The mocap rigid body is a different point/frame from the ENU PX4-IMU truth — needs alignment.
mocap_tag_board is a static board (pos σ < 0.6 mm) tracked the entire run, and its pose is identical across transition_1/2/3 to 0.6 mm / 0.04° — i.e. the mocap-world frame and the fiducial field are the same physical rig on all three transitions.
This is the keystone for reconstructing the transition handoff: a tag map / anchor built on one transition transfers to the others.
All published mocap pose streams are strictly time-ordered. Across every mocap_vehicle_data.csv and mocap_tag_board.csv, diff(t) > 0 for every row, so there are no duplicate timestamps and no backward steps; the smallest observed positive step is 7.510185241699219e-05 s on indoor_2/indoor_2_sensors/mocap_vehicle_data.csv.
True rate ~334 Hz median (≈3.0 ms spacing) on every run; the README per-run gross rates (297.88–358.80 Hz) reflect leading/trailing dead time, not different clocking.
Mocap is not gap-free: occlusion dropouts to 152 ms indoors, 864 ms (transition_1), 225 s (transition_3).
For the static mocap_tag_board lane specifically, transition_1 and transition_2 have no gaps larger than 50 ms, while transition_3 has seven such gaps totaling 2.3508646488189697 s.
transition_3 has one stray trailing sample 110.8 s past run end — clip it.
The per-run README.txtDistance value on mocap-backed runs is not the raw integrated 3-D path length of mocap_vehicle_data.csv. Exact mismatches are:
indoor_1 README 59.7771 m vs raw mocap path 128.58901563922984 m
indoor_225.2334 m vs 49.81123596722689 m
indoor_322.5216 m vs 46.214818576145845 m
transition_18.4006 m vs 18.9871245167508 m
and transition_29.6913 m vs 18.83194186346261 m
The README number is a derived summary with a different definition, not integrated mocap arc length.