PERCEPTION OF SAFETY AND ROBOT INTELLIGENCE IN PHYSICAL HUMAN-ROBOT INTERACTION

Abstract

This thesis focuses on the study of two important aspects of physical human-robot interaction (pHRI): perceived safety– the subjective feeling of safety of a human operator in the physical presence of a robot– and perceived robot intelligence. The first part of the thesis focuses on reviewing published papers with the goal of understanding what factors influence these two aspects. It was found that, in general, factors influencing the perception of safety are human-robot distance, robot speed, direction of approach, robot size and appearance, motion fluency and predictability, communication and smooth contacts. On the other hand, factors influencing the perceived intelligence of a robot are transparency, animacy, trust, human-like appearance and gestures; other aspects such as adaptability are also influencing perceived intelligence not only for robots, but for intelligent agents in general. Habituation also seems to influence perceived intelligence in some cases, causing it to increase. The second part of the thesis is related to experiments to study the influence of the above-mentioned factors. Experiments were run in which a human subject shared the workspace with a collaborative manipulator while carrying out independent tasks. Different algorithms were used to plan the motion of the robot, all of which applied the safety standard known as speed and separation monitoring (SSM), i.e., the robot speed was decreased proportionally to the distance with the human to guarantee that the robot could stop before a collision occurred. One algorithm generated a fixed path (FP) of the robot with SSM-based modulated speed. A second algorithm (based on model predictive control, or MPC in short), kept updating the robot motion based on the current human location, aimed at increasing productivity compared to the FP case. Two variants of these algorithms (namely, FP-HR and MPC-HR) were specifically developed in this thesis, to further decrease the robot speed based on heart rate measurements. The analysis of the experimental results obtained for 48 subjects showed that MPC was perceived as less safe than FP, which in turn was perceived as safer than FP-HR. The first result was expected, as the MPC-generated motion is in general less predictable, while the second was unexpected, and probably due to frequent pauses of the robot in the FP-HR case. In general, it was observed that further reducing robot speed based on heart-rate measurements did not improve perceived safety; this can be explained by the presence of a lightweight collaborative robot and by the application of SSM, factors that made the non-HR variants of the algorithms already perceived as sufficiently safe, so that no improvement was noticed when introducing the HR-based variants of the same algorithms. Also, participants did not find the robot more intelligent when its motion was governed by more complex algorithms; this can be explained by a relative lack of transparency. In other words, participants had no actual insight in the motion planning algorithms, and did not manage to fully understand their differences during experiments; this caused all algorithms to be perceived as equally intelligent. Apart from differences between single algorithms, it was found that habituation improved both perceived safety and perceived intelligence– indeed, each participant interacted with the robot in four subsequent sub-sessions– and that the previous experience of the participants in interacting with robots played no role in their perception of safety and robot intelligence.