Developing the behavioural repertoire of a robot assistant for professional painters

Abstract

Professional painting is a physically demanding job that frequently leads to long-term health issues. To reduce this physical strain, prior research developed a motorized robot cart assistant. However, this prototype lacked a defined behavioural repertoire, limiting its practical usability regarding control, feedback, and autonomy. This thesis addresses this gap by exploring how this robot can be transformed into a transparent and collaborative tool through appropriate autonomy design and multimodal interaction.

The research followed a Double Diamond design process, combining literature research, a benchmark of existing autonomous systems, and stakeholder input. These insights were subsequently explored in an exploratory field study with professional painters using a Wizard-of-Oz approach. The findings demonstrated a clear user preference for the robot to function as a responsive, predictable tool carrier rather than a proactive autonomous collaborator. Furthermore, the study highlighted the strict necessity for simple, physically grounded interaction over complex digital interfaces, as painters must maintain visual focus on their work. Based on these insights, the project scope was refined to focus specifically on the active work scenario within a constrained, one-dimensional workspace alongside a wall.
A functional motorized prototype was developed to evaluate the proposed interaction concepts in practice. The system incorporated a Time-of-Flight sensor for distance regulation, three distinct behavioural modes (Manual, Continuous Follow, and Segmented Follow), a wearable tactile remote, and a digital dashboard. A validation user study was then conducted, comparing the autonomous prototype against a manual cart baseline using abstract wall-based tasks and NASA-TLX workload assessments. The results showed that autonomous assistance successfully reduced physical effort but redistributed the workload by increasing cognitive demand, as users initially needed to monitor the robot's state. Additionally, the autonomous cart induced a shift in user workflow from sequential task execution to more efficient batch-processing strategies.

To address the observed increase in cognitive load, the interfaces were redesigned to include distinct tactile buttons, a physical mode slider, haptic vibration feedback, and a simplified dashboard. An extended familiarisation study was subsequently conducted to investigate interaction over repeated trials. This follow-up study revealed that the previously elevated mental workload was primarily a temporary learning effect. With repeated exposure, this participant's perceived mental demand dropped significantly, interaction efficiency improved, and they successfully adapted their workflow to naturally integrate the autonomous system. The thesis concludes that effective robotic assistance in this context depends on predictable robot behaviour, explicit state communication, and a shared-control approach that enables simple tactile interaction and immediate physical user override.