An automated picking workstation for healthcare applications Paolo Piccinini a , Rita Gamberini b,⇑ , Andrea Prati c , Bianca Rimini b , Rita Cucchiara a a Department of Information Engineering, University of Modena and Reggio Emilia, Via Vignolese, 905/b, Modena, Italy b Department of Engineering Sciences and Methods, University of Modena and Reggio Emilia, Via Amendola, 2, Padiglione Morselli, Reggio Emilia, Italy c Department of Design and Planning in Complex Environments, University IUAV of Venezia, Santa Croce 1957, Venezia, Italy article info Article history: Received 17 February 2012 Received in revised form 1 August 2012 Accepted 3 November 2012 Available online 7 December 2012 Keywords: Centralised distribution centre Automated picking workstation Pharmaceuticals Object segmentation Computer vision abstract The costs associated with the management of healthcare systems have been subject to continuous scru- tiny for some time now, with a view to reducing them without affecting the quality as perceived by final users. A number of different solutions have arisen based on centralisation of healthcare services and investments in Information Technology (IT). One such example is centralised management of pharmaceu- ticals among a group of hospitals which is then incorporated into the different steps of the automation supply chain. This paper focuses on a new picking workstation available for insertion in automated phar- maceutical distribution centres and which is capable of replacing manual workstations and bringing about improvements in working time. The workstation described uses a sophisticated computer vision algorithm to allow picking of very diverse and complex objects randomly available on a belt or in bins. The algorithm exploits state-of-the-art feature descriptors for an approach that is robust against occlu- sions and distracting objects, and invariant to scale, rotation or illumination changes. Finally, the perfor- mance of the designed picking workstation is tested in a large experimentation focused on the management of pharmaceutical items. Ó 2012 Elsevier Ltd. All rights reserved. 1. Introduction International healthcare systems are under increasing pressure to reduce waste and eliminate unnecessary costs while still improving the quality of patient care. Healthcare logistics and sup- ply chain management are therefore coming under a great deal of scrutiny from both practitioners and academics (Battini, Faccio, Persona, & Sgarbossa, 2009; Bradley, 2000). For example, Bowman (1997) cites that healthcare supply chain inefficiencies amount to $11 billion (or 48%) of the total annual cost of $23 billion. Among those processes with potential for improvement, pharmaceutical logistics is significant in terms of the resources and costs involved and their effect on the final per- ceived service level (de Vries, 2011). Traditional management is therefore currently under scrutiny and innovative models are being implemented on a continuous ba- sis. For example, some contributions have recently appeared which describe methodologies and the effects of the introduction of cen- tralised drugs management on order issue and receipt, centralised warehouse management, distribution to end users or simply regarding centralised logistics management. In this context it is clear that the new emerging figure of central distributor is appre- ciated for the benefits it brings when introduced into the supply chain, both when managing ward stocks (hence the reference to a Ward Stock Drug Distribution System – WSDDS) and when dis- pensing unit dose (hence the reference to a Unit Dose Drug Dis- pensing System – UDDDS) (Summerfield, 1983). The process of picking items is critical to the performance of distribution centres. This paper therefore focuses on an automated pick-and-place workstation whose operative behaviour is based on innovative computer-vision-driven robotics (as defined in Asadi, 2011) and whose main potentialities are described below:  Different object types and a highly variable appearance: pharma- ceuticals are characterised by highly variable appearance in terms of colour, size, primary packages (i.e. regular-shaped boxes, irregular-shaped boxes with transparents, flowpacks, etc.). The workstation manages different types of object of dif- ferent dimensions and complexity, available for both WSDDS and UDDDS.  Randomly available objects: most of the picking systems con- sider the scenario of well-separated objects, well-aligned on the belt and with synchronised grasping of the objects. Pharma- ceuticals are sometimes characterised by irregular dispositions in a bin or on a conveyor belt, especially when items in flow- packs are handled or UDDDSs are served.  Multiple instances and distractors: while in other image process- ing and computer vision applications the basic objective is to identify the single best instance of the target/query object, with 0360-8352/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.cie.2012.11.004 ⇑ Corresponding author. Tel.: +39 0522 522633; fax: +39 0522 522609. E-mail address: rita.gamberini@unimore.it (R. Gamberini). Computers & Industrial Engineering 64 (2013) 653–668 Contents lists available at SciVerse ScienceDirect Computers & Industrial Engineering journal homepage: www.elsevier.com/locate/caie