Guoliang, thanks for your question. For what concerns odometry, you can evaluate noise by comparing the reconstructed pose of the robot (included in the datasets) with the pose from the ground truth stream. (You can find an in-depth analysis of the precision of the GT data in Deliverable AD2.3, which you can download from the "Documents" section of this website.) By the way, our datasets also include the raw data from the wheel encoders (tick count), so you could even perform your own calibration of odometry -if you so desire- instead of using ours. For other sensor streams, the answer to your question depends on what stream you are thinking of and on what meaning you give to the word "noise". We will be happy to answer to more specific questions about this :-)
Hi, Thanks for your answer. I am testing my slam algorithm which fusion bearing only camera with odometry. So far I am using entended kalman filter. The motion model in EKf need the noise parameters, such as noise covariance assuming the noise is Gaussian distribution. Unfortunately the ground truth data is not available when robot start to move. But your idea is great, maybe I can calibrate it using ground truth data. Thanks.