← Back to project

X2 data pipeline: x2_native_v3 at official-grade smoothness

X2 SONIC · GMR retarget + wrist fix + 6 Hz smoothing · 2026-07-04
dataretargetingjerk

1 · Result

Canonical pipeline = GMR retarget + wrist-order fix + 6 Hz smoothing → x2_native_v3: 2,497 clips, jerk 7.0 ≈ 6.9 (NVIDIA official G1 data). All X2 training data now flows through this pipeline.
Data-quality comparison: official data vs x2_native_v3 (after smoothing).

2 · What the pipeline fixes

defect in raw GMR outputfix
wrist 6-joint permutation mismatchexplicit joint-order remap
high-frequency jitter (jerk far above official)6 Hz low-pass smoothing
foot skate / contact driftcontact-fix foot-lock (v3), root-only floor correction

Passed visual review before adoption. Note: even the official dataset carries ~5 cm/s residual foot skate — that is the reference bar, not zero.

3 · Sample gallery — G1 official ↔ X2 canonical

Left = NVIDIA official G1 data, right = the same motion in the X2 dataset. Note the embodiment differences playing out: shorter legs (smaller stride), narrower shoulder range.

turning walk · v3.1 canonical
dynamic upper-body sequence · v3.1 canonical
jump · high-dynamic sample (07-03 build, pre-final smoothing)

4 · Sample gallery — retarget stage (human source → X2)

The GMR stage of the pipeline: human mocap skeleton (left) driving the X2 body (right). These renders are from the retarget stage itself, before the final smoothing pass.

forward jog
walk loop 360°
sideways walk 45°
backward jog
SBS clips: left = reference motion data, right = policy rollout (Isaac Lab, full physics) · yellow ball = reference anchor · all runs single RTX 3090