From Prediction to Action: Learning Physical Context for Deformable Linear Object Manipulation
Organizations: The authors are with DEI - Department of Electrical, Electronic and Information Engineering, University of Bologna, 40136 Bologna, Italy.
Abstract
A representation that accurately predicts a deformable object's response is not necessarily the one most useful for choosing manipulation actions. We study this distinction between predictive utility and decision utility in planar single-grasp shape control of Deformable Linear Objects (DLOs). As a common decision backbone, we introduce Geometry-Pointer-Flow (GPF), a geometry-grounded policy that separates grasp-location selection from grasp-conditioned motion generation. Its geometry-only variant reduces one-step executed shape error by 32.8% relative to a DLO-specific diffusion policy, while using 18.7 times fewer parameters and achieving approximately 193 times faster inference. On this backbone, we compare physical context represented by ground-truth parameters, estimated parameters, and learned interaction-history representations, for both forward prediction and executed control. For prediction, ground-truth physical parameters reduce next-shape error by 15.8% over geometry alone, and estimated parameters are slightly more predictive than interaction history. This ordering does not carry over to control: estimated parameters provide no detectable one-step control gain, whereas history-conditioned GPF reduces executed error by 9.9% relative to geometry-only GPF across 470 paired simulation trials. In closed loop, interaction history improves strict-threshold attainment in simulation and yields the strongest zero-shot performance on hardware. All policies are trained exclusively in simulation and transferred to the physical robot without real-world training or fine-tuning. These results show that physical context should be evaluated by the decisions it supports rather than by predictive accuracy alone. Code will be released upon acceptance.
Figures & tables
| Method | Error | 95% CI | Params | Lat. | |
|---|---|---|---|---|---|
| (mm) | (mm) | (mm) | (M) | (ms) | |
| DLO-DP | 13.782 | 75.66 | 851.04 | ||
| MLP-BC | 14.317 | 3.88 | 0.32 | ||
| Transformer-BC | 13.957 | 3.76 | 1.37 | ||
| Global-Flow | 13.485 | 3.79 | 3.88 | ||
| GPF-Geom | 9.257 | – | – | 4.05 | 4.41 |
| Baseline | Matched | Same cond. | Other DLO | Other surface | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DLO | Surface | None | Est. | Hist. | Est. | Hist. | Est. | Hist. | Est. | Hist. |
| Soft | Smooth | 8.91 | 9.90 | 8.24 | 10.03 | 8.34 | 10.10 | 8.97 | 10.02 | 8.60 |
| Soft | Rough | 8.72 | 9.32 | 7.76 | 9.24 | 7.72 | 9.32 | 9.23 | 9.24 | 8.14 |
| Medium | Smooth | 10.29 | 11.77 | 8.62 | 11.80 | 9.14 | 11.83 | 9.70 | 10.19 | 8.73 |
| Medium | Rough | 10.84 | 13.91 | 9.44 | 13.88 | 9.66 | 13.95 | 10.33 | 13.51 | 9.66 |
| Stiff | Smooth | 11.53 | 11.50 | 9.96 | 11.52 | 10.04 | 11.47 | 10.11 | 11.61 | 9.90 |
| Controller | Error | Gain | Gain | 95% CI |
|---|---|---|---|---|
| (mm) | (mm) | (%) | (mm) | |
| GPF-Geom | 9.257 | – | – | – |
| Oracle-concat | 8.678 | 0.580 | 6.3 | |
| Estimated-concat | 8.821 | 0.436 | 4.7 | |
| Hist-concat | 8.494 | 0.763 | 8.2 | |
| Hist-FiLM | 8.336 | 0.921 | 9.9 |
| Best error (mm) | Reach 10 mm (%) | Passage (steps) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DLO | Surface | DLO- DP | GPF- Geom | Hist- FiLM | DLO- DP | GPF- Geom | Hist- FiLM | DLO- DP | GPF- Geom | Hist- FiLM |
| Soft | Smooth | 10.40 [9.82] | 7.97 [7.86] | 7.85 [7.28] | 50 | 90 | 90 | 9.2 | 6.7 | 6.3 |
| Soft | Rough | 10.32 [9.64] | 6.31 [6.62] | 5.73 [6.16] | 60 | 100 | 100 | 5.5 | 5.0 | 5.5 |
| Medium | Smooth | 8.78 [9.69] | 8.36 [8.01] | 6.73 [7.10] | 50 | 70 | 100 | 8.2 | 5.7 | 4.6 |
| Medium | Rough | 16.54 [19.37] | 10.48 [10.38] | 7.42 [7.07] | 20 | 50 | 90 | 6.0 | 8.0 | 7.2 |
| Stiff | Smooth | 9.31 [9.16] | 7.06 [7.45] | 6.48 [6.02] | 60 | 90 | 90 | 6.0 | 5.1 | 5.7 |
Deformable Object Manipulation
Deformable Object Manipulation studies how robots perceive, model, and control flexible objects such as cloth, ropes, cables, and soft tissue.
See all 112 papers