Datasets:

---

RoboDojo-Benchmark

/

RoboDojo

Like 8

Follow

RoboDojo-Benchmark 7

Replace data/RoboDojo_depth with the shift-fixed HDF5.

Color and approximate_depths drop frame 0 and shift forward; state, action,
and other time-aligned fields drop the last frame. Episode length becomes T-1.
Color streams are re-encoded as standard RGB JPEGs with the XPL-RGB1 marker.
Preview videos are regenerated from the shifted RGB frames.

Source: /personal/skl/data/hf/RoboDojo_depth_fixed
Script: fix_obs_shift_robodojo_depth.py