Developing and evaluating a model for human motion to facilitate low degree-of-freedom robot imitation of human movement

Imitation, Human Preference, Mobile Robots, Motion Capture

Kaushik, Roshni ; LaViers ; Amy

关键词: Imitation, Human Preference, Mobile Robots, Motion Capture;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/104951/KAUSHIK-THESIS-2019.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Imitation of human motion is a necessary activity for robots to integrate seamlessly into human-facing environments.While perfect replication is not possible, especially for low degree-of-freedom (DOF) robots, this thesis presents a model for human motion that achieves perceptual imitation. Motion capture data of dyadic interactions was first analyzed to quantify a characteristic of human motion observed in the movement.The leaning of the spine, or verticality, was found to correlate with these movement observations. Verticality was then used to inspire a low-DOF model of human motion using motion capture that can be used to command the movement of simulated robots.Experiments were developed to test users’ perception of the imitation by these 3 and 4-DOF simulated robots of human motion.Verticality was preferred in an initial study over artificially generated motion for the higher DOF robot, Broombot, which was preferred over the lower DOF robot, Rollbot.A study was developed to test the preferences of users when the mapping between human and robot motion was changed for variable human motion.Motion capture-based motion was preferred over artificially generated motion, and a sub-group of respondents who preferred verticality and were more engaged in the survey was found.Since the experiments were performed using motion capture data from a trained ballet dancer, a discussion of the differences between two Indian classical dance styles is included that shows that verticality alone is not representative of all motion and prompts a further analysis to develop socially adaptive robot behavior.In-progress and future work include a hardware implementation that will allow real-time motion capture data to drive simulated and/or physical robots. Menagerie is an in-development performance using the tools developed in this thesis that can include a human with simulated and/or physical robots moving together.