# Boundary Exploration of Next Best View Policy in 3D Robotic Scanning

arXiv: https://arxiv.org/abs/2412.10444 · HTML: https://mlsanity.com/paper/2412.10444/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 30 references that are in the map.
- First posted on arXiv 11 Dec 2024.

## Builds on (most cited first, 12 of 12)

- [ShapeNet: An Information-Rich 3D Model Repository](https://mlsanity.com/paper/1512.03012/) (2015)
- [GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction](https://mlsanity.com/paper/2402.16174/) (2024)
- [Supervised Learning of the Next-Best-View for 3D Object Reconstruction](https://mlsanity.com/paper/1905.05833/) (2019)
- [Online Next-Best-View Planner for 3D-Exploration and Inspection With a Mobile Manipulator Robot](https://mlsanity.com/paper/2203.10113/) (2022)
- [A Shadowcasting-Based Next-Best-View Planner for Autonomous 3D Exploration](https://mlsanity.com/paper/2109.09323/) (2021)
- [The Surface Edge Explorer (SEE): A measurement-direct approach to next best view planning](https://mlsanity.com/paper/2207.13684/) (2022)
- [Pred-NBV: Prediction-guided Next-Best-View for 3D Object Reconstruction](https://mlsanity.com/paper/2304.11465/) (2023)
- [Surface Edge Explorer (SEE): Planning Next Best Views Directly from 3D Observations](https://mlsanity.com/paper/1802.08617/) (2018)
- [Predicting the Next Best View for 3D Mesh Refinement](https://mlsanity.com/paper/1805.06207/) (2018)
- [Proactive Estimation of Occlusions and Scene Coverage for Planning Next Best Views in an Unstructured Representation](https://mlsanity.com/paper/2009.04515/) (2020)
- [Bag of Views: An Appearance-based Approach to Next-Best-View Planning for 3D Reconstruction](https://mlsanity.com/paper/2307.05832/) (2023)
- [Semantic-aware Next-Best-View for Multi-DoFs Mobile System in Search-and-Acquisition based Visual Perception](https://mlsanity.com/paper/2404.16507/) (2024)

## Abstract (arXiv)

The Next Best View (NBV) problem is a pivotal challenge in 3D robotic scanning, with the potential to significantly improve the efficiency of object capture and reconstruction. Existing methods for determining the NBV often overlook view overlap, assume a fixed virtual origin for the camera, and rely on voxel-based representations of 3D data. To address these limitations and enhance the practicality of scanning unknown objects, we propose an NBV policy in which the next view explores the boundary of the scanned point cloud, with overlap intrinsically considered. The scanning or working distance of the camera is user-defined and remains flexible throughout the process. To this end, we first introduce a model-based approach in which candidate views are iteratively proposed based on a reference model. Scores are computed using a carefully designed strategy that accounts for both view overlap and convergence. In addition, we propose a learning-based method, the Boundary Exploration NBV Network (BENBV-Net), which predicts the NBV directly from the scanned data without requiring a reference model. BENBV-Net estimates scores for candidate boundaries, selecting the one with the highest score as the target for the next best view. It offers a significant improvement in NBV generation speed while maintaining the performance level of the model-based approach. We evaluate both methods on the ShapeNet, ModelNet, and 3D Repository datasets. Experimental results demonstrate that our approach outperforms existing methods in terms of scanning efficiency, final coverage, and overlap stability, all of which are critical for practical 3D scanning applications. The related code is available at github.com/leihui6/BENBV.