Human Prediction of Robot's Intention in Reach Movements

Humans can predict another person’s intentions from observed movement patterns when they are working together. This intentions prediction allows team members to plan and perform actions in anticipation of other team members. However, the prediction of team members’ intentions does not come naturally in human-robot teaming. The first objective of this dissertation research is to examine human prediction of robot’s intentions by studying the effects of optimal feedback control laws in a robotic arm on its predictability and perceived human-likeness. Three in-laboratory studies were conducted to examine human prediction of robot’s intentions by studying the effects of optimal feedback control laws in a robotic arm on its predictability and perceived human-likeness. These three studies differ in either the number of targets or path planning constraints. Participants observed a robotic arm as it moved toward an object on a shelf. The results showed that low energy expenditure may enable the participants to predict the robot’s target quicker. The speed of predictions was significantly affected by the trajectory characteristics of the end-effector. To reach a more diverse group of participants, online studies with online crowdworkers were also proposed. This also brings up a question about the impact on experimental research results introduced by switching the format of interaction from observing a robot’s physical movements to watching their videos. This question led to the second objective of the dissertation. An additional study was conducted and compared to one of the in-laboratory studies to investigate the similarities and differences in task quality, subjective experience, motivation, and perceived payment fairness between observing a robot’s physical movements and watching their videos in the context of HRI judgment tasks. The findings showed that participants rated the robot’s physical movement as less human-like and less life-like, but reported feeling safer, compared to watching the robot’s movement through videos. Observing a robot’s physical movements yielded better task quality, in terms of accuracy and the desired response time, than watching their video recordings. Additional studies were conducted to investigate similarities and differences between an in-laboratory study with college students as participants and online studies. The online studies utilized an online crowdsourcing platform, on which online crowdworkers were recruited to observe the movements of a robot. The findings revealed that students may produce higher-quality data than crowdworkers. Students were also motivated to complete a study out of interest while the crowdworkers were more focused on compensation. The use of an online crowdsourcing platform also supports the investigation of a research gap in macroergonomics related to the compensation issue on online crowdsourcing platforms, including compensation scheme and fairness and their effects on task performance and subjective experience. This investigation forms the third objective of the dissertation. Four studies were conducted on Amazon Mechanical Turk to explore the effects of participant location, payment method, and payment rate on online crowdwork task quality, subjective experience, motivation, and perceived payment fairness results showed that participant location affects the rate that participants followed written task descriptions. Task accuracy was comparable between the quota and piece-rate payment methods. High-paying tasks were perceived as more fair, however, it did not lead to a higher number of completed tasks in the piece-rate condition. A lower-paying task in the quota condition and a higher-paying task in the piece-rate condition attracted more fraudulent participants.PhDIndustrial & Operations EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/178137/1/tsorat_1.pd