Current work
Programming by demonstration. The cost of (re)configuration of low-volume production is currently too high for automation to be economically viable. One way of lowering the cost of reconfiguration is to enable non-experts to do fast robot programming. Collaborative robots can be programmed using kinesthetic teaching, but the performance of the derived robot program is highly dependent on the type of task and the experience of the person doing the demonstration. Therefore, we aim to make programming by demonstration less sensitive to the demonstrator and make it applicable to high precision assembly tasks where the selection of robot compliance parameters is pivotal for obtaining a satisfactory performance.
Mathematical Modelling of Manufacturing Processes. Models are important to understand different observed phenomena. Also, it provides means for simulating a system’s behavior, and optimize the behavior with respect to some performance measure. Models of assembly processes are currently being used for reinforcement learning of insertion strategies that minimize contact forces and robust towards pose uncertainties. In addition, model-based control is used for layup process of large flexible skins.
Physical Human-Robot Interaction. It can be too expensive to fully automate many industrial processes; hence, it may be appropriate to use the strengths of both robots and humans in a collaborative robot setup. Many aspects need to be considered to facilitate seamless human-robot collaboration; these include safety of collaborative robots when colliding with humans and human ergonomics. Physical human-robot interaction is currently being analyzed in a newly established laboratory featuring e.g. a motion capturing system and a sEMG sensing system.
Future work
Data-Driven Performance Enhancement. Work is currently being conducted on incrementally improving of robot performance based on execution data and to detect abnormalities in robot systems. In addition, efforts are being put on estimating the human compliance during demonstration; it is the intention to transfer the human compliance to the robot. So far, the group has focused on kinesthetic teaching, but this work will be extended to direct teaching, where the full body pose of the human is recorded by a motion capturing system.
Model-Based Interaction Design. We aim at using simulation models of physical human-robot collaboration scenarios to propose strategies that lead to improved ergonomics and efficiency. In addition, time-optimal trajectories will be designed that comply with safety requirements.