By Sascha Feldhorst | Fraunhofer IML – Within the context of the MotionMiners start-up initiative at Fraunhofer Institute for Material Flow and Logistics IML, research demonstrates how machines can support people meaningfully. Automated activity recognition will improve both processes and ergonomics in intralogistics ­– a benefit for companies and employees alike.

High levels of sick leave in logistics show how important it is to improve working conditions in the industry.

Logistics jobs number among the professions with particularly high sick leave rates according to the current health report by the umbrella association of the company health insurance funds in Germany. Employees often suffer from muscular and skeletal disorders. Developments like these show how important it is to improve working conditions for people in logistics. The miracle word is »MotionMining«. This is all about automated activity identification as already known in sports and the leisure industry. In logistics, however, this topic is new and by and large unresearched.

Uses for automatic activity identification are wide-ranging: MotionMining can be used for anything from implementing single analyses to clarify specific questions to long-term learning installations which give recommendations for action based on existing data and which then evaluate their success. As a result, process times in companies decrease and/or workloads for employees are reduced. In addition to these use cases, it is also possible to achieve technical interactions, which, for example, could enable gestures to be used within an order picking system.

Wearables and beacons gather »real« data

The MotionMiners initiative aims to collect movement data in the warehouse, and then to analyze it. Mobile sensors, wearables and beacons – all common in this business – initially help to record »real« process data, for example, recording unhealthy movements or plotting order-picking route times for employees. Up to now, such data is often collected by hand – with the help of a stopwatch or a camera.

Employee data protection and the necessity of these measurements also have to be taken into account and weighed up in each case.

The crux of such a procedure is that manual process gathering can only collect data selectively – otherwise it would be too big an effort. Moreover, it is often very inconvenient for most employees to be filmed during work or be observed by others – especially since the acquired data are also clearly personalized. Company databases also only deliver snapshots. What happens between the entries is completely unknown. To receive real – or more specifically – realistic data, employees are equipped with small ordinary wearables, for example an intelligent bracelet. Movements are recorded and environment information like temperature and light are also identified. The additional use of mini transmitters, so-called beacons, helps to obtain further contextual knowledge about manual processes. This way a complete process in order-picking can be screened without having to carry out long-term manual analyses. All data are gathered anonymously. This is important because in Germany legislature distinctly limits any employee observation – especially if it leads to any sort of individual evaluation of performance. It is necessary to weigh up the need for an employee’s data protection and the necessity of making such measurements in each case.

A specifically developed machine-learning solution is available to evaluate the »real« data acquired in day-to-day practice. It allows data to be analysed automatically by means of mechanical learning algorithms. First of all, relevant movements are identified in the data and afterwards processed with the help of deep neural networks (deep learning). In this way, findings with respect to process efficiency and ergonomics can be obtained and unhealthy movements can be analysed. The data is subsequently summarised and prepared for process analysis. Together with contextual information, the identified movements are allocated to logistic activities, i.e. single process steps. For the first time there is now an efficient alternative for the evaluation of manual activities in intralogistics – and this is only one of many examples of what MotionMining can be used for.

About the author

Sascha Feldhorst is a scientific employee at the Fraunhofer Institute for Material Flow and Logistics IML and one of the heads of MotionMiners, a Fraunhofer IML initiative that is sponsored in the context of the F-Days run by Fraunhofer Venture. More information can be found at