Cosmos CV4x8x8 rollouts — 4-way ablation on mv1_v2_stride2 (with perceptual metrics)
Diffusion-forcing world model rollouts. Each sample is ~2.7 s (16 raw frames @ 6 fps).
Side-by-side videos show ground truth on the left , model prediction on the right .
First half of each clip = history (GT), second half = autoregressively sampled future.
Click a sample header to expand its videos (default-collapsed to keep the page light).
4-way Cosmos comparison with LPIPS/SSIM/PSNR . p01 (fixed ref), p01rand (buggy threshold=0.01), vision_only (num_tactiles=0), tactile_only (num_views=0). Same architecture; only tokenizer + modality mix differ.
Other Sections
Setup comparison Click a run name to jump to its rollout videos. ✓ = feature on, ✗ = off. Lower val_loss is better.
Perceptual metrics (decoded videos, 10 val samples) LPIPS lower=better; SSIM higher=better. Best-of-class in green . Future-only frames (history skipped). Run name = same anchor as in setup table above.
Rollout-latent metrics (10 val samples) Latent-space MSE of predicted future vs ground truth. Contact/no-contact split uses tactile latent-to-reference energy (threshold 0.05).
run view MSE TL MSE TR MSE tactile contact MSE tactile no-contact MSE p01_cosmos 0.071850 0.038378 0.032779 0.042333 0.006308 p01rand_cosmos 0.062772 0.039231 0.032639 0.042614 0.006994 vision_only_cosmos 0.161045 0.000000 0.000000 n/a 0.000000 tactile_only_cosmos n/a 0.036167 0.034861 0.041334 0.010296
p01_cosmos
sample_000 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_001 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_002 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_003 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_004 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_005 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_006 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_007 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
p01rand_cosmos
sample_000 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_001 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_002 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_003 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_004 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_005 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_006 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_007 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
vision_only_cosmos
sample_000 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_001 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_002 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_003 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_004 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_005 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_006 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_007 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
tactile_only_cosmos
sample_000 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_001 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_002 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_003 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_004 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_005 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_006 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
sample_007 (click to expand videos) view — GT | rollout tactile_left — GT | rollout tactile_right — GT | rollout
