Gupta_amit

Last weekend I was in adaptive cruise control in very slow and heavy traffic on the highway. At one point, the car in front of me moved over to the right lane, and by this time the next car in front of me had come to a complete stop (50 meters ahead). Given “adaptive” cruise control, I expected my car to accelerate just a little to get closer to the next car in front and then decelerate gradually and eventually brake if the car in front of me had not yet moved.


Instead, my car accelerated very aggressively, as if there were no cars in front of it and as if it was accelerating to reach the set cruise speed of 50 mph. I waited a second or two to see if it slows down, but it didn’t and continued accelerating, so I had to apply brakes aggressively to slow down. This was a very scary experience and I felt that the sensors or the algorithms of adaptive cruise control had failed for some reason in this case. Of course my wife sitting next to me told me that’s why she doesn’t trust these sensors and things.

I continued to use cruise control for the rest of the trip and did not have a repeat experience after that and forgot about it until I read this article a couple of days later.

https://www.wired.com/story/tesla-au...y-crash-radar/

This is when I realized that this is essentially an issue with all such current systems and it manifests in very specific situations when a stationary vehicle appears in front, after the vehicle in front of you moves away to the other lane. It basically ignores the stationary vehicle in front (considers it part of landscape such as highway dividers or overhead signs) and does not consider it as an obstacle that it has to adapt for. Very scary and as per the article this limitation impacts all adaptive cruise controls in use today.

Another reminder to not get distracted when using adaptive cruise control or other semi-autonomous systems....results can be dangerous.

SCarGuy

all depends on what settings you have in both MMI (under "Vehicle - ACC" and, what distance you are asking it to follow, since it defaults to distance before speed.

The system is not there to be autonomous, and it will only do what it is being asked to do. The onus is still on the driver

njspeedfreak

I purposely left DAP off my build sheet. I build software for a living, and while I think these semi-autonomous systems are very very good, they are not perfect and probably never will be without some huge advancement in AI technology. The car can only determine what's going on based on the inputs to it's sensors and the algorithms that the software designers built. And those algorithm's are going to be limited by the number of possible situations and combinations of factors that the designers originally planned for in their design.

One thing they will always lack is the ability to predict future events based on non obvious clues. For example, you see cars in the distance with their brake lights on and you can anticipate there may be some road hazard ahead. Or you see someone behind you and you start to prepare for the possibility he will cut you or a nearby driver off. Or you see some kids playing near the street and not paying attention to traffic and you prepare for the possibility that one of them will jump into the street chasing after an errant ball. Those things are going to be near impossible to encode in an algorithm.

Now the car may be able to REACT to something much faster than you can, but sometime that's not going to be enough. The best that can be said about these systems is that they never get lazy and never get distracted, but the problem is they will make us lazy and allow us to be more distracted as a result. That's the other reason I did not get DAP. I was worried that I would fall into that trap and let my mind wander or pay more attention to the phone instead of the road. And eventually that would lead to a situation where the car couldn't cope with a situation and forcing me to take over. But unless I was already paying full attention, I would have no situational awareness and no ability to react in the necessary time.

I would love to see the day where truly autonomous transportation comes along, I just don't think we are there yet and I think semi-autonomous can actually increase risk.

superswiss

Yes, this is something that everyone with ACC should know. It's explained in the user manual. It's not an issue specifically to ACC per say, but ACC uses radar and radar is not very good at detecting stationary objects. It's very good at detecting moving objects. Radar is based on the doppler effect. ACC therefore is a system that monitors moving objects ahead of you and slows down if they slow down, but it won't stop for walls or other stationery objects. In order for ACC to slow down it has to first see an object move and then see it slow down. If a stopped car head of you appears out of nowhere, it won't get detected. It really pays to read the owner's manual and fully understand the limits of these systems. No reason to panic, but a reminder that there's a learning curve to everything and ACC is merely an assistance feature. At no time does it relieve the driver from paying attention and manually intervene if necessary.

Chris

I don't know anything about programming. But I DO know that human beings are unpredictable, and do stupid, unpredictable things. I can't imagine any autonomous system would be able to compensate for that. So as long as humans are unpredictable and stupid ... lol

superswiss

These systems are meant to be used as tools. I build software for a living, too, and I use technology simply as a tool to let me complete a task more efficiently than I could on my own. I've used ACC for the past 5 years. It hasn't made me more lazy, on the contrary. It has made me less fatigue on long journeys and therefore more alert, which is rather a good thing.

benz240

njspeedfreak

I would imagine it would, but that could create a new problem. The sudden braking that prevents you from hitting the car in front might cause the guy behind to hit you, especially if he doesn't have such technology.

ArvX147

Uh, radar is not good at detecting objects that are stationary relative to its position. If the radar is moving (ala mounted to car) its just as good at detecting a stationary object as a moving one.

superswiss

That's in principal true, however, it wouldn't be able to distinguish between a fixed wall along a road, for example in a turn or a stationary car. Would have way too many false positives if stationary objects would be included in the algorithm. Needs cameras or something to figure where the road goes and what exactly the detected object is to decide if it is a "threat". But even cameras are not a 100% because they can be blinded by direct sunlight as some of the Tesla accidents have demonstrated, so a full area of radar, lidar and cameras would be needed for starters to properly map and recognize the surroundings.