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

**Authors:** Zikang Yuan, Ruiye Ming, Chengwei Zhao, Yonghao Tan, Pingcheng Dong, Hongcheng Luo, Yuzhong Jiao, Xin Yang, Kwang-Ting Cheng  
**arXiv:** [2503.22926](https://arxiv.org/abs/2503.22926)  
**Date:** 2025-03-29  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2503.22926v2/

## Key points

- LiDAR-Inertial Odometry (LIO) aims to improve the output frequency of 3D LiDAR systems
- SR-LIO++ system doubles the output frequency on limited hardware like Raspberry Pi 4B
- SR-LIO++ utilizes a sweep reconstruction methodology to enhance LiDAR sweep frequency and generate high-frequency reconstructed sweeps
- Introduces a caching mechanism for intermediate results to reduce redundant processing of common segments in adjacent reconstructed sweeps
- Implements quantized map point management to reduce memory usage and optimize computationally intensive calculations

## Summary

, , , , 


In the field of LiDAR-Inertial Odometry (LIO), addressing the challenge of low acquisition frequency in 3D LiDAR systems to achieve high-frequency output has become a crucial area of research. Real-time performance demands processing each sweep within a reduced timeframe, posing significant challenges for deployment on platforms with limited computational power. To overcome these limitations, the innovative SR-LIO++ system is introduced, capable of doubling the output frequency relative to the input frequency on resource-constrained hardware such as the Raspberry Pi 4B. <dk>LiDAR-Inertial Odometry</dk> is an important area of research that aims to improve the output frequency of 3D LiDAR systems. The newly developed <dk>SR-LIO++</dk> system addresses this challenge by doubling the output frequency on limited hardware like Raspberry Pi 4B. SR-LIO++ utilizes a <dk>sweep reconstruction methodology</dk> to enhance LiDAR sweep frequency and generate high-frequency reconstructed sweeps. This approach effectively reduces redundant processing and decouples processing time from traditional linear dependence on reconstructed sweep frequency. To further optimize efficiency, SR-LIO++ introduces a <dk>caching mechanism</dk> for intermediate results, reducing redundant processing of common segments in adjacent reconstructed sweeps. Additionally, it implements <dk>quantized map point management</dk>, which significantly reduces memory usage and optimizes computationally intensive calculations. The collaborative efforts of authors Zikang Yuan, Ruiye Ming, Chengwei Zhao, Yonghao Tan, Pingcheng Dong, Hongcheng Luo, Yuzhong Jiao, Xin Yang, and Kwang-Ting Cheng have resulted in a groundbreaking advancement in LiDAR-Inertial Odometry technology with SR-LIO++. Extensive experimental evaluations across three distinct computing platforms and four public datasets demonstrate its state-of-the-art accuracy and remarkable 20Hz state output on Raspberry Pi 4B hardware.