WeCo-SLAM: Wearable Cooperative SLAM System for Real-Time Indoor Localization Under Challenging Conditions

Viachaslau Kachurka, Bastien Rault, Fernando Israel Ireta Muñoz, David Roussel, Fabien Bonardi, Jean-Yves Didier, Hicham Hadj-Abdelkader, Samia Bouchafa, Pierre Alliez, Maxime RobinIEEE Sensors Journal 2021 · DOI

TL;DR

This paper introduces WeCo-SLAM, a novel wearable system that combines LiDAR-Inertial SLAM (LI-SLAM) and Visual-Inertial SLAM (VI-SLAM) to achieve robust real-time indoor localization in challenging environments where GPS is unreliable. By enabling these two SLAM systems to communicate and share optimization events, WeCo-SLAM significantly improves localization accuracy and resilience, crucial for applications like emergency response.

What Problem Does This Paper Address?

Accurate and continuous localization is vital for applications like autonomous navigation and emergency response. However, traditional systems often rely on GPS, which is notoriously unreliable or completely unavailable in indoor environments, especially those with challenging conditions like smoke, poor lighting, or complex structures. Existing SLAM (Simultaneous Localization and Mapping) systems, while capable of mapping, can struggle with robustness and drift over time, particularly in dynamic or feature-poor environments. The MALIN Challenge specifically targets this problem, aiming to localize agents in such difficult scenarios. This paper addresses the need for a robust, real-time localization system that can maintain accurate positioning and mapping even when faced with indoor/outdoor transitions, sensor failures, and other environmental challenges that confound single-sensor or single-SLAM approaches.

Methods

The core of WeCo-SLAM is a cooperative SLAM framework that leverages two semi-independent SLAM systems:

  • LiDAR-Inertial SLAM (LI-SLAM): Implemented using an enhanced version of LOAM (Lidar Odometry and Mapping), this system focuses on building dense 3D LiDAR maps. It extracts geometric features (planar and edge points) and uses ICP (Iterative Closest Point) variants for pose estimation and map building.
  • Visual-Inertial SLAM (VI-SLAM): Based on the VINS-Fusion framework, this system utilizes visual and inertial data for robust loop closure detection and relocalization. It employs a sliding window bundle adjustment for efficient pose estimation.

The novelty lies in the communication and cooperation strategy between these two SLAM systems:

  1. Initial Alignment: The system first estimates the extrinsic transformation between the LiDAR and camera sensors.
  2. Pose Fusion: Poses from both SLAMs are fused using an Error State Kalman Filter (ESKF), integrating IMU data for a more robust state estimation.
  3. Quality Indicators: Each SLAM system provides a "quality indicator" reflecting its confidence in its current pose estimation. These indicators are used to weight the contribution of each SLAM during fusion, prioritizing the more reliable system.
  4. Event Propagation: Crucially, loop closure events detected by the VI-SLAM are propagated to the LI-SLAM. This helps correct drift in the LiDAR map, a task that is challenging for LI-SLAM alone.
  5. Failure Recovery: The system also incorporates mechanisms for pose re-initialization, using externally provided data when one SLAM system fails, to maintain continuity.
  6. GPS Integration: When available, GPS data is used to register the relative SLAM map to a global UTM (Universal Transverse Mercator) frame, further improving accuracy and providing a global reference.

Datasets

The paper evaluates the system on a real-time 950m run conducted as part of the MALIN Challenge. This scenario was designed to include various challenging conditions such as:

  • Indoor/outdoor transitions
  • Dark environments
  • Narrow passages
  • Smoke
  • Climbing stairs
  • Repetitive patterns
  • Extreme lighting changes

The paper does not explicitly mention using pre-existing public datasets for training or evaluation, but rather focuses on testing within the context of this specific, challenging real-world scenario.

Key Insights and Ideas

  • Cooperation is Key: The most significant insight is that combining the strengths of different SLAM modalities (LiDAR for dense mapping, Vision for loop closure) through intelligent communication and fusion leads to a system far more robust than individual components.
  • Quality Indicators Drive Fusion: Using self-assessed quality metrics from each SLAM system to dynamically weight their contributions to the fused pose is a critical mechanism for handling varying environmental conditions and sensor performance.
  • Bridging the Gap in Loop Closure: VI-SLAM's inherent strength in loop closure detection is leveraged to improve the often-difficult loop closure task for LI-SLAM, directly addressing a known limitation of LiDAR-only systems.
  • Wearable Form Factor: The system's design as a wearable unit highlights its applicability for human-centric applications like emergency response, where portability and unobtrusiveness are paramount.
  • Challenging Environments are the True Test: The paper emphasizes that standard SLAM struggles in realistic, dynamic, and unpredictable environments, underscoring the need for advanced cooperative approaches.

Key Results

  • Absolute Trajectory Error (ATE): The system achieved an ATE of 0.48% of the total traveled distance (950m), with positional errors never exceeding 1.1%. The root mean square error (RMSE) for position was 3.14 m, and for orientation, it was 8.46°.
  • Pose Update Rate: The system operates at a pose update rate of 15-20 Hz, demonstrating real-time performance.
  • Robustness: The fused trajectory was significantly more consistent and accurate than individual LI-SLAM or VI-SLAM trajectories, especially in challenging sections of the run.
  • Mapping: The system successfully builds a global consistent 3D LiDAR map, which can be post-processed for 3D reconstruction at different levels of detail (LOD0 and LOD1 were demonstrated).

Limitations and Future Work

  • Homogenization of Quality Indicators: The authors suggest that further work could focus on homogenizing the quality indicators across different SLAM systems for even more seamless fusion.
  • Common Optimization: Exploring a common optimization strategy for both SLAM systems could potentially save computational resources.
  • LiDAR Loop Closure: Enhancing the loop closure capabilities of LiDAR-based SLAM remains an open area for future research, which could further benefit the cooperative system.
  • Hardware Constraints: While the system is designed to be wearable, the current hardware setup weighs under 9 kg, which might still be a consideration for prolonged use by emergency responders.
  • Autonomy: The system's autonomy is limited to approximately 1 hour, suggesting a need for improved power management or battery solutions for extended operations.

Summary updated

Abstract

Real-time globally consistent GPS tracking is critical for an accurate localization and is crucial for applications such as autonomous navigation or multi-robot mapping. However, under challenging environment conditions such as indoor/outdoor transitions, GPS signals are partially available or not consistent over time. In this paper, a real-time tracking system for continuously locating emergency response agents in challenging conditions is presented. A cooperative localization method based on Laser-Visual-Inertial (LVI) and GPS sensors is achieved by communicating optimization events between a LiDAR-Inertial-SLAM (LI-SLAM) and Visual-Inertial-SLAM (VI-SLAM) that operate simultaneously. The estimation of the pose assisted by multiple SLAM approaches provides the GPS localization of the agent when a stand-alone GPS fails. The system has been tested under the terms of the MALIN Challenge, which aims to globally localize agents across outdoor and indoor environments under challenging conditions (such as smoked rooms, stairs, indoor/outdoor transitions, repetitive patterns, extreme lighting changes) where it is well known that a stand-alone SLAM will not be enough to maintaining the localization. The system achieved Absolute Trajectory Error of 0.48%, with a pose update rate between 15 and 20 Hz. Furthermore, the system is able to build a global consistent 3D LiDAR Map that is post-processed to create a 3D reconstruction at different level of details.
BibTeX
@article{kachurka2021weco,
  author = {Viachaslau Kachurka and Bastien Rault and Fernando Israel Ireta Muñoz and David Roussel and Fabien Bonardi and Jean-Yves Didier and Hicham Hadj-Abdelkader and Samia Bouchafa and Pierre Alliez and Maxime Robin},
  title = {{WeCo-SLAM: Wearable Cooperative SLAM System for Real-Time Indoor Localization Under Challenging Conditions}},
  journal = {IEEE Sensors Journal},
  year = {2021},
  doi = {10.1109/jsen.2021.3101121},
}
Figure