Robocars - Autonomous Drive, Self-driving, Driver-less cars

Didn't it happen at night? And involve a pedestrian? And she came from the side?

Yes, yes, and - given the relative speeds involved - no.

A 3m blindspot around a vehicle is irrelevant when it's travelling at nearly 40mph. By that point any collision is completely unavoidable. Even at 20mph 3m is only ~1/3 of a second. If a someone or something hasn't been detected before then, it's still probably too late to do anything about it.

It's not clear where the 3m blind spot is. I take it to mean it is a ring in cantact with, and completely surrounding, the car, with an outer diameter or radius of 3m - caused by the shadow of the body of the car, blocking the roof mounted LIDAR array. If so, then it would seem to be irrelevant to this incident.

Makes sense. It still seems very bad that they'd allow a blind spot that big though, no? In terms of lower speed incidents? Then admit that.

If she was killed by a cyclist, the rider would already have been lynched..

3m before impact they were 3m from the car. Yet no braking occured.

Seems to me the front sensor was fucking useless and the lack of side sensors also equates to being fucking useless. Conclusion: Uber are dodgy, cheapskate pricks and their bullshit vehicle killed someone.

unavoidable

Yet, early statements said there was no braking. They probably mean, given lack of skid marks or whatever, but the data logger should show if there was ANY braking, even if it was just prior to impact. No braking attempt = sensors not doing their job.

IN my head I'm comparing this to airbags - how fast do airbags detect impact and deploy?

There was no braking but was there any trigger to brake sent by their sensors? If not, then it seems whatever detection system they are using is not fit for purpose, if there was then why was there a delay? What limitation caused the lack of recognition? Lack of sensors? Poor sensors? Poor code? etc?

Makes sense. It still seems very bad that they'd allow a blind spot that big though, no? In terms of lower speed incidents? Then admit that.

Cars don't need LIDAR. One can do SLAM without it-- see. for example, https://vision.in.tum.de/research/vslam/lsdslam.
You don't need LIDAR to "see". LIDAR is not magic. 850-950 nm is not really ideal-- except that one can get emitters out of silicon. Radar, ultrasonic, optical cameras and other sensors should-- and I think we'll increasingly see this-- be sufficient. Radar works better than 950nm LIDAR in snow, rain and fog. Optical cameras can "see" color and constrast. Ultrasonic has very short distances but works pretty well in all weather conditions.
Including a LIDAR on the roof is, I think, a good solution this week. Even if they get down to under $250 (and the Velodyne units start at around $4k and some are as much as $80k) and reduced in size there is still no need to have them by the 6 pack.. Don't underestimate the power and potential of optical image capture.
There are also a number of interesting optical camera capture techniques on the horizon. See: http://web.media.mit.edu/~guysatat/fog/

https://www.theguardian.com/world/2018/apr/28/autopilot-driver-who-sat-in-passenger-seat-is-banned-for-18-months

https://gizmodo.com/report-ubers-self-driving-car-sensors-ignored-cyclist-1825832504

Slipped out in the news yesterday - unsurprisingly little coverage. Uber really are schmucks.

"Uber's Self-Driving Car Sensors Ignored Cyclist In Fatal Accident"

So, yeah, they made it too human-like, too realistic.
Maybe dial it back to talky toaster.

https://www.bbc.co.uk/news/technology-44574290

"
The "safety operator" of a self-driving Uber car was watching TV just before the vehicle was involved in a fatal collision, a police report reveals.
"

"

The "safety operator" of a self-driving Uber car was watching TV just before the vehicle was involved in a fatal collision, a police report reveals.
"

This is part and parcel of what I seee as the biggest technical hurdle to self driving cars: getting systems good enough for Level 4 without crashing and burning. As ADAS improves the cognitive demands for the control driver increases as the task become more and more boring-- perhaps something for a specific profile within the autistic spectrum.

Report in the Guardian about auto cars and Holland uses the phrase "cyclists loose adherence to types of the road".

Daniel Boffey is evidently a twat.

This one:

https://www.theguardian.com/world/2019/feb/13/bikes-put-spanner-in-works-of-dutch-driverless-car-schemes

In Holland we'd say they're making an elephant out of a mosquitto. From what I've read city centres are the last place autonomous vehicles are expected to be implemented anyway. Oddly, despite our high number of cyclists we're still ranked highest on preparedness, making the entire article look a bit silly. Which leads me to believe the writer of the article is an anti-cycling lobbying Tory (is he? I've always wanted to use that word) cunt.

I've come down on the side of those who think that automated cars will be a reality in city centres, and not in the too distant future, too. I think all those who claim that 'it's too difficult' will simply realise that if it's too difficult it'll be made 'easier'. That means regulation of walking and cycling.

Not only has @Uber's liability been eliminated in the fatal crash w/ a
pedestrian, but the "safety driver" who was watching The Voice when
the crash happened might get off too. So it's a free for all...

https://twitter.com/designjunkie8/status/1103346232674668544

It's hard to overstate how important this moment is. We're already
losing.

I posted this in the Road Danger Reduction thread, but I might as well repost it here:

https://www.theguardian.com/technology/2019/oct/03/collision-course-pedestrian-deaths-rising-driverless-cars

,

Well, what's your comma?

https://www.theguardian.com/technology/2021/jan/03/peak-hype-driverless-car-revolution-uber-robotaxis-autonomous-vehicle

https://www.theguardian.com/technology/2021/apr/19/tesla-crash-texas-federal-safety-regulators-investigation