|Abstract:||We explore the impacts of applying a vision-based picking system, based upon the application of head-mounted displays (HMD), to order picking processes, as a means of eliminating errors. The evaluation in based upon the implementation of our experiment as well as an in-depth analysis of previous works done on this topic.
Our results show the effectiveness of such systems, when compared to established methods, such as paper-picking, voice-picking and using cart mounted displays. The results are less conclusive, when comparing HMD-based systems to light-based systems.
We identify a possible trend of users making mistakes, significantly more frequently while using vision-based order picking in comparison to other methods. However, a vast majority of those are identified by the system and rectified immediately. Error detection is extremely efficient, up to the point that the remaining error frequency surpasses other prevention-based attempts. This characteristic is neither insignificant nor without implications for the efficiency of the work as well as motivation of the worker.
Previous studies on the topic identified the most frequent errors as: mispick, wrong quantity and omission error. Less common were different procedural errors. However, as the setup of our experiment differs greatly, many error types were not possible to make. That is why our experiment identified condition error as the dominant error type.
Tough the consequences for detected errors are mostly seen in the time lost, the impacts of undetected errors can be much more severe in real life application.
That notwithstanding, our experimental design was modified in a way, that allowed for continuous work over a period of four hours. This meant, that a more immediate observation of the impacts of undetected errors, was possible. In this way we observed 60 tasks, that were affected by a total of 33 undetected errors. On average each one added additional 6.77 second to the total completion time.
I contrast to with our expectation, we were unable to conclude, that the frequency of errors falls in conjunction with time and the gained experience. Additionally, we were unable to show any significant difference in error rates when using an AR-enriched picking system as opposed to the simpler vision-picking system.|