Views
No views yet
GNSDynamicsModel — Graph Network Simulator for variable-vertex cloth meshes.
Predicts normalised per-node acceleration from a 5-frame velocity history
and node types (free / grasped), integrated as pos += v * dt with dt = 0.1.dexgarmentlab_dual_gps2048_clean.h5, Cloth-splatters/dexgarmentlab-dual-meshes-clean)GNSRegressionLoss)node_input_size 17)velocity_noise_std settings; Val/accel_mse in physical units was 0.163)demos2h_dyn_gns_k5_vn0p03_2026-08-21_17-27-35_776088 (full config in config.yml)K=5 / vn=0.03 gave the best
rollout error (4.40 cm at frame 6) against 4.74 cm for K=5 / vn=0.1 — which
is what the older dexgarmentlab-folding-lifting-dynamics-gns checkpoint uses
— and 5.31 cm for the best K=1 cell. The noise optimum is interior: 0.1
over-smooths, 0.01 is too weak. Noise is applied once per step with the target
built from the noised velocity, so it trains error correction rather than
acting as input jitter.input_sequence_length: 1 /
node_input_size: 5.max_grippers: 2. The action is one token stream per hand, with a learned
gripper-ID embedding and per-gripper grasped-flag node channels. A trajectory
that moves only one hand parks the other and flags it in the recording's
hand_active attribute; the loader masks that slot out of both gripper_mask
and the translation-centering centroid, so it behaves exactly like a padded
slot.gripper_pos / actuated_vertices fields, but on a two-handed
trajectory those are a lossy back-compat projection — the legacy point is a
virtual gripper between the hands, ~14 cm from the vertices it nominally
drives (the real per-hand positions sit 3-4 cm away), and the legacy mask
covers only ~85 of the ~153 vertices the two hands actually hold. Use
gripper_pos_2h / actuated_vertices_2h, as ClothDynamicsVariableDataset
and ClothDynamicsGraphDataset now do.scripts/benchmark_dynamics_accuracy.py, validation split (24 held-out
garments — the split is by garment, baked into the dataset file, so no
trajectory of a validation cloth is ever seen in training), 50 denoising
steps, one trajectory per cloth for the rollout:| model | one-shot (5 frames) | rollout mean | rollout end |
|---|---|---|---|
-dynamics-gps (DDPM) | 1.25 cm | 1.86 cm | 3.00 cm |
-dynamics-gps-flow | 1.24 cm | 1.84 cm | 3.10 cm |
-dynamics-gns | 0.52 cm | 3.75 cm | 5.72 cm |