Re:SR-71's “R2-D2” Could Be the Key to Winning Fights in GPS...
by planteen on Saturday March 30, 2019 @09:25AM
(#19529281)
Standard error decreases by the square root of the number of samples. So an order of magnitude improvement would need 100 sensors!
[ Parent
| on HN ]
Re:SR-71's “R2-D2” Could Be the Key to Winning Fights in GPS...
by justinjlynn on Saturday March 30, 2019 @11:20AM
(#19529838)
Of course, when you decrease the size of the sensor by half in each dimension while maintaining its error rate you've quadrupled the number of sensors. If you can increase the computing power similarly, or decrease the computing requirements for combining those sensor measurements - it quickly becomes feasible to use massive numbers of sensors to improve the standard error significantly - but it is a lot of ifs.
Practical example, you'd only have to shrink the sensor by half 6 or 7 times to fit 100 in the same area as 1.
Re:SR-71's “R2-D2” Could Be the Key to Winning Fights in GPS...
by mattkrause on Saturday March 30, 2019 @09:48AM
(#19529370)
That’s true when averaging together iid samples, but you could probably do something smarter if you know about the likely errors and the sensors’ covariance. For example, if three of the four sensors agree, you should probably completely disregard the fourth, rather than averaging it in.
Re:SR-71's “R2-D2” Could Be the Key to Winning Fights in GPS...
by robotresearcher on Saturday March 30, 2019 @11:52AM
(#19530006)
> you could probably do something smarter
Assuming measurements are independent and errors are Gaussian, the Kalman filter is this smarter thing. Variants exist for nonlinear systems.