Improving safety and productivity in manufacturing by studying human-robot interaction when using collaborative robots.


Flexible machining systems are key production assets in situations requiring reconfiguration under flexible production pressure and failures. The assistance of collaborative robots is transforming the role of operators and increasing the number of supervisory added-value tasks on larger portions of systems.

The project is studying the safety-related human-robot behaviours (movements, layout occupation, voluntary/accidental contacts, near misses, etc) in a prototype machining production setup. Typical situations will include frequent reallocation of human/robot tasks, uneven distribution of human location, and subtask-dependent potential physical interaction with machines.

Project progress

The team has secured a partner and site for experiments: the machining department in a medium-size manufacturing company located in Northern Italy. This company makes extensive usage of flexible manufacturing systems supplied by MCM and robot applications. This site allows the team to record the behaviour of human workers along their interaction with the robot system.

Experiments for tracing and analysing safety-critical behaviours on the real machining shop floor depends on:

  • a suitable set of human-robot collaborative tasks, including repeated, yet varied, routines and frequent changes in workflow/materials configuration
  • a production environment where external instrumentation can be deployed 
  • a robot system which is updated to the needs of production in order to maximise the usage, and consequently the expected body of knowledge

The RECOLL team have identified the best location and parts in the production line together with the end user, establishing the background for the introduction of robot-assisted activities, the expected impact on the behaviour and operations of workers, and the necessary training on safety (including administrative measures to be aware of when sharing the collaborative tasks).

The preparation of the site / work cell allowed the team to consider all intended uses by the operators. This has been coded as the default expected pattern to be recorded with external sensors. A model of the collaborative task, including steps, notable positions, expected occupancy of the layout, etc. has been formalised for tracing the online behaviour of both the robot and the human operator(s). From such records, the team expects to identify the principal components of the regular figures of human-robot collaboration (rate of task sharing, occupancy, distances, contacts, etc.) and the potential deviations, which are hopefully precursors of hazardous situations.

The optimised displacement of the sensors combined with a software tool that tracks human movements provides the work cell with the capability of recognising situations that can lead to undesired risks. Given the robustness of the deployed RAS, in the next step a considerable amount of data related to a group of operators will be collected. The forthcoming analysis will be designed to determine the probability of a hazard and its correlation with an incorrect task execution by the operator or/and an incorrect task planning by the system itself.

Project team