichna]

AP does require lane markings to initiate lane keeping. So if it starts with lane markings, it can continue driving even when lane markings disappear. But AP cannot initiate on a road with no lane markings. There are plenty of residential streets and campus streets where I cannot use AP.

Thanks for the correction. I sometimes could not enable the autopilot and I already suspected the lack of a dividing line, but it wasn't clear to me. I will keep testing though. Perhaps the autopilot can be started if the road is wide enough.

In any case, it seems the autopilots in other cars are less capable.

It seems to be difficult to find direct comparisons or even just demos of other cars' autopilots under difficult conditions. I only see the occasional fair-weather demo in very light traffic. Perhaps the Tesla fans rarely drive other cars while fully exploiting their autopilot, and the drivers of other cars don't know the Tesla autopilot.

 

[powertoold]

In any case, it seems the autopilots in other cars are less capable.

Let's not kid ourselves, AP is *by far* the most capable and reliable consumer automation system.

If anyone doesn't think so, feel free to post a video.

There's a reason why there's 20+ youtubers dedicated mostly to stress testing AP, and essentially none doing it for any other car.

 

[Daniel in SD]

EAP includes all those things. So yes, it's better than just those things alone.

Better, sure, I like it better too. But safer? I'm not so sure.
The way I look at it is when you use automatic emergency braking, lane departure avoidance, etc. the car is monitoring you. When you use Autosteer and TACC you are monitoring the car. Humans are currently better than machines at controlling the car so it seems like there are fewer errors to correct when controlling the car myself. I have to intervene using Autosteer and TACC way more than the car intervenes when I'm controlling the car myself.

 

[Eno Deb]

EAP includes all those things. So yes, it's better than just those things alone. The same way a pizza and a carton of triple-chocolate lactose-free ice cream is better than just a pizza alone

That doesn't really compute, since we have seen a number of accidents that wouldn't have happened without Autopilot (e.g. crashes into stationary objects).

The overall effect on safety is impossible to judge without proper statistics. On one hand, I'm sure that Autopilot prevents some fender benders in stop-and-go traffic (which often happen because the driver isn't paying attention) and perhaps by reducing fatigue on long highway drives. On the other hand, the car on Autopilot sometimes shows unexpected behaviors (such as sudden swerving when lanes split or merge, phantom braking, or drifting dangerously close to trucks in neighboring lanes) that may cause accidents, and may contribute to accidents for drivers that don't understand its limits and have a false sense of safety.

Tesla could easily shine some light by releasing honest statistics, i.e. accidents per mile driven with and without AP on the same type of road.

 

[DanCar]

In order for Tesla statistics to be more believable as previous poster indicated, we would need to know lots of details. For example the crashes into stationary objects: did those count against AP or were they not counted because AP disengaged 10 ms before accident happened?

 

[hgmichna]

That doesn't really compute, since we have seen a number of accidents that wouldn't have happened without Autopilot (e.g. crashes into stationary objects).

The overall effect on safety is impossible to judge without proper statistics. On one hand, I'm sure that Autopilot prevents some fender benders in stop-and-go traffic (which often happen because the driver isn't paying attention) and perhaps by reducing fatigue on long highway drives. On the other hand, the car on Autopilot sometimes shows unexpected behaviors (such as sudden swerving when lanes split or merge, phantom braking, or drifting dangerously close to trucks in neighboring lanes) that may cause accidents, and may contribute to accidents for drivers that don't understand its limits and have a false sense of safety.

Tesla could easily shine some light by releasing honest statistics, i.e. accidents per mile driven with and without AP on the same type of road.

Autopilot driving into large, bright, unstructured, stationary surfaces is a special case. Such accidents are utterly rare. I have heard only of two. Apparently the autopilot has a problem with these, possibly because the radar cannot differentiate stationary objects from the background, and the cameras have problems recognizing large, unstructured, bright objects. Some work may have to be done in this respect.

As to the unexpected behaviors you mention, I don't like these either, but I don't remember any that came anywhere close to a collision. A Tesla may be "swerving when lanes split or merge", but never dangerously so. It never swerves into another car.

My Tesla is not "drifting dangerously close to trucks in neighboring lanes" either. Quite to the contrary, I only observe trucks drifting dangerously close to and sometimes into my lane. In these cases the Tesla brakes and, if possible and necessary, steers away. Normally Teslas steer fairly precisely within the driving lane. They don't drift. I cannot confirm the descriptions you wrote.

I would also welcome good statistics. I hope we will see some over time. My feeling is that my Tesla is far safer than any car I have ever driven, and I hope the statistics will bear this out.

 

[powertoold]

Tesla autopilot is the only autosteer system that I'm aware of that nudges away from large trucks while lane keeping.

Just another example of how much better Tesla is vs the competition.

 

[daniel]

It does autosteer, just horribly. Do some more research into the Honda LKAS

Thanks for the clarification. I don't feel the need to research it further since I know from their video that it would not be an improvement over my Tesla Model 3.

Let's not kid ourselves, AP is *by far* the most capable and reliable consumer automation system.

^ This!

Better, sure, I like it better too. But safer? I'm not so sure.
The way I look at it is when you use automatic emergency braking, lane departure avoidance, etc. the car is monitoring you. When you use Autosteer and TACC you are monitoring the car. Humans are currently better than machines at controlling the car so it seems like there are fewer errors to correct when controlling the car myself. I have to intervene using Autosteer and TACC way more than the car intervenes when I'm controlling the car myself.

Yes, I believe it to be much safer, with one caveat: When used as intended. See below

That doesn't really compute, since we have seen a number of accidents that wouldn't have happened without Autopilot (e.g. crashes into stationary objects).

As far as I'm aware, accidents while AP is engaged have all or nearly all been the result of drivers neglecting their responsibility to keep their frigging eyes on the road and intervene when the car is obviously unaware of an obstacle.

And we get no statistics about the accidents prevented when a driver is momentarily distracted. But I know that I'm more rested and therefore more alert when using EAP. It's not great at all in the city (LKAS doesn't even function at city speeds) but on the highway the only time I have to intervene is when there's a red light and no car stopped ahead of me. Since Tesla says AP/EAP is intended only for highway use, other than red lights, I never have to intervene when using it where Tesla says to use it. ("FSD" is another matter entirely, and one I have no experience with because I don't see much utility in it at present.)

 

[Daniel in SD]

Yes, I believe it to be much safer, with one caveat: When used as intended. See below

If automobiles were “used as intended” there would be 90% fewer accidents so I don’t think that’s realistic. Haha.