Wang, Ker-Jiun
(2020)
Adaptive Control and Cooperative Learning of Symbiotic Behavior of Human-Machine-Interaction.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Building a Human-Robot Symbiotic environment that robots could live side-by-side with humans, perform joint actions to achieve common goals, and further augment human’s existing capabilities is always our dream. Besides numerous benefits, it poses many challenges as well. Most importantly, the robot should have learning and adaptability to coordinate its actions with the human. It should take Human-in-the-Loop co-learning, co-adaptation, and the prediction of mutual consensus behaviors into account to foster a stable closed-loop interaction, such that the robot can have more flexibility to perform a broader range of sensorimotor skills in diverse interaction contexts. Moreover, human and robot need to have a seamless communication interface, with which it can identify a wide spectrum of interaction features, from kinematic trajectories and dynamics profiles, to eye gazes, facial expressions and emotion states, etc., to effectively convey multi-modal intentions.
In this dissertation, we tried to address these challenges by developing a biomimic learning and adaptive control framework, which allows wearable robots to cooperate with humans seamlessly. We used Co-Adaptive Optimal Control and Nonzero-Sum Differential Game to describe the human-machine neuromuscular coordination skills, and utilized iterative Inverse Optimal Control and Inverse Differential Game theory to model the cerebellum cooperative learning procedures. The mathematical derivations of theorems as well as the real human subject research on simulated double-inverted pendulum for human-exoskeleton cooperative balancing task were conducted, where we demonstrated promising results that our decentralized cooperative learning and control model is comparable to the centralized optimal control strategy. In the meanwhile, we have developed a compact, non-obtrusive and ergonomic wearable Human-Machine Interface, which observes the physiological gestures (i.e., eye/facial expressions, hand/body movements, somatosensory stimulations, etc.) with over 95% accuracy, based on our developed Deep Multi-Spectrogram Convolutional Neural Network decoder, to interpret and communicate multi-modal human intentions with the machines. It allows the end-users to interact with the machines seamlessly using nature and intuitive commands with engaging and immersive experiences. Hopefully our developments can lead to the next generation intelligent symbiotic machines that enable us to go beyond existing cognitive and physical limitations, achieving superior performance in motor generation and perceptual capabilities in the near future.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
29 January 2020 |
Date Type: |
Publication |
Defense Date: |
9 October 2019 |
Approval Date: |
29 January 2020 |
Submission Date: |
25 November 2019 |
Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
Number of Pages: |
142 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
Swanson School of Engineering > Bioengineering |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Human-Robot Interaction, Human-Machine Symbiosis, Optimal Control, Adaptive Control, Differential Game Theory, Inverse Optimal Control, Inverse Differential Game, Brain-Computer Interface, Biosignal Processing, Deep Learning, Wearable Device, Exoskeleton |
Date Deposited: |
29 Jan 2020 15:22 |
Last Modified: |
29 Jan 2024 06:15 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/37890 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
|
View Item |