Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

LVI-SAM: Tightly-coupled Lidar-Visual-Inertial Odometry via Smoothing and Mapping (2104.10831v2)

Published 22 Apr 2021 in cs.RO

Abstract: We propose a framework for tightly-coupled lidar-visual-inertial odometry via smoothing and mapping, LVI-SAM, that achieves real-time state estimation and map-building with high accuracy and robustness. LVI-SAM is built atop a factor graph and is composed of two sub-systems: a visual-inertial system (VIS) and a lidar-inertial system (LIS). The two sub-systems are designed in a tightly-coupled manner, in which the VIS leverages LIS estimation to facilitate initialization. The accuracy of the VIS is improved by extracting depth information for visual features using lidar measurements. In turn, the LIS utilizes VIS estimation for initial guesses to support scan-matching. Loop closures are first identified by the VIS and further refined by the LIS. LVI-SAM can also function when one of the two sub-systems fails, which increases its robustness in both texture-less and feature-less environments. LVI-SAM is extensively evaluated on datasets gathered from several platforms over a variety of scales and environments. Our implementation is available at https://git.io/lvi-sam

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Tixiao Shan (11 papers)
  2. Brendan Englot (33 papers)
  3. Carlo Ratti (88 papers)
  4. Daniela Rus (181 papers)
Citations (266)

Summary

An Overview of LVI-SAM: Tightly-Coupled Lidar-Visual-Inertial Odometry via Smoothing and Mapping

Introduction

The paper proposes a new framework labeled LVI-SAM, devised to address the challenges of simultaneous localization and mapping (SLAM) using a multi-sensor approach. This framework synergistically integrates data from lidar, visual, and inertial sensors, enabling real-time state estimation and map generation with substantial accuracy and reliability. A core premise of this research is that an integrated approach can surpass the limitations inherent to SLAM systems that depend solely on single-sensor inputs like lidar or vision.

System Architecture

LVI-SAM comprises two subsystems: a Visual-Inertial System (VIS) and a Lidar-Inertial System (LIS), both structured in a tightly-coupled configuration. This system aims to facilitate robustness across diverse environments, including those with minimal textures or features. The VIS, built on monocular visual information complemented by IMU data, enhances its depth estimation by incorporating lidar inputs. Meanwhile, the LIS employs scan-matching techniques supported by visual odometry's initial guesses, thus refining accuracy through mutual system interdependence.

A factor graph is central to LVI-SAM's construction, underpinning both the VIS and LIS subsystems. It supports the reconciliation of various constraints—IMU preintegration, visual and lidar odometry, and loop closure detection—into a unified global optimization task. This integration is crucial for eliminating drift commonly encountered over time and for ensuring a stable and accurate trajectory in varied operational contexts.

Performance and Robustness

Evaluations across diverse datasets confirm LVI-SAM's capacity to deliver high precision in environments with extensive variations in feature density and scale. LVI-SAM demonstrated resilience in scenarios where either the VIS or LIS would independently fail, underpinning a key advantage of the proposed integrated approach. Their experiments illustrate that LVI-SAM maintains robust functionality even when faced with degraded sensor inputs, leveraging the remaining operational sub-system to sustain reliable state estimation.

The subsystem separation ensures that failures in one sensor do not compromise the combined sensor data interpretation, enhancing navigation accuracy in both structured urban and unstructured open-field environments. These achievements highlight LVI-SAM's potential for deployment in complex SLAM applications where single-sensor approaches exhibit limitations.

Numerical Results and Comparisons

The paper provides comparative analyses showing how LVI-SAM's tightly-coupled design contributes to substantial improvements in end-to-end translational and rotational errors, measured against GPS-based ground truth. LVI-SAM consistently achieves lower root mean square error (RMSE) between estimated and true trajectories compared to stand-alone VIO, LIO, or other competing LVIO solutions.

Notably, experiments utilizing self-collected datasets across different platforms demonstrated LVI-SAM's superior accuracy under varied conditions, including the presence of dense overhead foliage and texture-less open fields. These findings underscore the adaptability of the framework to different SLAM tasks while maintaining computational efficiency conducive to real-time applications.

Implications and Future Directions

The LVI-SAM framework offers a significant contribution to multi-sensor SLAM by promoting enhanced localization accuracy and robustness through strategic sensor integration. In practical robotics and autonomous navigation, where diverse environmental conditions pose a continuous challenge, these insights could inform more resilient SLAM solutions.

Looking forward, further investigation into more advanced fusion and optimization techniques, such as incorporating additional absolute sensor measurements (e.g., GPS, altitude), may enhance LVI-SAM's adaptability to even broader applications. The paper's open-source platform provides a promising basis on which the research community can extend LVI-SAM's capabilities and contribute to the evolving SLAM paradigms.

In conclusion, LVI-SAM represents a discerning enhancement in the lidar, visual, and inertial fusion domain, offering a viable path towards more efficient and reliable robotic localization and mapping solutions.