Human-Interpretable Uncertainty Explanations for Point Cloud Registration

2025-09-23

AI Summary

Point cloud registration suffers from severe uncertainty under sensor noise, pose estimation errors, and partial occlusion—leading to degraded performance of conventional methods such as ICP. To address this, we propose GP-CA (Gaussian Process Concept Attribution), the first interpretable uncertainty modeling framework specifically designed for point cloud registration. GP-CA integrates Gaussian process regression, concept attribution analysis, and active learning to jointly quantify overall registration uncertainty, decompose it into contributions from specific error sources (e.g., noise, occlusion), and adaptively discover previously unknown uncertainty patterns. Extensive experiments on three public benchmarks and a real-world robotic platform demonstrate that GP-CA significantly outperforms state-of-the-art approaches: it achieves higher registration accuracy, faster runtime, and superior sampling efficiency. Crucially, it enhances robotic perception robustness and enables effective failure recovery by providing actionable, source-aware uncertainty estimates.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

In this paper, we address the point cloud registration problem, where well-known methods like ICP fail under uncertainty arising from sensor noise, pose-estimation errors, and partial overlap due to occlusion. We develop a novel approach, Gaussian Process Concept Attribution (GP-CA), which not only quantifies registration uncertainty but also explains it by attributing uncertainty to well-known sources of errors in registration problems. Our approach leverages active learning to discover new uncertainty sources in the wild by querying informative instances. We validate GP-CA on three publicly available datasets and in our real-world robot experiment. Extensive ablations substantiate our design choices. Our approach outperforms other state-of-the-art methods in terms of runtime, high sample-efficiency with active learning, and high accuracy. Our real-world experiment clearly demonstrates its applicability. Our video also demonstrates that GP-CA enables effective failure-recovery behaviors, yielding more robust robotic perception.

Problem

• Addressing point cloud registration failures caused by sensor noise and occlusion
• Quantifying and explaining registration uncertainty by attributing error sources
• Developing active learning to discover new uncertainty sources in real scenarios

Innovation

• GP-CA attributes uncertainty to known error sources
• Active learning discovers new uncertainty sources
• Outperforms methods in runtime and sample-efficiency