3 ms·
Standard error decreases by the square root of the number of samples. So an order of magnitude improvement would need 100 sensors!
by planteen 8y ago
Standard error decreases by the square root of the number of samples. So an order of magnitude improvement would need 100 sensors!
- mattkrause 8y agoThat’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.
- robotresearcher 8y ago> 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. https://en.wikipedia.org/wiki/Kalman_filter https://en.wikipedia.org/wiki/Kalman_filter
- justinjlynn 8y agoOf 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.