Jump to content


Photo
- - - - -

Uber Tempe AAV crash ntsb report.


  • Please log in to reply
4 replies to this topic

#1 mariner

mariner
  • Member

  • 1,837 posts
  • Joined: January 07

Posted 11 November 2019 - 11:19

My apologies for starting a new AV thread but it seemed simpler.

 

The preliminary NTSB report into the crash in Tempe, Arizona where an Uber modified XC90 hit and killed a pedestrian pushing a bicycle has been published.

 

https://www.ntsb.gov...H010-prelim.pdf

 

The media reports suggested the Uber AV systems where not set up to recognise a pedestrian I don't think the report says that.

 

What it does say is that the AV system detected an object 6 seconds from impact but took until 1.3 seconds pre impact to determine emergency braking was needed.

 

What is rather weird is that  "Uber had disabled the AV emergency braking to "to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. The system is not designed to alert the operator"

 

For me two things stand out.

 

1) how long the system takes , albeit at night , to respond - 4.7 seconds - that's  about 120 metres at 60 mph.

 

2) They relied totally on the permanent alertness of a driver who is basically not driving the car to actually initiate the emergency braking as the system gives no warning to her.

 

Any thoughts?



Advertisement

#2 NotAPineapple

NotAPineapple
  • Member

  • 375 posts
  • Joined: July 13

Posted 11 November 2019 - 17:55

Sounds likely that they were having the usual problems that are faced by AV's which is the triggering of false positives on the AEB systems which presents a significant danger in itself. It's a fair reason to have the system turned off. 

 

The driver slacking off in a proto is inexcusable. I always drive proto's with gritted teeth, waiting for the next thing to go wrong. 

 

I've seen enough of silicon valleys automotive engineering to not trust it on it's own for more then a second or two. 

 

 



#3 Greg Locock

Greg Locock
  • Member

  • 5,698 posts
  • Joined: March 03

Posted 11 November 2019 - 18:29

According to someone who has (presumably) read the whole report, the tracking computer kept recognising her alternately as a bike and a pedestrian, and started a new track each time. According to somebody else who has presumably read the report, there was no attempt to save target history, so there was no tracking between updates.

 

One or other or both of these interpretation is missing something.

 

As to drivers slacking off when driving protos. Hmm, this is exactly what is wrong with Tesla's L2 masquerading as L5. People are not very good at just monitoring fairly reliable systems.



#4 Charlieman

Charlieman
  • Member

  • 1,804 posts
  • Joined: October 09

Posted 12 November 2019 - 16:11

1) how long the system takes , albeit at night , to respond - 4.7 seconds - that's  about 120 metres at 60 mph.

What is the expectation for human beings under the same circumstances?



#5 Greg Locock

Greg Locock
  • Member

  • 5,698 posts
  • Joined: March 03

Posted 13 November 2019 - 00:19

2.3 seconds is the average according to https://copradar.com...A2000_ABS51.pdf

 

That is rather long, but it is only half of Uber's fail.