Probabilistic Local Navigation and Behavior Model for a Warehouse Robot Following an Order Picker
Year: 2026
Type: article
Abstract: Autonomous carts that follow order pickers and transport collected goods reduce walking time in warehouses. However, existing person-following methods have limitations in handling temporary occlusions, identifying the correct person among similarly dressed workers, and adapting safety distances to localization uncertainty, motion prediction errors, and payload variations. This study proposes a probabilistic framework comprising six interconnected components: the robot and its payload, warehouse environment, perception system, order picker, uncertainty propagation, and adaptive safety constraints. Two algorithms are developed for local motion planning and behavior switching. The proposed approach was evaluated in Gazebo and ROS 2 using a four-wheeled mobile robot equipped with a 2D LiDAR, depth camera, and display. Across 180 trials, the method achieved the highest overall task success rate, 20.0%, compared with 4.4–17.8% for three baseline methods; however, only the difference from the weakest baseline was statistically significant (p = 0.047), whereas the difference from the closest baseline (17.8%) was not. Success reached 53.3% in the basic scenario, while in the scenario with identically dressed workers no method completed the task. The share of time spent following the wrong person was 12.6%, compared with 16.4–22.4% for the baselines. Additional identification cues are proposed to distinguish order pickers wearing identical uniforms.
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7226008732
