bias stability or random walk parameter
rising-turtle opened this issue · 1 comments
Hi All:
By comparing the content in README with the IMU noise model described in https://github.com/ethz-asl/kalibr/wiki/IMU-Noise-Model , it seems that the parameters Gyroscope/Accelerometer "bias Instability" (in the README) are the same with the Gyroscope/Accelerometer "random walk" (in the link). Since the unit 1/(s^2*sqrt(HZ)) = 1*sqrt(HZ)/(s). Correct me if I make any mistake.
If these two parameters are the same, according to the README, the Gyroscope/Accelerometer "bias Instability" are estimated by the minimum value of the Allan variance.
On the contrary, the link identifies the Gyroscope/Accelerometer "random walk" as the value of the (fitted) "random walk" diagonal (with slope +1/2) at an integration time of t=3 second. Which is the same approach used in the paper "The TUM VI Benchmark for Evaluating Visual-Inertial Odometry" (displayed by Fig. 5).
So, which is right? Any suggestion or correction is greatly appreciated, thanks in advance.
The Kalibr wiki page is right, and the code here and the documentation is wrong. Check out my fork of this repo where I fixed this and many other problems: