Hacker News new | past | comments | ask | show | jobs | submit login
Automatic Emergency Braking: A Case Study (medium.com)
18 points by cma 7 days ago | hide | past | web | favorite | 7 comments





>Car manufacturers also lack the ability to issue Over-The-Air (OTA) software updates, which is another strong limitation in developing and deploying new features.

I am actually very happy about this. I want definite times (ie after servicing) that my car is going to change algorithms, not whenever the car company decides.

In addition, I think this is a very large potential security hole.


I’ve seen it pointed out in the case of Tesla’s OTAs, for instance, drivers may get used to the behavior of Autopilot for their car in a certain segment of road and then have an OTA change that behavior without much warning.

(Of course, there are trade-offs here in that OTAs could also bring important safety functionality, such as Tesla’s fix for braking distance on the Model 3.)


That's what emergency recalls are for. It has the benefit of making it very clear to the owner that something in the vehicle is changing because you have to physically bring it to the shop. Not that it can't be solved by software, but it's not solved well today.

I disagree that the "turning vehicle" scenarios represent false positives. The driver can't possibly know that there isn't e.g. a scooter in front of the turning vehicle. It isn't safe to drive full speed ahead with the assumption that once the turning vehicle is gone, there isn't another obstacle in front of it, and the EAB is correct to brake in these instances.

Not sure in those cases, but we make assumptions all the time based on not having seen a scooter previously, and assuming a scooter couldn’t have gotten into a car-occluded area without moving faster than X mph. Most AEB maybe doesn’t use long term temporal data to make decisions or trust its own non-observations as being definitive (it may assume there was a strong chance of a false negative in detecting a scooter in a previous frame too easily and not allow that non observation to rule it out in the current frame; it may also have enough of a false negative detection rate that what it is doing is rational).

> This is about one AEB event every 3,300 driven miles:

What the hell are people doing when they're driving? I don't mean the ones distracted by cell phones or whatever. I mean the ones who appear to be looking out their windshield, but simply don't react when something unexpected starts to happen in front of them.

I routinely find myself braking before the driver ahead of me. Passengers have even commented on it because -- I guess -- it's... unusual? So I can't fathom what. are. people. looking. at?


Do comma.ai make this data available for researchers, or is this something that they are hoping to monetise?



Applications are open for YC Winter 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: