Discussion in 'Clarity' started by Clarity_Newbie, May 9, 2019.

  1. Sandroad

    Sandroad Well-Known Member


    The CMBS does detect and respond to some stationary objects. I don't know what the algorithm is, but I have had my Clarity detect and warn me with BRAKE in the instrument cluster and shake of the steering wheel in some stationary object circumstances, including a large road sign in a sharp right hand curve that is on a regular route of mine.
  2. Clarity_Newbie

    Clarity_Newbie Active Member



    In this case...the Clarity had clear sailing ahead as it were. I believe the sudden appearance of an object regardless of position relative to the Clarity activated the CMBS...which it is designed to do as I understand it. worked this go-a-round.
  3. rodeknyt

    rodeknyt Active Member


    The Doppler effect, which is what radar uses, doesn't depend on movement of two objects—only movement by one toward another is necessary. That's how radar speed signs (and police radar) calculate your speed (only one object, your car, is moving). Yes, I know there is "moving radar" that is used by police to calculate speed of vehicles when everyone is in motion. Those simply determine the difference in speed between the two vehicles and add back in the speed of the car that carries the radar. In the case of a car moving perpendicular to another, the one moving toward the (broad)side of the other will sense the object briefly while it's within the "field of view" of the radar unit. In that case, how long the perpendicular car is in that "window" will determine whether a warning is issued and the system reacts.
    KentuckyKen likes this.
  4. MrFixit

    MrFixit Active Member

    This is not entirely true. The RADAR used here is undoubtedly very basic. Your vehicle is moving, and therefore yes, stationary objects (like the road, signs, trees, etc) will have a doppler. The trouble is that the signal from this stationary "clutter" is very large. A stopped vehicle will likely not be separable from the stationary clutter, and therefore the system will probably blank out the vehicle along with the stationary clutter. It can likely only detect another vehicle that has a different doppler relative to the stationary clutter in order to process it. ie: only objects that are moving relative to the road.
  5. rodeknyt

    rodeknyt Active Member

    Roadsigns, trees etc. are farther away than a car on the road in front of you. If the system can't discern a stopped car directly in front of you, then it is not actually using radar.
  6. MrFixit

    MrFixit Active Member

    This is a function of how far away, the angular displacement (antenna pattern), the size of the objects (RADAR cross-section), and many other factors. The designers must be very careful to maximize performance while minimizing false alarms. At best, detection of a stationary vehicle is desensitized by the many other stationary objects. There are LOTS of stationary objects that will be farther away, but cause a greater radar return (like a metal overpass). These must be somehow discerned without being confused as a vehicle. It is much easier and more robust to detect other vehicles that are also moving with respect to their surroundings. Remember the incident last year where a Tesla ran directly into a stationary firetruck? Here is a link to an article about that Tesla incident:

    I am not saying it is impossible to detect a stationary vehicle with radar. It is just much more difficult to do it reliably, and thus I think there is a caveat about the performance of the CMBS systems in this regard.
  7. JCA

    JCA Member

    I think this gets into the debate over whether self driving can be achieved with just radar and cameras, or if LIDAR (laser range detection that can be pinpointed to the area in front of the car and thus can better differentiate stopped obstacles vs large fixed objects that are just nearby ) is required. Tesla claims they can do it without the more expensive LIDAR; other companies in the field disagree. And Teslas have crashed into stopped objects (and into trucks moving sideways across the path, which is similar from a radar detection point of view).

    Humans aren't perfect by far, and there's a lot of room for automated driving to compensate for that and do better in the long term, but human perception and intuition when paying attention is actually really good and hasn't been exactly replicated with camera, radar, or LIDAR for that matter technology yet.

    I notice that when using ACC -- when I notice the driver in the next lane in front of me start glancing and looking like they want to change lanes, I realize that I subconsciously either back off to give them a clear obvious lane change, or I speed up a bit to get out of their blind spot (where I try not to drive in the first place) so they know to wait. It makes me uncomfortable and want to take over when the ACC doesn't do the same, but waits until the car is most of the way in my lane before slowing abruptly. But of course I know the systems just aren't that sophisticated and use them accordingly.

    In the future, vehicles that communicate with each other and negotiate maneuvers, let each other know when stopped etc, will vastly increase safety, but until then a combination of humans and computers driving is going to be an interesting mix.
  8. 2002

    2002 Well-Known Member

    LIDAR is affected by weather, so Musk might be on the right track. Using multiple radar detectors with sophisticated enough software could possibly, eventually approximate the capability of LIDAR, maybe not quite but it would work in all weather. Then again maybe multiple LIDAR detectors at different wavelengths can mitigate weather somewhat. Of course running both LIDAR and radar together would be ideal but currently cost prohibitive.

    The Wired article is one of the rare examples of realistic pessimism about the timeline. I have just about given up on telling people that I think fully autonomous driving is at least a decade away because everyone is convinced it's just around the corner. Just like twenty years ago everyone thought that fully developed artificial intelligence in our everyday lives was just around the corner.

    You don't even need to see the driver, just the suble motions of the car in the next lane including changes in speed are good indicators, combined with an awareness of that driver's situation, i.e. their lane is moving slower that yours, someone changing a flat tire up ahead of them, etc.

    I don't know of any system that looks through the windshield of the car in front of you to notice what the car in front of them is doing, much less looking even farther ahead to see what traffic is doing. The current systems are good at helping us to keep from hitting the car directly in front of us, sometimes even faster reacting than even an attentive driver. Everything else however is really only useful as a backup when a driver is not paying attention. For example I estimate that my LKAS detects the lane markings about 75% of the time. Which means if I got distracted, there is a three out of four chance that LKAS will keep me from going out of my lane. That is certainly better than not having LKAS, but it's far short of allowing me to do something else even for a brief period of time.

    The real dilemma about the continuing improvements in these systems is that the better the systems get the more people start to become less concerned about looking away from the road for several seconds. That's when the accidents occur that we keep reading about.
  9. MPower

    MPower Active Member

    Some of the radar detectors would have to be much higher than our current one. A little drive in the slush and snow will demonstrate that radar is affected by weather. (Pull over and wipe off the big H.)
  10. JulianClarity

    JulianClarity Active Member

    The brake on Clarity works really well.
  11. 2002

    2002 Well-Known Member

    Correct, I was referring to the potential of radar compared to what is currently used on cars like Clarity by using upgraded hardware, multiple sensors, and more robust software. Even what is currently on Tesla will be out of date in a few years as upgrades and improvements are implemented. Musk is going for full autonomous, which I doubt his timeline but any improvements made towards autonomous will trickle down to the safety features that are designed to assist an active driver.
  12. rodeknyt

    rodeknyt Active Member

    And I’m sure you’ll find that the radar used in these systems is looking in a rather narrow field ahead at street level and not up in the trees, up at overpasses or to the signs at the side of the road. These things (on all brands of vehicles) are designed to see what is directly in front of the car. If it isn’t seeing a stationary object then it isn’t designed or working properly.

    I get an alert every time I come up behind a stopped car when the system doesn’t think I’ll stop in time. I don’t wait to see if the car will stop itself in time, but if it didn’t that would indicate a problem with the algorithms involved in emergency braking, not with the radar.
  13. Danks

    Danks Member

    On my way home Monday afternoon I drove this stretch of road. When I approached the narrow part the CMBS beeped, shook the wheel, and displayed BRAKE. Straight road, no traffic either direction. LKAS and Road Departure were both off. CMBS would have been off too if I had remembered to turn it off when I started for home.

    Clarity Warn.jpg
  14. petteyg359

    petteyg359 Active Member Subscriber

    FCW (forward collision warning) can be triggered by a falling leaf and a shadow in the wrong place for a moment, and I've had it happen a couple times, but those both of those obviously move relative to the car as it moves, so they don't actually trigger the car to brake on its own.
  15. Danks

    Danks Member

    Interesting about the leaf and shadow. As I remember it - could be misremembering - mine did drop out of cruise control when it triggered.
  16. AveSharia

    AveSharia New Member

    I have had mine trigger when approaching a bridge with no traffic in front of me. I don't think it was the bridge itself, but the shadow under the bridge. The screenshot you posted could have had the same effect if it was a similar time of day to the street view camera: there is a shadow from the trees on the right that appears abruptly.

    I assume these things are trained to not mistake those for cars, but like all "smart" systems... they ain't perfect. :/
  17. KentuckyKen

    KentuckyKen Well-Known Member

    At least this and other reports are about false positives or unintended activations instead of false negatives where the system fails to activate.
    Since no system can be perfect (as @AveSharia points out) and all these systems are in their early stages, I can accept the few erroneous activations since a false positive is mostly annoying while a false negative or lack of activation is potentially catastrophic.

    I appreciate how the engineers set up all this Honda Sense because I am safer with it and only get a very, very few false activations. These are usually on an inside curve with oncoming traffic or on an outside curve with a car pulling into the right lane to exit.

    So I don’t expect the Honda Sense to be perfect and I sure wouldn’t bet my life on it like a few idiots do by sleeping while their Tesla drives itself, but I do use it keeping in mind its limitations (and mine!).
    MPower likes this.
  18. AveSharia

    AveSharia New Member

    Yeah you wouldn't make it 30 seconds. :) The lane-keeping on the Clarity will turn itself off if you don't move the wheel for 15-20 seconds, and while the 'lane departure' shake will stay active, it won't keep you in-lane or re-activate the lane-keeping.

    Not that I've tried. Swear.
  19. Groves Cooke

    Groves Cooke Active Member

    The Clarity is my second car with LKAS. My 2015 Acura TLX also had it. Worked about the same. Will ask you to steer after 15 to 20 seconds. The Acura also had the forwIt too had false positive, usually when driving from bright sunshine into a shadow. I am used to it and don't mind the false positives at all.
  20. 2002

    2002 Well-Known Member

    Even if they removed the automatic shut-off you couldn't autodrive for very far because even when active it doesn't always detect the lanes and it starts to go out of the lane. Depends very much on the roadway markings and other conditions including lighting, and shadows as mentioned. And it can only handle shallow curves, anything more sharp and the system doesn't have enough torque to turn the wheel, or at least there is a limit to the torque the system will use. So you really have to be watching at all times just the same as if you weren't using the system, with your hands either on or very close to the steering wheel and ready to grab the wheel at a moments notice when the car begins drifting.

    I actually think if the system worked better it would ironically be more dangerous because a lot of people would then get complacent after experiencing that it rarely fails to detect the road, and then do dumb things like try and dig out some missing item from their glove compartment or tend to a child in the back seat. Or of course send a long text message. I like the idea that openpilot has where the system has a camera watching your face and making sure that you have your eyes on the road, if you look away for more than I think ten seconds it beeps and eventually shuts off. Actually I think it should be much tighter like two seconds because that is the recommended maximum time you should ever take your eyes off the road.

    Tesla autopilot is much more sophisticated than Clarity including using downloaded information on road conditions. But it still is not 100% reliable, although I don't know if Tesla has a camera watching the driver or if drivers are just on the honor system when using it and are expected to heed all of the warnings that Tesla gives about not relying on it.

Share This Page