Wrist-stream ablation

Checkpoint mot_df_vt_a9dfilm_dforce_isolate_adaln_wrist_mixed (mixed wrist-availability training: 32 no-wrist + 21 wrist episodes, per-clip wrist dropout p=0.2), evaluated with and without the wrist-camera stream on three held-out splits. "Without" forces the model into its 3-stream (view + tactile L/R) mode at inference (--force_no_wrist); it never removes tactile or view. All pixel metrics use the pad-latent-restored Wan2.2 decode (fixed 2026-09-15 -- see the pad_latent note below); frames 0-7 are clean GT context, 8-15 are predicted.

Every episode below is from the test split (never used in training, never used to pick the checkpoint). Full manifest (mixed_mb_wrist_holdout.json, 49 train / 4 val / 7 test episodes, 162912/7726/19707 windows):

train 49 episodes, 162912 windows -- gradient updates only val motherboard_0510_episode_005, motherboard_0510_episode_006, motherboard_0911_episode_002_seg00, motherboard_0912_episode_003 -- checkpoint selection only (never shown here) test motherboard_0511_episode_009, motherboard_0519_episode_002, motherboard_0909_episode_000, motherboard_0909_episode_001, motherboard_0909_episode_002, motherboard_0911_episode_008_seg00, motherboard_0912_episode_002 -- everything shown below

no wrist

No-wrist episodes (0511_ep009, 0519_ep002)

motherboard_0511_episode_009 test motherboard_0519_episode_002 test

32 held-out windows, predicted frames only (context frames excluded)
streamwith wrist
camera view24.76 dB
SSIM 0.83 · LPIPS 0.09
tactile left41.66 dB
SSIM 0.98 · LPIPS 0.06
tactile right38.62 dB
SSIM 0.97 · LPIPS 0.09

Sample rollouts

motherboard_0511_episode_009   t_start=1577   test

camera view
with wrist · PSNR 27.7 dB
tactile left
with wrist · PSNR 37.1 dB
tactile right
with wrist · PSNR 29.9 dB

motherboard_0511_episode_009   t_start=1658   test

camera view
with wrist · PSNR 24.7 dB
tactile left
with wrist · PSNR 49.9 dB
tactile right
with wrist · PSNR 41.5 dB

motherboard_0511_episode_009   t_start=2067   test

camera view
with wrist · PSNR 26.0 dB
tactile left
with wrist · PSNR 51.0 dB
tactile right
with wrist · PSNR 32.7 dB

motherboard_0519_episode_002   t_start=13   test

camera view
with wrist · PSNR 23.3 dB
tactile left
with wrist · PSNR 37.1 dB
tactile right
with wrist · PSNR 31.9 dB

motherboard_0519_episode_002   t_start=821   test

camera view
with wrist · PSNR 21.6 dB
tactile left
with wrist · PSNR 31.6 dB
tactile right
with wrist · PSNR 37.8 dB

motherboard_0511_episode_009   t_start=412   test

camera view
with wrist · PSNR 26.3 dB
tactile left
with wrist · PSNR 44.1 dB
tactile right
with wrist · PSNR 44.6 dB

wrist indist

Wrist episodes, in-distribution (0911_ep008_seg00, 0912_ep002)

motherboard_0911_episode_008_seg00 test motherboard_0912_episode_002 test

32 held-out windows, predicted frames only (context frames excluded)
streamwith wristwith wrist
camera view24.64 dB
SSIM 0.81 · LPIPS 0.09		24.59 dB
SSIM 0.81 · LPIPS 0.09
wrist left15.04 dB
SSIM 0.42 · LPIPS 0.35		n/a
(no wrist cameras)
wrist right16.79 dB
SSIM 0.52 · LPIPS 0.30		n/a
(no wrist cameras)
tactile left43.42 dB
SSIM 0.98 · LPIPS 0.04		43.63 dB
SSIM 0.98 · LPIPS 0.04
tactile right46.79 dB
SSIM 0.99 · LPIPS 0.03		46.80 dB
SSIM 0.99 · LPIPS 0.03

Sample rollouts

motherboard_0912_episode_002   t_start=2105   test

camera view
with wrist · PSNR 24.7 dB
with wrist · PSNR 24.8 dB
wrist left
with wrist · PSNR 15.1 dB
no wrist cameras
wrist right
with wrist · PSNR 15.6 dB
no wrist cameras
tactile left
with wrist · PSNR 47.6 dB
with wrist · PSNR 47.8 dB
tactile right
with wrist · PSNR 43.5 dB
with wrist · PSNR 42.0 dB

motherboard_0912_episode_002   t_start=2930   test

camera view
with wrist · PSNR 25.4 dB
with wrist · PSNR 25.5 dB
wrist left
with wrist · PSNR 17.0 dB
no wrist cameras
wrist right
with wrist · PSNR 14.7 dB
no wrist cameras
tactile left
with wrist · PSNR 45.9 dB
with wrist · PSNR 44.2 dB
tactile right
with wrist · PSNR 45.5 dB
with wrist · PSNR 45.4 dB

motherboard_0912_episode_002   t_start=1883   test

camera view
with wrist · PSNR 24.5 dB
with wrist · PSNR 25.7 dB
wrist left
with wrist · PSNR 16.9 dB
no wrist cameras
wrist right
with wrist · PSNR 15.0 dB
no wrist cameras
tactile left
with wrist · PSNR 42.7 dB
with wrist · PSNR 43.5 dB
tactile right
with wrist · PSNR 51.3 dB
with wrist · PSNR 51.5 dB

motherboard_0912_episode_002   t_start=5141   test

camera view
with wrist · PSNR 25.0 dB
with wrist · PSNR 24.3 dB
wrist left
with wrist · PSNR 14.7 dB
no wrist cameras
wrist right
with wrist · PSNR 17.0 dB
no wrist cameras
tactile left
with wrist · PSNR 41.4 dB
with wrist · PSNR 41.3 dB
tactile right
with wrist · PSNR 46.3 dB
with wrist · PSNR 47.0 dB

motherboard_0912_episode_002   t_start=4726   test

camera view
with wrist · PSNR 23.7 dB
with wrist · PSNR 22.9 dB
wrist left
with wrist · PSNR 13.2 dB
no wrist cameras
wrist right
with wrist · PSNR 18.0 dB
no wrist cameras
tactile left
with wrist · PSNR 50.3 dB
with wrist · PSNR 50.1 dB
tactile right
with wrist · PSNR 32.8 dB
with wrist · PSNR 33.1 dB

motherboard_0912_episode_002   t_start=4928   test

camera view
with wrist · PSNR 22.6 dB
with wrist · PSNR 23.1 dB
wrist left
with wrist · PSNR 15.6 dB
no wrist cameras
wrist right
with wrist · PSNR 15.9 dB
no wrist cameras
tactile left
with wrist · PSNR 37.9 dB
with wrist · PSNR 39.0 dB
tactile right
with wrist · PSNR 48.7 dB
with wrist · PSNR 49.8 dB

wrist ood

Wrist episodes, val0909 session -- OOD (hardware/calibration session held out of training)

motherboard_0909_episode_000 test motherboard_0909_episode_001 test motherboard_0909_episode_002 test

32 held-out windows, predicted frames only (context frames excluded)
streamwith wristwith wrist
camera view20.86 dB
SSIM 0.78 · LPIPS 0.15		21.49 dB
SSIM 0.79 · LPIPS 0.14
wrist left13.51 dB
SSIM 0.42 · LPIPS 0.48		n/a
(no wrist cameras)
wrist right12.91 dB
SSIM 0.38 · LPIPS 0.50		n/a
(no wrist cameras)
tactile left31.95 dB
SSIM 0.97 · LPIPS 0.15		32.08 dB
SSIM 0.97 · LPIPS 0.15
tactile right28.59 dB
SSIM 0.95 · LPIPS 0.17		33.08 dB
SSIM 0.96 · LPIPS 0.16

Sample rollouts

motherboard_0909_episode_001   t_start=2995   test

camera view
with wrist · PSNR 20.5 dB
with wrist · PSNR 21.8 dB
wrist left
with wrist · PSNR 11.6 dB
no wrist cameras
wrist right
with wrist · PSNR 11.7 dB
no wrist cameras
tactile left
with wrist · PSNR 29.5 dB
with wrist · PSNR 29.2 dB
tactile right
with wrist · PSNR 30.6 dB
with wrist · PSNR 38.1 dB

motherboard_0909_episode_001   t_start=3318   test

camera view
with wrist · PSNR 20.8 dB
with wrist · PSNR 21.2 dB
wrist left
with wrist · PSNR 15.3 dB
no wrist cameras
wrist right
with wrist · PSNR 12.0 dB
no wrist cameras
tactile left
with wrist · PSNR 32.4 dB
with wrist · PSNR 32.2 dB
tactile right
with wrist · PSNR 27.9 dB
with wrist · PSNR 30.4 dB

motherboard_0909_episode_000   t_start=2281   test

camera view
with wrist · PSNR 20.2 dB
with wrist · PSNR 21.4 dB
wrist left
with wrist · PSNR 13.0 dB
no wrist cameras
wrist right
with wrist · PSNR 13.9 dB
no wrist cameras
tactile left
with wrist · PSNR 31.6 dB
with wrist · PSNR 32.2 dB
tactile right
with wrist · PSNR 27.8 dB
with wrist · PSNR 32.2 dB

motherboard_0909_episode_001   t_start=1765   test

camera view
with wrist · PSNR 21.0 dB
with wrist · PSNR 21.6 dB
wrist left
with wrist · PSNR 13.8 dB
no wrist cameras
wrist right
with wrist · PSNR 13.9 dB
no wrist cameras
tactile left
with wrist · PSNR 32.5 dB
with wrist · PSNR 32.7 dB
tactile right
with wrist · PSNR 25.9 dB
with wrist · PSNR 27.7 dB

motherboard_0909_episode_000   t_start=1649   test

camera view
with wrist · PSNR 21.2 dB
with wrist · PSNR 21.6 dB
wrist left
with wrist · PSNR 13.3 dB
no wrist cameras
wrist right
with wrist · PSNR 11.4 dB
no wrist cameras
tactile left
with wrist · PSNR 30.4 dB
with wrist · PSNR 30.4 dB
tactile right
with wrist · PSNR 29.5 dB
with wrist · PSNR 35.4 dB

motherboard_0909_episode_002   t_start=1136   test

camera view
with wrist · PSNR 20.4 dB
with wrist · PSNR 21.6 dB
wrist left
with wrist · PSNR 14.6 dB
no wrist cameras
wrist right
with wrist · PSNR 11.7 dB
no wrist cameras
tactile left
with wrist · PSNR 32.4 dB
with wrist · PSNR 32.7 dB
tactile right
with wrist · PSNR 25.8 dB
with wrist · PSNR 29.3 dB

Contact scenarios

The random N=32 windows above mix in a lot of free-motion / no-contact time. These are the top-16 (per group) windows by fraction of the 8 predicted fps15 frames with max(force_left, force_right) > 0.5N (contact_frac 0.88-1.00), spread across episodes -- see select_contact_test_windows.py. Same with/without-wrist toggle, same pad-latent decode.

no wrist

No-wrist episodes

motherboard_0511_episode_009 test motherboard_0519_episode_002 test

16 held-out windows, predicted frames only (context frames excluded)
streamwith wrist
camera view23.36 dB
SSIM 0.80 · LPIPS 0.12
tactile left38.84 dB
SSIM 0.97 · LPIPS 0.07
tactile right36.23 dB
SSIM 0.96 · LPIPS 0.11

Sample rollouts (all genuine contact)

motherboard_0511_episode_009   t_start=21   test

camera view
with wrist · PSNR 23.7 dB
tactile left
with wrist · PSNR 50.7 dB
tactile right
with wrist · PSNR 32.2 dB

motherboard_0511_episode_009   t_start=383   test

camera view
with wrist · PSNR 25.8 dB
tactile left
with wrist · PSNR 47.1 dB
tactile right
with wrist · PSNR 34.0 dB

motherboard_0511_episode_009   t_start=1225   test

camera view
with wrist · PSNR 24.6 dB
tactile left
with wrist · PSNR 41.5 dB
tactile right
with wrist · PSNR 50.4 dB

motherboard_0511_episode_009   t_start=1891   test

camera view
with wrist · PSNR 25.4 dB
tactile left
with wrist · PSNR 49.1 dB
tactile right
with wrist · PSNR 35.0 dB

motherboard_0519_episode_002   t_start=112   test

camera view
with wrist · PSNR 21.0 dB
tactile left
with wrist · PSNR 34.6 dB
tactile right
with wrist · PSNR 34.3 dB

motherboard_0519_episode_002   t_start=366   test

camera view
with wrist · PSNR 22.9 dB
tactile left
with wrist · PSNR 28.8 dB
tactile right
with wrist · PSNR 34.5 dB

motherboard_0519_episode_002   t_start=630   test

camera view
with wrist · PSNR 22.6 dB
tactile left
with wrist · PSNR 31.6 dB
tactile right
with wrist · PSNR 34.2 dB

motherboard_0519_episode_002   t_start=782   test

camera view
with wrist · PSNR 20.4 dB
tactile left
with wrist · PSNR 32.7 dB
tactile right
with wrist · PSNR 38.4 dB

wrist indist

Wrist episodes, in-distribution

motherboard_0911_episode_008_seg00 test motherboard_0912_episode_002 test

16 held-out windows, predicted frames only (context frames excluded)
streamwith wristwith wrist
camera view25.46 dB
SSIM 0.84 · LPIPS 0.08		25.30 dB
SSIM 0.83 · LPIPS 0.08
wrist left15.80 dB
SSIM 0.41 · LPIPS 0.32		n/a
(no wrist cameras)
wrist right18.23 dB
SSIM 0.60 · LPIPS 0.21		n/a
(no wrist cameras)
tactile left35.84 dB
SSIM 0.96 · LPIPS 0.09		35.53 dB
SSIM 0.96 · LPIPS 0.09
tactile right47.29 dB
SSIM 0.99 · LPIPS 0.03		46.81 dB
SSIM 0.99 · LPIPS 0.03

Sample rollouts (all genuine contact)

motherboard_0911_episode_008_seg00   t_start=11   test

camera view
with wrist · PSNR 24.7 dB
with wrist · PSNR 24.2 dB
wrist left
with wrist · PSNR 13.6 dB
no wrist cameras
wrist right
with wrist · PSNR 17.7 dB
no wrist cameras
tactile left
with wrist · PSNR 32.3 dB
with wrist · PSNR 32.0 dB
tactile right
with wrist · PSNR 51.7 dB
with wrist · PSNR 51.6 dB

motherboard_0911_episode_008_seg00   t_start=254   test

camera view
with wrist · PSNR 26.2 dB
with wrist · PSNR 26.2 dB
wrist left
with wrist · PSNR 15.7 dB
no wrist cameras
wrist right
with wrist · PSNR 19.2 dB
no wrist cameras
tactile left
with wrist · PSNR 30.4 dB
with wrist · PSNR 30.1 dB
tactile right
with wrist · PSNR 51.9 dB
with wrist · PSNR 51.8 dB

motherboard_0911_episode_008_seg00   t_start=442   test

camera view
with wrist · PSNR 23.9 dB
with wrist · PSNR 24.7 dB
wrist left
with wrist · PSNR 16.2 dB
no wrist cameras
wrist right
with wrist · PSNR 16.1 dB
no wrist cameras
tactile left
with wrist · PSNR 33.8 dB
with wrist · PSNR 30.8 dB
tactile right
with wrist · PSNR 45.3 dB
with wrist · PSNR 45.8 dB

motherboard_0911_episode_008_seg00   t_start=818   test

camera view
with wrist · PSNR 25.0 dB
with wrist · PSNR 24.8 dB
wrist left
with wrist · PSNR 14.6 dB
no wrist cameras
wrist right
with wrist · PSNR 19.9 dB
no wrist cameras
tactile left
with wrist · PSNR 34.9 dB
with wrist · PSNR 35.6 dB
tactile right
with wrist · PSNR 51.0 dB
with wrist · PSNR 51.3 dB

motherboard_0912_episode_002   t_start=371   test

camera view
with wrist · PSNR 24.9 dB
with wrist · PSNR 25.5 dB
wrist left
with wrist · PSNR 17.3 dB
no wrist cameras
wrist right
with wrist · PSNR 15.4 dB
no wrist cameras
tactile left
with wrist · PSNR 36.0 dB
with wrist · PSNR 37.6 dB
tactile right
with wrist · PSNR 47.5 dB
with wrist · PSNR 46.8 dB

motherboard_0912_episode_002   t_start=2086   test

camera view
with wrist · PSNR 23.7 dB
with wrist · PSNR 23.4 dB
wrist left
with wrist · PSNR 16.2 dB
no wrist cameras
wrist right
with wrist · PSNR 17.7 dB
no wrist cameras
tactile left
with wrist · PSNR 38.1 dB
with wrist · PSNR 38.9 dB
tactile right
with wrist · PSNR 48.4 dB
with wrist · PSNR 46.6 dB

motherboard_0912_episode_002   t_start=4145   test

camera view
with wrist · PSNR 25.9 dB
with wrist · PSNR 25.8 dB
wrist left
with wrist · PSNR 16.3 dB
no wrist cameras
wrist right
with wrist · PSNR 17.9 dB
no wrist cameras
tactile left
with wrist · PSNR 30.0 dB
with wrist · PSNR 30.1 dB
tactile right
with wrist · PSNR 45.3 dB
with wrist · PSNR 45.3 dB

motherboard_0912_episode_002   t_start=6094   test

camera view
with wrist · PSNR 24.3 dB
with wrist · PSNR 27.3 dB
wrist left
with wrist · PSNR 14.0 dB
no wrist cameras
wrist right
with wrist · PSNR 15.5 dB
no wrist cameras
tactile left
with wrist · PSNR 39.5 dB
with wrist · PSNR 40.6 dB
tactile right
with wrist · PSNR 43.3 dB
with wrist · PSNR 42.1 dB

wrist ood

Wrist episodes, val0909 session -- OOD

motherboard_0909_episode_000 test motherboard_0909_episode_001 test motherboard_0909_episode_002 test

16 held-out windows, predicted frames only (context frames excluded)
streamwith wristwith wrist
camera view20.81 dB
SSIM 0.77 · LPIPS 0.15		21.48 dB
SSIM 0.79 · LPIPS 0.14
wrist left12.96 dB
SSIM 0.38 · LPIPS 0.48		n/a
(no wrist cameras)
wrist right13.32 dB
SSIM 0.38 · LPIPS 0.48		n/a
(no wrist cameras)
tactile left31.37 dB
SSIM 0.96 · LPIPS 0.15		31.56 dB
SSIM 0.96 · LPIPS 0.15
tactile right27.90 dB
SSIM 0.94 · LPIPS 0.19		31.82 dB
SSIM 0.95 · LPIPS 0.18

Sample rollouts (all genuine contact)

motherboard_0909_episode_000   t_start=113   test

camera view
with wrist · PSNR 21.0 dB
with wrist · PSNR 21.4 dB
wrist left
with wrist · PSNR 14.8 dB
no wrist cameras
wrist right
with wrist · PSNR 14.2 dB
no wrist cameras
tactile left
with wrist · PSNR 31.4 dB
with wrist · PSNR 32.1 dB
tactile right
with wrist · PSNR 26.8 dB
with wrist · PSNR 29.4 dB

motherboard_0909_episode_000   t_start=1562   test

camera view
with wrist · PSNR 19.8 dB
with wrist · PSNR 20.6 dB
wrist left
with wrist · PSNR 14.1 dB
no wrist cameras
wrist right
with wrist · PSNR 16.0 dB
no wrist cameras
tactile left
with wrist · PSNR 31.3 dB
with wrist · PSNR 31.5 dB
tactile right
with wrist · PSNR 27.5 dB
with wrist · PSNR 30.5 dB

motherboard_0909_episode_000   t_start=2890   test

camera view
with wrist · PSNR 21.0 dB
with wrist · PSNR 21.7 dB
wrist left
with wrist · PSNR 11.7 dB
no wrist cameras
wrist right
with wrist · PSNR 12.2 dB
no wrist cameras
tactile left
with wrist · PSNR 31.1 dB
with wrist · PSNR 30.8 dB
tactile right
with wrist · PSNR 29.0 dB
with wrist · PSNR 34.3 dB

motherboard_0909_episode_001   t_start=159   test

camera view
with wrist · PSNR 20.8 dB
with wrist · PSNR 21.6 dB
wrist left
with wrist · PSNR 11.6 dB
no wrist cameras
wrist right
with wrist · PSNR 12.8 dB
no wrist cameras
tactile left
with wrist · PSNR 32.2 dB
with wrist · PSNR 32.1 dB
tactile right
with wrist · PSNR 25.3 dB
with wrist · PSNR 27.8 dB

motherboard_0909_episode_001   t_start=2048   test

camera view
with wrist · PSNR 20.9 dB
with wrist · PSNR 21.3 dB
wrist left
with wrist · PSNR 14.1 dB
no wrist cameras
wrist right
with wrist · PSNR 11.9 dB
no wrist cameras
tactile left
with wrist · PSNR 32.5 dB
with wrist · PSNR 32.8 dB
tactile right
with wrist · PSNR 27.7 dB
with wrist · PSNR 30.9 dB

motherboard_0909_episode_002   t_start=121   test

camera view
with wrist · PSNR 21.7 dB
with wrist · PSNR 22.1 dB
wrist left
with wrist · PSNR 15.0 dB
no wrist cameras
wrist right
with wrist · PSNR 12.8 dB
no wrist cameras
tactile left
with wrist · PSNR 32.3 dB
with wrist · PSNR 32.7 dB
tactile right
with wrist · PSNR 27.0 dB
with wrist · PSNR 27.8 dB

motherboard_0909_episode_002   t_start=1540   test

camera view
with wrist · PSNR 21.4 dB
with wrist · PSNR 22.4 dB
wrist left
with wrist · PSNR 14.6 dB
no wrist cameras
wrist right
with wrist · PSNR 14.8 dB
no wrist cameras
tactile left
with wrist · PSNR 32.5 dB
with wrist · PSNR 33.0 dB
tactile right
with wrist · PSNR 27.9 dB
with wrist · PSNR 31.2 dB

motherboard_0909_episode_002   t_start=2468   test

camera view
with wrist · PSNR 21.3 dB
with wrist · PSNR 21.8 dB
wrist left
with wrist · PSNR 12.2 dB
no wrist cameras
wrist right
with wrist · PSNR 13.5 dB
no wrist cameras
tactile left
with wrist · PSNR 28.4 dB
with wrist · PSNR 28.7 dB
tactile right
with wrist · PSNR 27.9 dB
with wrist · PSNR 32.4 dB

8s long-horizon rollouts

Keep-1 autoregressive sliding-window rollout driven by the episode's recorded actions, one real long-horizon episode per split. Decoded with the same pad-latent-restored convention: decoded frame i is exactly raw frame i from the seed window's start (no more 3-frame offset / blurred first frame).

long  /  motherboard_0511_episode_009   test   (8s keep-1 rollout)

with wrist

long  /  motherboard_0519_episode_002   test   (8s keep-1 rollout)

with wrist

long  /  motherboard_0909_episode_000   test   (8s keep-1 rollout)

with wrist
with wrist

long  /  motherboard_0909_episode_001   test   (8s keep-1 rollout)

with wrist
with wrist

long  /  motherboard_0909_episode_002   test   (8s keep-1 rollout)

with wrist
with wrist

long  /  motherboard_0911_episode_008_seg00   test   (8s keep-1 rollout)

with wrist
with wrist

long  /  motherboard_0912_episode_002   test   (8s keep-1 rollout)

with wrist
with wrist