Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

IIHS implementing new safeguard rating for advanced driver assist systems

This site may earn commission on affiliate links.

diplomat33

Average guy who loves autonomous vehicles
Aug 3, 2017
12,558
18,389
USA
Saw this news that IIHS is implementing a new safeguard rating for advanced driver assist systems like Autopilot, SuperCruise etc... Here is the key part:

The safeguards will be rated good, acceptable, marginal or poor. To earn a good rating, systems will need to ensure that the driver’s eyes are directed at the road and their hands are either on the wheel or ready to grab it at all times. Escalating alerts and appropriate emergency procedures when the driver does not meet those conditions will also be required.

IIHS expects to issue the first set of ratings in 2022. The precise timing is uncertain because ongoing supply chain woes in the auto industry have made it more difficult to obtain vehicles for testing.

 
Looks like they are following (maybe?) in Consumer Reports footsteps for something similar.

Yeah. Driving monitoring is seen as a key safety feature for advanced driver assist systems. So I think there is a push for advanced driver assist systems to have a really solid driver monitoring system that includes a driver facing camera and escalating alerts when the driver is not paying attention.
 
  • Like
Reactions: K5TRX
So the rating is only about the effectiveness of "nannying", not controlling the vehicle?

Apparently yes. It is not a rating of the driver assist itself, like how well it does lane keeping. It is a rating of the safeguard mechanism, ie how good is the driver assist is at keeping the driver engaged so that they can take over if something goes wrong.
 
  • Like
Reactions: K5TRX
Yeah and driver engagement will be a huge part of ADAS for a long time yet, since we're nowhere near the point of vehicles that can drive themselves in most situations with an inattentive driver

Beyond that, I think we'll start seeing more vehicles with driver monitoring built in regardless of ADAS. Vehicles will roll off the line with systems that are designed to detect distraction, intoxication, fatigue, etc when driven manually and throw out warnings or otherwise take action in response. When lane keeping is turned on in my 2021 F-150 rental, it'll detect your ability to stay inside lane lines and warn you to pull over when it thinks you're slipping up.
 
  • Like
Reactions: K5TRX
IIHS has put up a video that seems to show some "good" and "bad" examples?

There was an example of "good" Autopilot slowing down when unbuckling as well as "bad" Autosteer deactivating when steering. Although for the latter, "Lane centering does not discourage steering by driver" would seem to have IIHS prefer drivers needing to manually cancel Autopilot with the stalk before making steering correction instead of just directly taking over?

The video also shows Autopilot triggering an audible double beep before resuming cruise control while displaying a "small red hands" message to pay attention to the road:
resume acc.jpg


I'm guessing IIHS would take away points for that. I wonder if Tesla could still get the highest "safeguard" rating if there were options to control these behaviors where the standard could be the ones IIHS wants, e.g., default is to require hands on wheel or looking up to resume.
 
The eye tracking from supercruise was pretty good. I remember it had IR transmitters so the camera could capture the pupils even behind sunglasses and at night. If IIHS considers that "good", then I predict all automakers, including Tesla, will just implement that, in addition to whatever exists today (seat sensor, steering wheel torque, seat belt, etc). Then this issue of driver monitoring becomes moot, and we can all stop discussing it, the same way we never discuss seat belt safety concerns because everyone has adopted seat belts.
 
The IIHS says that no vehicle on the market currently meets all of the criteria proposed under this new series of tests. The institute plans to issue the first set of ratings this year. The requirements for a “good” rating are listed below.

Don't quite understand that for a 100% rating you will need to have all of these... But I can understand that many will say that this skips the point of using ADAS in the first place. Understandable from an insurancce point of view however. IIHS is the Insurance Institute for Highway Safety.

2022-IIHS-Vehicle-Automation-Misuse-safeguards_2.jpg
 
  • Informative
  • Like
Reactions: nvx1977 and K5TRX
good point - WHY does there have to be a hard mode where its ON or OFF? it should just gravitate toward lane centering when it can, like a gravity 'snap'. and if the car is so smart like elon keeps bragging, it should not be hard to know if its supposed to be centering or if a driver is admin'ing it himself.

in fact, it does not sound all that hard, now that I think about it. and there would not be the hard 'clunk' to get out of the AP mode, which IS a huge problem (it causes some to overcomp and run off the road just trying to disengage AP via the wheel torque).

if I'm angling and I initiated it, that's an admin event, let me continue to have control. when I stop doing things, that's a sign that I like where it is and 'keep it going' where it is. does not sound like ... rocket science.

I object to cameras in the cabin. very very low res sensors, maybe, but I can't trust vendors on what tech they use and how securely they keep that data stream (and how locally). vendors dont have a good track record of trust, tbh.

I'm pretty sure the wheel torque is still a valid way to show driver==alive. cameras are a surveillance state nightmare and should be avoided at all costs. they just are not necessary and are huge overkill.

besides, lets address an elephant in the room: who says that 'driver eyes pointing forward' is going to get you a safer driving experience. some people can pass that test but still be nearly brain-dead (not literally, of course). I dont think this is a metric that is really worth the privacy invasion it costs.

a final thought - the 'enforcement mechanism' has zero to do with self driving tech. its a policing measure and adds nothing to the algorithm or speed of decision making on the part of the car. its a purely admin function and not everyone agrees that it offers enough value to justify itself.

to be clear, I'm not one of those 'mah freedumbs!' kind of people. but I also dont think we should just blindly go down the surveillance path and install cabin cameras and make them mandatory.

there are some public safety things we opted, as a society, NOT to do. [cough cough]. for one reason or another, we decided not to do things that would ensure a higher level of societal safety. and yet, people are demanding that all cars have surveillance cams in them 'so that we'll all be safer'. the doublethink is strong, in this, dontchathink? again, I'm not one of the freedumbs gang, but I dont like cameras forced on me 'because of public safety'. I dont buy that and I dont like where it can lead.
 
there would not be the hard 'clunk' to get out
I believe this turning force threshold is software actively applying the steering wheel motor as physical feedback that the driver is indeed about to cancel Autosteer. Personally I think I would prefer Autosteer to remain active without the "clunk" as at least with FSD Beta, deactivating Autosteer while keeping TACC active can lead to somewhat unexpected behaviors, and that I also typically reactivate Autosteer anyway, so it would save on a lot of double-stalk-tap activations.

I would guess why Tesla doesn't do this is if one steers left slightly to change lanes, Autosteer's initial reaction would be to steer back right to stay in the current lane. Then it gets confusing about how much steering force one should one apply potentially overriding Autosteer's desire to stay in lane; or if Autosteer should have a "backoff" of some amount of time to allow the human to stop applying force to return steering to the car.

Maybe alternatively if visual and audible indicators are sufficient foregoing the physical feedback, Autopilot behavior could be to have Autosteer still deactivate on human steering but automatically reactivate when force is not detected on the wheel? Too confusing?

I suppose what IIHS might specifically be considering "discourage steering" is indeed the "clunk," so potentially at minimum removing that additional force even without keeping Autosteer active might be "sufficient" for a better rating.
 
I would guess why Tesla doesn't do this is if one steers left slightly to change lanes, Autosteer's initial reaction would be to steer back right to stay in the current lane. Then it gets confusing about how much steering force one should one apply potentially overriding Autosteer's desire to stay in lane; or if Autosteer should have a "backoff" of some amount of time to allow the human to stop applying force to return steering to the car.

Maybe alternatively if visual and audible indicators are sufficient foregoing the physical feedback, Autopilot behavior could be to have Autosteer still deactivate on human steering but automatically reactivate when force is not detected on the wheel? Too confusing?
Well this is how every other manufacturer handles Lane Keep Assist. Put turn signal on, turn wheel with a little force to disengage lane keep(autosteer), once lane change is complete, lane keep turns back on once it detects lines on both sides. You don't have to re-engage autosteer every time you change lanes.