Cut a Wall, See a Wall: Identity Substitution in Occluded Point-Cloud Perception
Overview
This research addresses the challenge of point-cloud segmentation for construction robots in occluded environments. The study reveals that when entire regions are occluded during testing, segmentation networks mislabel these areas as a single class, leading to severe performance degradation. Introducing geometric point features (GPE) significantly improves segmentation accuracy, reducing whole-object losses and enhancing the reliability of perception in both synthetic and real-world scenarios.
From arXiv Robotics — research abstracts
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · test points: 4.5 other · Basis not reported
“column absorbs up to 70% of all test points; with randomized layouts the sink moves to the stair, predicted at 4.5-5.4x its true share across three training seeds. Baseline mIoU falls to 0.15-0.27 (fixed) and 0.36-0.72 (randomized), and up to 269 whole objects vanish in a single test set. Appending 14 hand-computed geometric features per point (GPE) restores mIoU to 0.54-0.92 and removes 64-100% of whole-object loss”
Exact source · revision 1Source owner not reported
Reported quantity · mIoU: 0.15 other · Basis not reported · Differing source assertions
“column absorbs up to 70% of all test points; with randomized layouts the sink moves to the stair, predicted at 4.5-5.4x its true share across three training seeds. Baseline mIoU falls to 0.15-0.27 (fixed) and 0.36-0.72 (randomized), and up to 269 whole objects vanish in a single test set. Appending 14 hand-computed geometric features per point (GPE) restores mIoU to 0.54-0.92 and removes 64-100% of whole-object loss”
Exact source · revision 1Source owner not reported
Reported quantity · mIoU: 0.36 other · Basis not reported · Differing source assertions
“column absorbs up to 70% of all test points; with randomized layouts the sink moves to the stair, predicted at 4.5-5.4x its true share across three training seeds. Baseline mIoU falls to 0.15-0.27 (fixed) and 0.36-0.72 (randomized), and up to 269 whole objects vanish in a single test set. Appending 14 hand-computed geometric features per point (GPE) restores mIoU to 0.54-0.92 and removes 64-100% of whole-object loss”
Exact source · revision 1Source owner not reported
Reported quantity · whole objects: 269 other · Basis not reported
“column absorbs up to 70% of all test points; with randomized layouts the sink moves to the stair, predicted at 4.5-5.4x its true share across three training seeds. Baseline mIoU falls to 0.15-0.27 (fixed) and 0.36-0.72 (randomized), and up to 269 whole objects vanish in a single test set. Appending 14 hand-computed geometric features per point (GPE) restores mIoU to 0.54-0.92 and removes 64-100% of whole-object loss”
Exact source · revision 1Source owner not reported
Evidence dependency changes (2)
- 2026-10-30T19:48:31.143Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T19:48:15.339Z · evidence updated · source revision 1. Evidence extraction was updated; current source attributions are shown above.
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsSignalCut a Wall, See a Wall: Identity Substitution in Occluded Point-Cloud Perception for Construction Robots
This research addresses the challenge of point-cloud segmentation for construction robots in occluded environments. The study reveals that when entire regions are occluded during testing, segmentation networks mislabel these areas as a single class, leading to severe performance degradation. Introducing geometric point features (GPE) significantly improves segmentation accuracy, reducing whole-object losses and enhancing the reliability of perception in both synthetic and real-world scenarios.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).