Task-Relevant Depth Quality Metrics for Suction Grasping

Shivansh Inamdar

Keywords: depth evaluation, suction grasping, task-relevant metrics, contact mechanics, depth estimation, perception

TL;DR: We propose physics-grounded depth quality metrics for suction grasping and show that methods with worse RMSE can produce better geometry for grasp success.

Abstract: Standard depth evaluation metrics (RMSE, MAE) measure global accuracy but fail to capture the local geometric properties that determine suction grasp success. These properties include surface planarity within the contact patch, surface normal accuracy at grasp points, and contact patch completeness near object boundaries. We propose four task-relevant depth quality metrics grounded in suction contact mechanics and evaluate three depth estimation methods on 1,200 images from the GraspNet-1Billion dataset. Our results reveal a consistent rank reversal: the raw depth sensor achieves two to three times better RMSE than learned methods, yet scores worse than at least one learned method on every task-relevant metric. Learned models produce geometrically coherent surfaces (smooth, complete, with consistent normals) despite worse metric accuracy, and suction grasping rewards coherence over accuracy. This suggests that standard metrics can mislead practitioners selecting depth methods for manipulation, and that hybrid pipelines using sensor depth for positioning and model depth for grasp evaluation and final approach may be beneficial.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 35