Shape Completion with Prediction of Uncertain Regions

TL;DR

The paper tackles shape completion from partial views under irreducible viewpoint-induced uncertainty caused by partial symmetry. It extends occupancy-based implicit-function learning with two methods to predict uncertain regions and evaluates their impact on grasp planning using a ShapeNet mug dataset with ground-truth uncertainty. The approach outperforms two adapted probabilistic baselines in both occupied and uncertain-region predictions, and accounting for uncertainty improves grasp safety in synthetic and real data, though sim2real gaps remain. Overall, the work provides effective strategies for predicting and leveraging uncertain regions to enhance manipulation reliability in cluttered or occluded settings.

Abstract

Shape completion, i.e., predicting the complete geometry of an object from a partial observation, is highly relevant for several downstream tasks, most notably robotic manipulation. When basing planning or prediction of real grasps on object shape reconstruction, an indication of severe geometric uncertainty is indispensable. In particular, there can be an irreducible uncertainty in extended regions about the presence of entire object parts when given ambiguous object views. To treat this important case, we propose two novel methods for predicting such uncertain regions as straightforward extensions of any method for predicting local spatial occupancy, one through postprocessing occupancy scores, the other through direct prediction of an uncertainty indicator. We compare these methods together with two known approaches to probabilistic shape completion. Moreover, we generate a dataset, derived from ShapeNet, of realistically rendered depth images of object views with ground-truth annotations for the uncertain regions. We train on this dataset and test each method in shape completion and prediction of uncertain regions for known and novel object instances and on synthetic and real data. While direct uncertainty prediction is by far the most accurate in the segmentation of uncertain regions, both novel methods outperform the two baselines in shape completion and uncertain region prediction, and avoiding the predicted uncertain regions increases the quality of grasps for all tested methods.

Figures (6)

• Figure 1: Shape completion of a mug. Left: If the handle is occluded from the camera view, we predict the uncertain region (red) that contains the handle, resulting from pose ambiguity. The mug is reconstructed in the region not affected by pose ambiguity (gray). Right: If any part of the handle is visible, also the handle is reconstructed, and no uncertain region is predicted.
• Figure 2: Slice of a side view from the predicted occupancy probability grid (a). Using a small lower threshold, a region possibly containing a handle appears behind the mug (light red), but the mug itself is also contained (dark red). Using the gradient of the predicted occupancy probability (b) with its average as upper threshold, this unwanted region can be discarded by only considering the intersection of regions shown in red as uncertain.
• Figure 3: Effect of varying the threshold parameter $\tau$ on IoU for occupied and uncertain region predictions.
• Figure 4: Grasping the mug with an occluded handle without filtering using the predicted uncertain region leads to collision with the handle (left). Discarding grasps that collide with the uncertain region avoids collision and thus improves the grasp quality (right).
• Figure 5: Qualitative sim2real results. From top to bottom: Input point cloud, predicted mesh, ground-truth mesh. From left to right: test datasets $\text{HB}_{\rm pri}$, $\text{HB}_{\rm kin}$, LM, TYOL, $\text{YCBV}_{48}$, $\text{YCBV}_{55}$.