Design and Evaluation of a Cost-Effective Autonomous Outdoor Cleaning Robot for Public Park Environments

Abstract

This study presents a practical approach to developing a prototype outdoor cleaning robot with reduced hardware and software complexity. While indoor cleaning robots are widely adopted, their outdoor counterparts remain costly and less accessible due to unstructured environments and increased system complexity. To address this, we retrofitted an off-the-shelf manual sweeper with minimal hardware: a single 3D LiDAR and a compact onboard computer running an open-source autonomy software stack. This LiDAR-centric design reduces hardware and software integration complexity while using the integrated IMU of the LiDAR for motion compensation. We focus on adapting three key autonomy modules, mapping, localization, and planning, for semi-structured park environments characterized by dense vegetation and GNSS-denied conditions. A consistent and loop-closed pointcloud map was generated using a LiDAR SLAM framework. Map-based pose initialization and LiDAR scan matching were evaluated under the tested conditions. Bidirectional path-following behavior was demonstrated through lanelet-based planning adapted for shared narrow paths. A field trial demonstrated autonomous execution of the predefined cleaning route, with a loop-closure residual of 0.072 m, low covariance-derived lateral uncertainty, and a mean deviation from the lanelet-centerline reference trajectory of approximately 0.14 m. These results demonstrate the feasibility of a low-complexity LiDAR-centric autonomy architecture for the tested outdoor cleaning scenario.