Timeline for Estimating velocity from known position and acceleration
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 6, 2012 at 12:30 | comment | added | c0dehunter | Thank you both, I hope this thread will help future visitors too. | |
| Sep 6, 2012 at 12:11 | comment | added | Jason R | If you say that the variance of a measurement is $0.00001$, then in the Kalman filter framework, you're saying that it is Gaussian distributed with a standard deviation of $\sqrt{0.00001}$. So, ~68% of the time, the measurement error magnitude will be less than $\sqrt{0.00001}$, ~95% of the time, it will be less than $2\sqrt{0.00001}$, and so on. | |
| Sep 6, 2012 at 12:08 | comment | added | c0dehunter | But what does this mean in practice? If I say my position sensor has 0.00001 variance (first diag. element in Q matrix) does that mean it's mistake is +- 0.00001 meters for example (I guess not)? I guess you choose smaller variance for a sensor that is more precise (position) and a bigger variance for a sensor with less precision (acceleration). | |
| Sep 6, 2012 at 12:01 | history | bounty awarded | c0dehunter | ||
| Sep 6, 2012 at 11:51 | comment | added | Peter K.♦ | Thanks! The reason for choosing 0.00001 over 2.84: Well, 2.84 is VERY large for a variance. I assumed that you had a valid reasoning behind the jerk number (0.001), and that the other numbers should be significantly smaller than this. If there is no real reason for choosing the jerk number, try the same value in all diagonal elements. | |
| Sep 6, 2012 at 6:09 | comment | added | c0dehunter | Nice explaination - but why choose 0.001, why not 2.84 for example? | |
| Sep 5, 2012 at 23:46 | history | edited | Peter K.♦ | CC BY-SA 3.0 | added 198 characters in body |
| Sep 5, 2012 at 21:17 | history | answered | Peter K.♦ | CC BY-SA 3.0 |