b-spline
Datasets
All datasets matching “b-spline”forceflow-bspline-knot-analysis
ForceFlow Adaptive B-spline Knot Analysis
Training-free adaptive cubic B-spline fitting results on episode 0 of
JokerESC/ForceFlow.
No policy, SmolVLA model, or GPU was used.
Question
When action changes little but force/torque changes substantially, does fitting
one vector-valued B-spline to [action, force, torque] allocate knots near the
force event that are absent from an action-only fit?
Both fits use one common knot vector per multichannel trajectory. For the… See the full description on the dataset page: https://huggingface.co/datasets/sungkyunner/forceflow-bspline-knot-analysis.univtac-bspline-budgeted-routing
UniVTAC shared-budget B-spline knot routing
This repository is a no-training diagnostic for multimodal B-spline representations on four UniVTAC episodes. It fits action and tactile marker-displacement splines separately while forcing them to share one temporal knot budget.
Experiment definition
Tasks: Lift Bottle, Lift Can, Insert HDMI, Pull Out Key; episode 0 from each task.
Inputs: 8-D joint action and 4,800-D marker displacement only. No tactile RGB is used.… See the full description on the dataset page: https://huggingface.co/datasets/sungkyunner/univtac-bspline-budgeted-routing.univtac-bspline-knot-analysis
UniVTAC × B-spline Policy fitting pilot
This repository fits cubic B-splines to an augmented target containing UniVTAC's next 8-D joint action and next 4,800-D GelSight Mini marker displacement. It deliberately performs fitting only, not policy training.
The completed pilot outputs are in results/. Each task/setting has:
a PNG of the complete episode with natural-language phase labels and action/tactile knot colors;
an NPZ containing the complete knot vector and every 4,808-D… See the full description on the dataset page: https://huggingface.co/datasets/sungkyunner/univtac-bspline-knot-analysis.bspline_fcmimic_cxr_bspline_518