QLIO: Quantized LiDAR-Inertial Odometry

TL;DR

QLIO tackles real-time LiDAR-Inertial Odometry on SWaP-constrained platforms by distributing processing between a coprocessor and a host, and introducing LiDAR-specific quantization along with an adaptive rQ-vector resampling scheme. It implements a Quantized MAP Estimation (QMAP) and structured point-to-plane residual quantization to compress observations while preserving geometry. Open-source implementation demonstrates substantial bandwidth reductions (approximately 14–15×) with preserved localization accuracy on real-world datasets, enabling edge deployment on drones and humanoid robots. By bridging quantization theory with 3D LiDAR geometry, QLIO offers a scalable, efficient solution for bandwidth-limited, real-time LIO in resource-constrained settings.

Abstract

LiDAR-Inertial Odometry (LIO) is widely used for autonomous navigation, but its deployment on Size, Weight, and Power (SWaP)-constrained platforms remains challenging due to the computational cost of processing dense point clouds. Conventional LIO frameworks rely on a single onboard processor, leading to computational bottlenecks and high memory demands, making real-time execution difficult on embedded systems. To address this, we propose QLIO, a multi-processor distributed quantized LIO framework that reduces computational load and bandwidth consumption while maintaining localization accuracy. QLIO introduces a quantized state estimation pipeline, where a co-processor pre-processes LiDAR measurements, compressing point-to-plane residuals before transmitting only essential features to the host processor. Additionally, an rQ-vector-based adaptive resampling strategy intelligently selects and compresses key observations, further reducing computational redundancy. Real-world evaluations demonstrate that QLIO achieves a 14.1% reduction in per-observation residual data while preserving localization accuracy. Furthermore, we release an open-source implementation to facilitate further research and real-world deployment. These results establish QLIO as an efficient and scalable solution for real-time autonomous systems operating under computational and bandwidth constraints.

Figures (6)

• Figure 1: Comparison of Traditional LIO deployment nguyen2021viralnguyen2022ntunguyen2024mcd vs. Proposed QLIO: Enhancing Efficiency with Distributed Processing.
• Figure 2: QLIO system architecture with a dual-processor setup. The coprocessor handles LiDAR preprocessing, residual quantization, and compression, while the host processor performs IMU propagation and quantized state estimation, optimizing efficiency for real-time localization.
• Figure 3: The residual vector space is quantized into a voxel space based on INT-2 quantization. 3D points with the same quantized residuals are grouped into the same hash index. Additionally, the point clouds can be compressed together according to the hash index.
• Figure 4: Quantization of the points and their association point-plane residuals.
• Figure 5: The test result on MCD-NTU dataset. (a) Conventional methods exhibit uniform planar point distributions, while our rQ-resample strategy preserves residual coherence, achieving lower ATE with fewer points. (b) Standard INT8 compression to the raw observation (left) induces discrete local map errors and causes point-plane association failure, whereas residual-aware quantization maintains structural fidelity.