Clicky

Jump to content

Archived

This topic is now archived and is closed to further replies.

Ruslan

Self-driving cars

Recommended Posts

Some suspect reliability stats:

  1. Uber claims to have completed 2,000,000 miles in autonomous mode. (as of 22 December 2017)
    1. They have run over and killed one person.
  2. Google:
    1. Google claims that they have done over 300,000 miles accident free (August 2012)
    2. Google claims they have driven their fleet 1,500,000 miles in autonomous mode (March 2016).
    3. During that time there were 14 collisions, 13 the fault of the other driver.
      1. This means on accident every 107,142 miles.
      2. Accident avoidance is a driver skill
      3. If all the cars on the road were automated, would this mean only one accident in 1,500,000 miles?
  3. Tesla claims 130 million miles driven by its customers with Autopilot engaged.
    1. Tesla claims only one person killed while driving in autopilot mode (in Florida).
    2. Other people claim the January 2016 fatality in Hubei China was the result of autopilot system error.
  4. In the U.S., there is a fatality every 94 million miles among all types of vehicles.
    1. This is according to Tesla. I have not independently verified it.
    2. This calculations I gather includes motorcycle drivers and pedestrians.
    3. Therefore, if Tesla only has one death, they are doing better than humans, if they had two, they are doing worse.
  5. So to summarize
    1. U.S. standard is one fatality every 94 million miles.
    2. Google has done 1.5 million miles without a fatality, but has had one accident every 107,142 miles.
    3. Tesla has done either 130 million miles per fatality or 65 million miles per fatality.
    4. Uber has 2 million miles per fatality.

So....close....but not quite there yet.

 

 

Share this post


Link to post
Share on other sites

Great research.

It is impressive that they can make robots so relatively safe. It can be difficult for me sometimes to see exactly where the road is in bad weather at night. To make a robot do that without fault must be a mighty task.

Share this post


Link to post
Share on other sites
12 hours ago, Robert Rick said:

Great research.

It is impressive that they can make robots so relatively safe. It can be difficult for me sometimes to see exactly where the road is in bad weather at night. To make a robot do that without fault must be a mighty task.

Hardly great research....I mostly just clipped it from Wikipedia.

It is the visibility that is an issue. That crash in Florida occurred because with the sun was backlighting a tractor trailer, and the Tesla did not see that it was there (could not distinguish it)...and drove right into it at speed. When humans get into a bad visibility situation, they tend to slow down (and least the ones who live a long life). The computer may not realize that it is in a bad visibility situation. As one unrelated blog post points out:  "It is how badly it gets it wrong when it makes a mistake that has me worried." Look at the picture of how a computer identified the difference between a Husky and a Wolf: http://www.dupuyinstitute.org/blog/2018/03/21/artificial-intelligence-ai-and-warfare/

I have no doubt that switching to self-driving cars today would greatly reduce the number of accidents. In the case of the Google car, they has 14 accidents, of which 13 were the fault of the other driver. The problem is that the system sometimes goes horribly wrong (not that humans don't).

But, the thing that gets my attention is what in 2014, the United States had 32,744 people killed in car accidents. This is 1.08 fatalities per 100 million miles. This is the lowest rate ever (it was 24.09 in 1921).From 2009 to 2014, we had between 32,479 to 33,883 people killed per year in a car. These are the lowest number of people killed in a year by cars since 1951 (the peak year was 1972 with 54,589 killed). But.....what gets my attention is that in 2016 it was 37,461 people killed by cars. This is 4,717 more Americans killed a year. This is more than we lost in the Iraq War. This is more than we lost on 9/11. It seems like we are becoming worse drivers. This is borne out by statistics and my observations while driving (and I am teaching my son to drive right now...just to drive this point home).  

See: https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year

 

Share this post


Link to post
Share on other sites
5 hours ago, Ruslan said:

 It seems like we are becoming worse drivers

 

Maybe we just drive more. Maybe some new fancy vehicle is dangerous. Maybe it's just within normal statistical fluctuations.

Regarding the wolf and the husky that is a classical problem with ai, that you never know what the computer actually thinks - more pictures of wolfs would have solved that problem. And the same goes with self-driving cars - the more experience these vehicles get collectively the safer they will be.

I live in Copenhagen. Here we have robotic trains in the metro. I feel perfectly safe taking them.

Share this post


Link to post
Share on other sites
35 minutes ago, Robert Rick said:

Maybe we just drive more. 

In the Wikipedia article, look at the third column, which is "Fatalities per 100 million VMT." It starts at 24.09 in 1921 and rather steadily declines from there. As it is tied to miles driven....then the accident rate per mile should not change much based upon usage. Normal statistical fluctuation seems very limited. It is higher than the previous year in only 19 cases (1930, 1933, 1934, 1940, 1941, 1943, 1950, 1955, 1962, 1963, 1964, 1966, 1977, 1979, 1980, 1986, 2005, 2015 and 2016), and often by only very small increments. It looks very steady from 2009 to 2014 (ranging from 1.15 to 1.08), then starts climbing. It could be statistical variation, but when one considers that there are over 30,000 fatalities and over 3 trillion miles traveled, I don't think there is a lot of statistical variation there. On the other hand, I see lots of people on their phone while I am driving. Keep in mind that this is an uptick of 0.1 from 2014 to 2016. The last time there was such a large uptick was from 1977 to 1980. While 0.1 may not sound significant, it is almost 5,000 additional people dead. That is a lot.

I gather it is lower in 2017, but not as low as 2014. In the end, we are seeing stories about teenagers driving and tweeting their mom just before a fatal accident. I kind of think now that the sooner we go to self-driving cars, the better.

Share this post


Link to post
Share on other sites

I've been wondering, won't it be easy to hack such a car? Or something like antivirus software will be used? I've just finished reading avast premium review and it seems like modern antiviruses are way better than they were back then.

Share this post


Link to post
Share on other sites

Re:  ... it seems like modern antiviruses are way better than they were back then
>That IS because, modern viruses are way better than they were back then

Ever heard the saying ‘Change is the only constant’?

 

Share this post


Link to post
Share on other sites
On 25/04/2018 at 9:39 AM, Robert Rick said:

Great research.

It can be difficult for me sometimes to see exactly where the road is in bad weather at night. To make a robot do that without fault must be a mighty task.

NOT really, the road has NOT moved.  The road IS a constant, the ONLY variable is, the conditions.
All that IS required is simply to preprocess for the different conditions.

Share this post


Link to post
Share on other sites

In my opinion, this is too unreliable. You know, I am a very careful person. Before ordering an essay I usually read academized reviews to make sure that I will get a good product. So, I will never go in this car till it is improved. And reading reviews actually always is a great thing to do before you pay for something.

Share this post


Link to post
Share on other sites
On 5/7/2019 at 5:21 PM, Albertino said:

In my opinion, this is too unreliable. You know, I am a very careful person. Before ordering an essay I usually read... to make sure that I will get a good product. So, I will never go in this car till it is improved. And reading reviews actually always is a great thing to do before you pay for something.

But I don’t read anything before ordering an essay, I just order it in the Wr1ter service and that’s it. I get the finished written work on time and it suits me. I always order my essays in the service of writers.

Share this post


Link to post
Share on other sites

Let's don't be too clever around here with double talk. How is promotion (sale) of literary services related to self-driving cars? 

Share this post


Link to post
Share on other sites
On 5/7/2019 at 4:21 PM, Albertino said:

In my opinion, this is too unreliable. You know, I am a very careful person. Before ordering an essay I usually read academized reviews to make sure that I will get a good product. So, I will never go in this car till it is improved. And reading reviews actually always is a great thing to do before you pay for something.

Re: Before ordering an essay I usually read academized reviews
> It IS great 2 C that UR readin skills HAVE progressed from Bill & Ben 

 
 

BillBenJpg.jpg

Share this post


Link to post
Share on other sites

×
×
  • Create New...