Study on Risk Reduction in Localization of Cloud-Supported Autonomous Mobile Robots

抄録

Recently, outdoor robotics applications have increasingly adopted the “cloud robotics” approach, offloading processing to computationally rich locations via cloud communication due to growing task complexity. However, network conditions in outdoor environments are often volatile, leading to significant performance degradation or unstable behavior in robots owing to poor cloud communication. To address this issue, this study proposes combining minimal self-localization capabilities on the robot side with advanced self-localization processing on the cloud side by integrating and interpolating the two. This paper presents the implementation of 2D self-localization on a robot and 3D self-localization on the cloud, clarifies their characteristics, and proposes a fusion method that combines both self-localization techniques to enhance accuracy and robustness against communication failures. Adaptive Monte Carlo localization (AMCL), a standard algorithm for 2D self-localization, was used on the robot side. Two fusion methods—time-varying weighted average (TVWA) and unscented Kalman filter (UKF)—were implemented. It was shown that the root mean squared error (RMSE) could be reduced to 0.096 m for TVWA and 0.094 m for UKF on closed paths within a 3 m × 4 m rectangle, compared to 0.124 m for AMCL. Furthermore, even with random data loss in the self-localization estimation results from Fast-LIO, the RMSE decreased to 0.102 m for TVWA and 0.097 m for UKF. In this case, the coordinate change before and after the time step due to data loss was reduced from 0.408 to 0.263 m for TVWA and 0.108 m for UKF, indicating that the proposed method reduces the sudden coordinate shifts caused by data loss.