Please excuse me. Some questions about algorithmic equation deducing and a certain experiment operation details. about r-vio

[muchen12]

Is there anyone that can help me please? I have pushed this issue here for almost a whole week. Could you please help me?

[huaizheng]

1. In our robocentric formulation, the frame of reference needs to be changed from one IMU frame to the next at every image time which is achieved through the composition step. In addition to the change of reference for the state (40), the covariance matrix should also be expressed in the new frame of reference. This is done by the so-called covariance propagation (41) with a corresponding Jacobian matrix (43). To derive it, you should start from the original composition equation whose form is the same as (40), however, now without "^hat" for each state. And for the relevant procedures you can refer to our companion tech report (please check the reference) where we have provided a lot of similar derivations for the other equations shown in this paper to get their Jacobians, which can help you to derive (43) yourself.
2. As R-VIO dose not need "SLAM wiggle" for initialization, you can skip all those stages in the dataset and then start it. This also works if you don't want to wait for a long stationary period. For example, the mav takes off after around 11 seconds in V1_02, so you can first play rosbag for 8 seconds and stop (with --pause option), then roslaunch R-VIO, and finally resume playing rosbag to let the estimation begin. I just tried this myself before writing this comment and it worked, and the result was just the same as what I showed you before. Actually, when you enable --pause option the rosbag will be preloaded so that you don't need to worry about the timestamp issue, because it will just replay all the messages included. So this is why I have no clue for the issue you met. If such thing occurs all the time, I may suggest you to try it on another computer to see what you can get.