FreeDOM: Online Dynamic Object Removal Framework for Static Map Construction Based on Conservative Free Space Estimation

TL;DR

The paper tackles the problem of dynamic objects corrupting maps used for localization and planning in unknown environments. It introduces FreeDOM, an online dynamic object removal framework built on conservative free space estimation, featuring a multi-resolution representation and two main components: a scan-removal front-end that integrates visibility via raycasting and a map-refinement back-end that incrementally cleans the static map using free-space updates. Key innovations include raycast enhancement to recover free space in unobserved directions and a refinement stage that leverages historical observations to remove residual dynamics, yielding state-of-the-art performance on SemanticKITTI, HeLiMOS, and indoor datasets with consistent real-time operation. The approach provides robust, training-free dynamic object removal across diverse sensors and environments, significantly improving static map quality for downstream localization and navigation tasks, with an average F1-score improvement of 9.7% reported.

Abstract

Online map construction is essential for autonomous robots to navigate in unknown environments. However, the presence of dynamic objects may introduce artifacts into the map, which can significantly degrade the performance of localization and path planning. To tackle this problem, a novel online dynamic object removal framework for static map construction based on conservative free space estimation (FreeDOM) is proposed, consisting of a scan-removal front-end and a map-refinement back-end. First, we propose a multi-resolution map structure for fast computation and effective map representation. In the scan-removal front-end, we employ raycast enhancement to improve free space estimation and segment the LiDAR scan based on the estimated free space. In the map-refinement back-end, we further eliminate residual dynamic objects in the map by leveraging incremental free space information. As experimentally verified on SemanticKITTI, HeLiMOS, and indoor datasets with various sensors, our proposed framework overcomes the limitations of visibility-based methods and outperforms state-of-the-art methods with an average F1-score improvement of 9.7%.

Figures (8)

• Figure 1: Comparison between (a) raw point cloud map and (b) map generated by FreeDOM (dynamic objects are highlighted in red).
• Figure 2: Overview of FreeDOM: the multi-resolution map, representing FreeSpace (green voxels) at voxel level and StaticSpace (rainbow-colored voxels) at subvoxel level, is shown at the top of Fig. \ref{['framework']}. Given the LiDAR scan ${}_SP$ and estimated pose ${}^W_ST$, the scan-removal front-end estimates FreeSpace and segments the LiDAR scan. The map-refinement back-end further removes residual dynamic points utilizing incremental FreeSpace.
• Figure 3: Effect of Raycast Enhancement on SemanticKITTI 07: Raycast Enhancement recovers free space information in directions without background, resulting in better free space estimation (green voxels) and fewer residual dynamic points (red points).
• Figure 4: Procedure of the scan-removeal front-end. (a) Free space estimation: integrates visibility information from each scan into FreeSpace using a conservative strategy. (b) Real-time dynamic object removal: assigns DynamicLevel labels to points based on the estimated FreeSpace.
• Figure 5: Procedure of map clearing: Sequence-wise removal is performed utilizing incremental FreeSpace, resulting in a clean and complete static map.