While non-embodied pre-specified planning methods, like simulation, are typically used for validating social robot behaviour, these methods generally take a long time to execute and require multidisciplinary expertise and sustained software development. Besides, the integration of humans in these methods remains complex. To address this, the Dramaturgy for Devices project utilises improvisation theatre as a simulation tool. Previous work of the research group aimed to develop a reconfigurable teleoperatable robot that is used within these improvisation experiments to allow quick iteration between robot behaviours and appearances, to extract valuable insights for robot (interaction) design. However, the system lacked the functional flexibility required to perform a wide variety of healthcare-related scenarios and expressiveness, limiting its effectiveness in generating design insights.
This work explored the design space of detachable modules for performative simulation of hospital scenarios with this improvotype platform. Nine potential modules and five shared control strategies were identified, of which a sound module, allowing two-way communication with sound, enabling telepresence, and a robotic arm module, enabling physical manipulation and gestures, were selected for development. These modules were fully integrated into the improvotype platform, utilising standardised mounting points and a newly developed ROS2 device manager, RViz2 plugin, and robot description logic to enable hot-swappable, automatic reconfiguration without the need for system restarts when reconfiguring the robot. User tests confirmed that the modules can be easily and quickly reconfigured by users with varying expertise. An operational test using the robot in an actual improvisation experiment showed that the system allowed for rapid real-world adaptation of the scenes, enabling interactions that were previously impossible, and identified behavioural and situational factors falling entirely outside the assumptions generated by pre-specified planning methods.
For example, how humans use voice and touch to communicate with the robot during edge-case interactions. During this same experiment, the robotic arm was removed in just 16 seconds to immediately test a different robot embodiment, demonstrating effective use in the real use case. The results underscore that a versatile, performative teleoperation platform can generate valuable design insights that graphical simulations typically do not capture. On top of that, this work bridges a research gap by moving beyond swappable appearance-only parts to hot-swappable mechatronic systems for HRI research robots.