Feasibility of Radar-Based Mapping in LiDAR-Optimized SLAM Frameworks

Abstract

This work investigates the feasibility of integrating 60 GHz FMCW radar sensors within established LiDAR-optimized Simultaneous Localization and Mapping (SLAM) frameworks. The research addresses the question of whether comprehensive preprocessing can convert sparse radar measurements into data streams that are compatible with existing SLAM implementations. A systematic sensor replacement methodology maintains identical SLAM parameters while developing a five-step radar preprocessing pipeline that features statistical target detection, temporal stabilization, angle binning, and interpolation to convert irregular radar detections into uniform laser scan messages that are compatible with unmodified SLAM algorithms. Experimental evaluation is conducted in controlled indoor environments to assess mapping performance for configurations with radar preprocessing, limited LiDAR, and full-coverage LiDAR. Performance metrics include scan quality, temporal stability, convergence properties, and map quality derived from synchronized data streams. The results show that despite comprehensive optimization of preprocessing, lowresolution radar systems only achieve 62-70% of the performance of high-resolution LiDAR systems. Dynamic scenarios show similar performance differences. Map analysis confirms that the occupancy grids generated by radar systems exhibit fragmented geometric representations and significant coverage limitations. The fundamental limitation lies in measurement density, as LiDAR provides thousands of measurements per scan, while radar systems only achieve dozens of detections. Processing techniques cannot compensate for missing data, resulting in an insurmountable information gap for dense mapping applications. While the basic scan quality remains comparable, radar systems have significant shortcomings in terms of map quality and temporal performance. The research concludes that optimizing preprocessing alone cannot overcome the inherent limitations of radar in LiDAR-centric SLAM systems. Successful radar-based SLAM requires specialized algorithmic architectures designed for sparse measurement patterns, rather than sensor replacement approaches within existing dense measurement frameworks.

Date of Award	2025
Original language	English
Supervisor	Florian Eibensteiner (Supervisor)