SR-LIO++: Efficient LiDAR-Inertial Odometry and Quantized Mapping with Sweep Reconstruction

TL;DR

SR-LIO++ tackles the bottleneck of low LiDAR data frequency by integrating sweep reconstruction with a cache-based surface parameter reutilization and a quantized map-point management scheme. The system doubles the effective sweep frequency to $20$ Hz on resource-limited hardware while maintaining state-estimation accuracy comparable to state-of-the-art LIO methods. Key contributions include (1) a surface parameter cache to remove redundant computations across overlapping reconstructed sweeps, (2) an index-table based 8-bit encoding of global map points with integer-domain nearest-neighbor search, and (3) comprehensive experimentation demonstrating substantial memory and compute savings and robust 20 Hz performance on a Raspberry Pi $4$B. Together, these innovations enable practical, high-frequency LIO on embedded platforms, with validated performance on multiple public datasets and a custom hardware platform for real-world urban mapping.

Abstract

Addressing the inherent low acquisition frequency limitation of 3D LiDAR to achieve high-frequency output has become a critical research focus in the LiDAR-Inertial Odometry (LIO) domain. To ensure real-time performance, frequency-enhanced LIO systems must process each sweep within significantly reduced timeframe, which presents substantial challenges for deployment on low-computational-power platforms. To address these limitations, we introduce SR-LIO++, an innovative LIO system capable of achieving doubled output frequency relative to input frequency on resource-constrained hardware platforms, including the Raspberry Pi 4B. Our system employs a sweep reconstruction methodology to enhance LiDAR sweep frequency, generating high-frequency reconstructed sweeps. Building upon this foundation, we propose a caching mechanism for intermediate results (i.e., surface parameters) of the most recent segments, effectively minimizing redundant processing of common segments in adjacent reconstructed sweeps. This method decouples processing time from the traditionally linear dependence on reconstructed sweep frequency. Furthermore, we present a quantized map point management based on index table mapping, significantly reducing memory usage by converting global 3D point storage from 64-bit double precision to 8-bit char representation. This method also converts the computationally intensive Euclidean distance calculations in nearest neighbor searches from 64-bit double precision to 16-bit short and 32-bit integer formats, significantly reducing both memory and computational cost. Extensive experimental evaluations across three distinct computing platforms and four public datasets demonstrate that SR-LIO++ maintains state-of-the-art accuracy while substantially enhancing efficiency. Notably, our system successfully achieves 20Hz state output on Raspberry Pi 4B hardware.

Figures (12)

• Figure 1: (a) Illustration of the intermediate result (i.e., surface parameter) reutilization for overlapping segments in sweep reconstruction. The proposed method employs minimal storage overhead to eliminate redundant processing of overlapping segments across consecutive reconstructed sweeps. (b) Illustration of quantized map point management, which restrict global map points to predefined locations (as illustrated by yellow points), thereby enabling enumeratively encoding 3D points with fewer bits and in turn optimizing computational efficiency and reducing memory overhead. (c) Benefiting from methods (a) and (b), SR-LIO++ is capable of real-time running at the frequency of 20 Hz on Raspberry Pi 4B. (d) We deployed our own platform for data collection and testing in real world. The reconstructed 3D point cloud and the trajectory overlaid on Google Earth are shown in (e) and (f) respectively.
• Figure 2: Comparison of the map point management of SR-LIO and SR-LIO++.
• Figure 3: System Overview of SR-LIO++. The overall system consists of a cloud processing module, a static initialization module, a state prediction module and an iterative state update module. The surface parameter reutilization and quantified map point management are two core contributions, where the former can significantly reduce computational overhead and the latter can reduce both computational and memory overhead simultaneously.
• Figure 4: Illustration of segmented distortion correction, which makes the specific point to be undistorted only once, thereby ensuring the consistency of state estimation.
• Figure 5: Schematic diagram of surface parameter reuse mechanism. (a)-(d) demonstrate the sequential processing steps applied to the new segment. During subsequent state update iterations, the current new segment transitions to the old segment classification, enabling direct retrieval of its corresponding surface parameters from the temporary cache, thereby eliminating redundant computational overhead.