CLOi-Mapper: Consistent, Lightweight, Robust, and Incremental Mapper With Embedded Systems for Commercial Robot Services

TL;DR

CLOi-Mapper tackles SLAM for commercial service robots operating on constrained embedded hardware by delivering consistent local and global pose estimation across diverse sensor configurations. It combines an extensible, zero-constraint graph generation, a semi-real-time Bayesian-like global pose estimator anchored to prior optimizations, and a memory-efficient back-end with temporal nodes and pruning via information- and geometry-based weights. The approach is demonstrated on real-world platforms (e.g., AirStar and R9 cleaning robot) across in-home and large indoor environments, achieving stable trajectories and robust maps with embedded resource usage within tight limits. The work offers practical, mass-production–friendly SLAM that generalizes across hardware variations and sensor suites, with clear potential for extending to multi-robot SLAM in commercial settings.

Abstract

In commercial autonomous service robots with several form factors, simultaneous localization and mapping (SLAM) is an essential technology for providing proper services such as cleaning and guidance. Such robots require SLAM algorithms suitable for specific applications and environments. Hence, several SLAM frameworks have been proposed to address various requirements in the past decade. However, we have encountered challenges in implementing recent innovative frameworks when handling service robots with low-end processors and insufficient sensor data, such as low-resolution 2D LiDAR sensors. Specifically, regarding commercial robots, consistent performance in different hardware configurations and environments is more crucial than the performance dedicated to specific sensors or environments. Therefore, we propose a) a multi-stage %hierarchical approach for global pose estimation in embedded systems; b) a graph generation method with zero constraints for synchronized sensors; and c) a robust and memory-efficient method for long-term pose-graph optimization. As verified in in-home and large-scale indoor environments, the proposed method yields consistent global pose estimation for services in commercial fields. Furthermore, the proposed method exhibits potential commercial viability considering the consistent performance verified via mass production and long-term (> 5 years) operation.

Figures (12)

• Figure 1: (a) Illustration of our SLAM applications arranged from bottom-left to top-right. Airstar is the world's first commercially available airport guidance robot utilized in an international airport. Although each robot has a different sensor configuration, computational performance, and operational condition, they all utilize the proposed framework. (b)--(c) Visualization of the global pose tracking output in an indoor space. Empirically, it was demonstrated that CLOi-Mapper exhibits less delayed pose tracking performance with the proposed embedded processor ($\leq$ 1 GFLOPS) (depicted in the red dotted rectangle), thus enabling the proposed system to enhance trajectory smoothness by mitigating duplicated trajectories and large corrections. In addition, the colors of nodes represent nodes within the same local map.
• Figure 2: Block diagram presenting an overview of CLOi-Mapper. The dashed box represents the novel and modified blocks proposed by this study. The functionalities of each block are described as follows: (a) Front-end of our framework with visual and LiDAR odometry, (b) semi-real-time global pose tracker with a simplified graph, (c) global mapper with pose-graph optimization, including a temporal node and the novel pruning method.
• Figure 3: Overview of a structurally extensible pose graph with cameras and a 2D LiDAR sensor. Visual and LiDAR odometry parts contain visual features and local maps, respectively, and the two parts are interconnected via the frame nodes set $\mathcal{V}$. Frame nodes comprise a combination of wheel odometry with IMU, visual odometry, and LiDAR odometry.
• Figure 4: Illustration of the proposed zero-constraint concept. (a) Regarding the robot with synchronized cameras, various studies employ image merging or fusion of odometry estimated by each camera. (b) In contrast, we propose a zero-constraint, setting it as a relative pose to enable the synchronization effect without altering the framework.
• Figure 5: Illustration of simplified pose graph generation between the reference pose ($\mathbf{x}_{0}$) and current pose ($\mathbf{x}_{k}$) in global coordinates. (L-R) By utilizing the metric embedding method lim2011online, we transform the original pose graph into the simplified pose graph with previously optimized nodes, odometry $\mathbf{o}_{k}$, measurements $\mathbf{z}_{k-1:k}$, and loop constraints in the SLAM submap ($\mathbf{m}_{s,k}$).