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

**Authors:** Chen Li, Wanlei Li, Wenhao Liu, Yixiang Shu, Yunjiang Lou  
**arXiv:** [2504.11073](https://arxiv.org/abs/2504.11073)  
**Date:** 2025-04-15  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2504.11073v1/

## Key points

- Online maps are crucial for autonomous robots to navigate through unfamiliar environments
- Dynamic objects in environments can lead to inaccuracies in maps, hindering localization and path planning processes
- FreeDOM framework developed for dynamic object removal in static map construction
- Framework comprises scan-removal front-end and map-refinement back-end components
- Multi-resolution map structure introduced for rapid computation and effective environment representation
- Advanced raycast enhancement techniques used in scan-removal front-end to enhance free space estimation
- Residual dynamic objects in the map eliminated using incremental free space information in the map-refinement back-end
- Extensive experimental validation on diverse datasets shows superior performance of FreeDOM framework with 9.7% average F1-score improvement

## Summary

In the realm of autonomous robotics, the construction of online maps plays a crucial role in enabling robots to navigate through unfamiliar environments. However, the presence of dynamic objects within these environments can lead to inaccuracies in the map, ultimately hindering localization and path planning processes. To address this challenge, a cutting-edge online dynamic object removal framework for static map construction has been developed, known as FreeDOM (Conservative Free Space Estimation). This innovative framework comprises two main components: a scan-removal front-end and a map-refinement back-end. The , developed by authors Chen Li, Wanlei Li, Wenhao Liu, Yixiang Shu, and Yunjiang Lou, represents a significant advancement in addressing challenges related to  in  for . The framework begins by introducing a multi-resolution map structure that facilitates rapid computation and ensures an effective representation of the environment. In the scan-removal front-end, advanced raycast enhancement techniques are employed to enhance free space estimation and segment  based on the estimated free space regions. Moving on to the map-refinement back-end, residual dynamic objects in the map are further eliminated by leveraging incremental free space information. Extensive experimental validation conducted on diverse datasets such as SemanticKITTI, HeLiMOS, and indoor environments equipped with various sensors has demonstrated the superior performance of the proposed FreeDOM framework. By surpassing traditional visibility-based methods, this framework showcases an impressive average F1-score improvement of 9.7%.