ARMOR: Adaptive Meshing with Reinforcement Optimization for Real-time 3D Monitoring in Unexposed Scenes

TL;DR

Unexposed underground environments pose severe challenges for real-time 3D mapping due to complex geometries and projection-based SDF label errors. ARMOR combines a spatial-temporal geometry smoothing module with a reinforcement learning-based parameter adaptation strategy to perform online, scene-aware meshing with a six-dimensional action space and a composite reward. Across more than 3,000 meters of tunnels, caves, and lava tubes, ARMOR reduces geometric error by 3.96% relative to state-of-the-art baselines while preserving real-time performance, demonstrating robustness across diverse underground environments. The approach enables accurate, automated 3D monitoring for construction safety, robotic inspection, and geospatial analysis in challenging unexposed settings.

Abstract

Unexposed environments, such as lava tubes, mines, and tunnels, are among the most complex yet strategically significant domains for scientific exploration and infrastructure development. Accurate and real-time 3D meshing of these environments is essential for applications including automated structural assessment, robotic-assisted inspection, and safety monitoring. Implicit neural Signed Distance Fields (SDFs) have shown promising capabilities in online meshing; however, existing methods often suffer from large projection errors and rely on fixed reconstruction parameters, limiting their adaptability to complex and unstructured underground environments such as tunnels, caves, and lava tubes. To address these challenges, this paper proposes ARMOR, a scene-adaptive and reinforcement learning-based framework for real-time 3D meshing in unexposed environments. The proposed method was validated across more than 3,000 meters of underground environments, including engineered tunnels, natural caves, and lava tubes. Experimental results demonstrate that ARMOR achieves superior performance in real-time mesh reconstruction, reducing geometric error by 3.96\% compared to state-of-the-art baselines, while maintaining real-time efficiency. The method exhibits improved robustness, accuracy, and adaptability, indicating its potential for advanced 3D monitoring and mapping in challenging unexposed scenarios. The project page can be found at: https://yizhezhang0418.github.io/armor.github.io/

Figures (15)

• Figure 1: Adaptive Meshing with Reinforcement Optimization for Real-time 3D Monitoring in Underground Sites. Our system processes sequential LiDAR frames through three key stages: (1) spatial-temporal normal vector smoothing to enhance geometric consistency, (2) reinforcement learning-based parameter optimization that adapts to local scene characteristics, and (3) high-fidelity mesh reconstruction. This pipeline enables accurate and robust real-time 3D monitoring in complex underground environments, which is critical for construction safety and progress assessment.
• Figure 2: Architecture of ARMOR. The ARMOR pipeline begins with the preprocessing of sequential LiDAR and IMU data, followed by a spatio-temporal geometry smoothing module that leverages consistency constraints to generate high-quality normal estimations. A reinforcement learning agent then analyzes the characteristics of local neural maps from the previous state $T_{i-1}$ and computes a multi-discrete probability distribution to infer optimal mapping parameters. These parameters guide the update of the local neural map and drive the adaptive meshing process. The updated neural representation is then used as the observation for the next timestamp $T_{i+1}$, forming a continuous loop for adaptive reconstruction.
• Figure 3: Scanblock formation process. Trajctory T(t) shows the continuous movement of LiDAR over interval [$t_0,t_K$). In each interval, consecutive scan frames will integrate through coordinate transformation, enhancing point cloud density and geometric completeness for improved normal estimation in the unexposed environment.
• Figure 4: Illustration of our normal-guided sampling strategy for improved reconstruction. Projective distances along the LiDAR ray can introduce inherent SDF label errors in irregular 3D unexposed environments. By integrating enhanced geometry smoothing with normal-guided sampling, our method enables more reliable SDF labels and improves reconstruction quality.
• Figure 5: The architecture of agent network.