umich-curly / cheetah_inekf_realtime Goto Github PK

Hi, can I get your (synced_proprioceptive_lcmt).data package mentioned in https://github.com/UMich-CURLY/cheetah_inekf_realtime/blob/main/src/communication/lcm_handler.cpp， I checked your data folder LCM file, which should be the packet of the previous version,

I have trouble when apply this method to estimating the 6-dim pose of quadruped A1. Specifically, the estimation of position along z direction ( gravity direction) is not accurate in my experiments, which decays as quadruped keep troting and the acceleration baise at z direction reaches to about -0.17 . I noticed that there are no such functions as below figure in RobotState.cpp . But I have set the value of covariances as the same with ones of this repository via inekf::NoiseParams.

It is known that inaccurate estimation of z position may happen when only IMU data is available. However, with additional information, such as kinematics data and contact information, is it possible to estimate the 6-dim pose enoughly ? Or, do I make mistakes somewhere ? Please help .