LEVIO: Lightweight Embedded Visual Inertial Odometry for Resource-Constrained Devices

TL;DR

LEVIO addresses infrastructure-free motion tracking on resource-constrained devices by delivering a fully featured six-DoF visual-inertial odometry pipeline optimized for ultra-low-power multicore systems. It combines established components (ORB features, 8-point motion, EPnP, pose-graph optimization) with hardware-aware optimizations, including a custom linear algebra library and Schur-complement-based marginalization, and validates the solution on a GAP9-based platform (<100 mW, 20 FPS). The authors present both a Python golden-model study and real hardware evaluation on EuRoC, demonstrating a favorable accuracy-efficiency balance (average RMSE around 3.46 m with proper tuning) and real-time capability for micro-drones and AR wearables, with the full implementation open-sourced. This work showcases that high-quality VI/O performance can be achieved on constrained devices, providing a practical route for infrastructure-less localization in lightweight autonomous systems.

Abstract

Accurate, infrastructure-less sensor systems for motion tracking are essential for mobile robotics and augmented reality (AR) applications. The most popular state-of-the-art visual-inertial odometry (VIO) systems, however, are too computationally demanding for resource-constrained hardware, such as micro-drones and smart glasses. This work presents LEVIO, a fully featured VIO pipeline optimized for ultra-low-power compute platforms, allowing six-degrees-of-freedom (DoF) real-time sensing. LEVIO incorporates established VIO components such as Oriented FAST and Rotated BRIEF (ORB) feature tracking and bundle adjustment, while emphasizing a computationally efficient architecture with parallelization and low memory usage to suit embedded microcontrollers and low-power systems-on-chip (SoCs). The paper proposes and details the algorithmic design choices and the hardware-software co-optimization approach, and presents real-time performance on resource-constrained hardware. LEVIO is validated on a parallel-processing ultra-low-power RISC-V SoC, achieving 20 FPS while consuming less than 100 mW, and benchmarked against public VIO datasets, offering a compelling balance between efficiency and accuracy. To facilitate reproducibility and adoption, the complete implementation is released as open-source.

Figures (5)

• Figure 1: Blockdiagram of the pipeline. The overview of the full pipeline is depicted on the left, with a detailed description of the segments to the right. The complete pipeline has been verified using a Python golden model and is deployed on the GAP9 .
• Figure 2: a) Block diagram presenting the hardware resources of the GAP9 and tasks of the pipeline, with the pose graph optimization running on the fabric controller and the mapped to the eight parallel processing cores. Non-parallelized parts of the are handled by the Master Core (M). b) Overview of the physical GAP9 Evaluation Kit (EVK).
• Figure 3: Relative and absolute translation RMS error of the golden model during the parameter sweep (a) and absolute RMSE error for the best performing configuration, when applied to each sub-dataset (b). Data is color-coded for the type of pipeline, and the shapes indicate the dataset it was evaluated on (EuRoC MH01-MH05).
• Figure 4: Average runtime of the parallelized LEVIO pipeline for different scenarios occurring during processing (either or 8-point, and with/without keyframe). Small numbers showing the execution time in ms. On average, the pipeline processes incoming camera frames with a runtime below 50ms, allowing for 20.
• Figure :