When to assist? - Modelling human behaviour for hybrid assembly systems

Over centuries the successful cooperation of foreman and assistant has been an important factor of efficiency, as the mutual understanding of each others actions and intentions facilitates a smooth work flow. Hence, a promising option in the field of automation is to develop models of intuitive and adaptive assistance. An artificial system equipped with such models could improve not only efficiency, but also acceptance and safety in human-robot interaction (HRI). As a first step towards this goal, we investigated the timing of the most common physical interaction occurring in HRI the handing over of items. In an assembly experiment we measured the duration of the different steps. We found that a linear dependency sufficiently describes the relation between the complexity of each working step and its duration. The parameters of the linear model however largely differ between the individual subjects. This linear dependency was used to develop Kalman filters to predict durations and complexities of different assembly steps. We showed that after a short adapting phase we received an accuracy of 18.03%, comparing the predicted assembling duration to the measured duration. The model was integrated into an assistive robot system to demonstrate its robustness and low requirements of the sensory systems.