3 ms·
Is it really so dismal? Let's run some numbers: The accident rate is around 74 per 100 million miles (and fatalities is 1.13). It's unclear exactly how to tur
by ars 11y ago
Is it really so dismal? Let's run some numbers:
The accident rate is around 74 per 100 million miles (and fatalities is 1.13).
It's unclear exactly how to turn that into a percentage, but no matter how you do it it's quite high.
Say an accident takes 5 minutes, and people drive 30 miles/hour. Then that works out to 99.999% for humans. If you use the numbers for fatalities then it's 99.99999%.
So the human track record for fatalities is a failure rate of 0.00001% - driverless cars are going to have a VERY hard time beating that.
Would you like to calculate what that works out to for the reliability of each individual computer, or order to have such good results in aggregate?
- yourapostasy 11y agoThis is an interesting analysis, but I'm not sure we have enough data on autonomous cars to know one way or the other. Perhaps failsafe protocols can help mitigate catastrophic effects from computer hardware and software failures, and that edges out humans who override equivalent "failures" like impaired driving capacity. The data on US traffic fatalities [1] is interesting, because you point out it will be hard for the autonomous vehicles to beat the average, but in some sparsely-populated states the rate is dramatically lower, which could potentially challenge autonomous vehicles' safety records even more in those states. If a lot of their safety is developed from training, and they don't train much in the sparsely-populated states because they're so low population, then maintaining the safety rate will depend a lot upon whether the training in more populous states adequately covers use cases encountered in the field in the sparsely-populated states. I suspect that at the end of the day, unless autonomous vehicles are much worse on the fatality rate, they only have to roughly match the human driver-based fatality rate, and commercial interests will push extremely hard for autonomous vehicles. The number of miles driven in the US is tracked by the Federal Reserve among others [2]. Current 12 month moving total is 3,147,579 millions of miles (3.14*10^12). Latest figure for just crashes (not only those with fatalities) is a 2011 article [3] that asserted 185 per 100 million Vehicle Miles Traveled (VMT). The Insurance Information Institute has a report [4] that cites a $277 billion in 2010 cost of crashes, not including quality of life valuations. Autonomous cars don't have to move the needle that much over the current regime before insurance will put in very strong incentives to use them; there is a ton of room to improve, and the insurance companies stand to gain a huge windfall by not lowering rates commensurately with actual cost savings. The trucking companies must be going bonkers over the possibilities to run diesel trucks 24x7. [1] http://www.iihs.org/iihs/topics/t/general-statistics/fatalityfacts/state-by-state-overview http://www.iihs.org/iihs/topics/t/general-statistics/fatalit... [2] https://research.stlouisfed.org/fred2/series/M12MTVUSM227NFWA https://research.stlouisfed.org/fred2/series/M12MTVUSM227NFW... [3] http://www.caranddriver.com/features/safety-in-numbers-charting-traffic-safety-and-fatality-data http://www.caranddriver.com/features/safety-in-numbers-chart... [4] http://www.iii.org/issue-update/auto-crashes http://www.iii.org/issue-update/auto-crashes