Adaptive Perception and Manipulation for Autonomous Robotic Kitting in Dynamic Warehouse Environments

Manual batch kitting in pharmaceutical manufacturing is labor intensive and prone to quality issues. Automation can improve productivity and reduce risks, but handling dynamic warehouse environments is challenging. This paper studies the feasibility of integrating autonomous industrial mobile manipulators (AIMMs) to enable flexible kitting without infrastructure changes. We propose an adaptive perception and manipulation system for localizing and grasping pharmaceutical containers from variable rack placements. Fiducial markers provide approximate navigation for the mobile base, then point cloud processing precisely estimates container poses for manipulation even under tight spacing and occlusion constraints. The key technical contribution is a segmentation approach leveraging the viewpoint to isolate the target container’s front plane for robust localization. The proposed system was evaluated in a mock pharmaceutical kitting environment where it localized containers with an average translation and orientation error of respectively 6mm and 1.6 degree. Finally, it achieved a success rate of 96.66% for the kitting of individual containers.