Autonomous Implicit Indoor Scene Reconstruction with Frontier Exploration

TL;DR

This work addresses autonomous indoor 3D scene reconstruction with implicit neural representations by integrating frontier-based exploration for global coverage and surface-uncertainty-guided implicit reconstruction, accelerated through color-uncertainty-based view selection. A multi-task framework alternates exploration and reconstruction tasks, guided by surface uncertainty and an adaptive mode-switching strategy that balances efficiency and surface quality. The approach employs MonoSDF with surface-uncertainty rendering, an ATSP-based planner, and explicit sampling schemes for frontiers and local surfaces, achieving superior reconstruction quality and planning efficiency across virtual and real-world experiments. The results demonstrate robust global coverage, high-fidelity geometry, and practical applicability for UAV-based indoor mapping, with potential extensions to multi-agent systems and pose-estimation-aware planning.

Abstract

Implicit neural representations have demonstrated significant promise for 3D scene reconstruction. Recent works have extended their applications to autonomous implicit reconstruction through the Next Best View (NBV) based method. However, the NBV method cannot guarantee complete scene coverage and often necessitates extensive viewpoint sampling, particularly in complex scenes. In the paper, we propose to 1) incorporate frontier-based exploration tasks for global coverage with implicit surface uncertainty-based reconstruction tasks to achieve high-quality reconstruction. and 2) introduce a method to achieve implicit surface uncertainty using color uncertainty, which reduces the time needed for view selection. Further with these two tasks, we propose an adaptive strategy for switching modes in view path planning, to reduce time and maintain superior reconstruction quality. Our method exhibits the highest reconstruction quality among all planning methods and superior planning efficiency in methods involving reconstruction tasks. We deploy our method on a UAV and the results show that our method can plan multi-task views and reconstruct a scene with high quality.

Figures (4)

• Figure 1: The pipeline of our proposed method.
• Figure 2: Overview of our method. Views and paths are planned in exploration mode (a) and merged mode (b). Candidate view sampling for each surface cluster is depicted in (c).
• Figure 3: Comparison with different methods. Top: novel view synthesis; Bottom: reconstructed meshes.
• Figure 4: Comparison of trajectories with different methods.