Human-Interpretable Uncertainty Explanations for Point Cloud Registration
This work addresses uncertainty in point cloud registration for robotics, providing interpretable explanations to improve robustness, though it is incremental as it builds on existing methods with a novel attribution approach.
The paper tackles point cloud registration under uncertainty from sensor noise and occlusion by developing Gaussian Process Concept Attribution (GP-CA), which quantifies and explains uncertainty, outperforming state-of-the-art methods in runtime, sample-efficiency, and accuracy on three datasets and a real-world robot experiment.
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.