long trajectory with big z drift dataset, converge not good in z direction

[DaojunZhu]

hi, thanks for your great work. we adapt hba in our own dataset, when the data has multilayer, the converge reuslt is good in xy direction, but is not good in z direction, even has bad effects. do you have any suggestion about this problem ?

we also confused about the hba code in PGO, which the first layer constraint build from initial pose not the optimized pose as measurement.

expect your reply, thanks!!