Uncertainty-Aware Stereo Grasp Point Selection for Deformable Linear Objects
Cristina Saccani, Alessio Caporali, Gianluca Palli
AI summary
Problem
Grasp point selection on thin, deformable linear objects like cables is highly sensitive to stereo depth estimation errors near occlusions and discontinuities, leading to unreliable grasps.
Approach
A five-stage stereo network jointly predicts disparity, semantic labels, and per-pixel uncertainties, which are then used to filter and rank candidate grasp points before final selection.
Key results
- Reduced mean grasp-point depth error from 4.19 mm to 1.55 mm
- Increased grasp success rate within 3 mm tolerance from 74.2% to 88.6%
- Lowered 90th percentile failure exceedance above 3 mm from 29.47 mm to 6.77 mm
- Achieved real-time inference at 32.4 FPS on a standard GPU
Why it matters
Provides a practical, real-time method for robots to safely manipulate cables and ropes by explicitly accounting for perception confidence during grasp planning.
Abstract
Reliable grasp point selection on deformable lin- ear objects, such as cables, requires not only accurate depth estimation but also awareness of prediction reliability. We present a five-stage stereo network for joint disparity, semantic, and uncertainty estimation, and use the predicted uncertainty to filter grasp candidates before geometric ranking. Disparity uncertainty is modeled via a Laplace negative log-likelihood, se- mantic uncertainty via the entropy of semantic predictions, with an alignment term enforcing consistency between them. Experi- ments on a synthetic stereo dataset show that uncertainty-aware selection reduces the mean grasp-point depth error from 4.19 mm to 1.55 mm, increases the success rate within a 3 mm tolerance from 74.2% to 88.6%, and lowers the 90th percentile of the failure exceedance above 3 mm from 29.47 mm to 6.77 mm. These results show that uncertainty is an effective cue for safer grasp selection on deformable linear objects.