Date of Award

8-2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Mechanical Engineering

Committee Member

Dr. Yue Wang, Committee Chair

Committee Member

Dr. Kapil Chalil Madathil

Committee Member

Dr. Ardalan Vahidi

Committee Member

Dr. John R. Wagner

Abstract

Despite great achievements made in (semi)autonomous robotic systems, human participa-tion is still an essential part, especially for decision-making about the autonomy allocation of robots in complex and uncertain environments. However, human decisions may not be optimal due to limited cognitive capacities and subjective human factors. In human-robot interaction (HRI), trust is a major factor that determines humans use of autonomy. Over/under trust may lead to dispro-portionate autonomy allocation, resulting in decreased task performance and/or increased human workload. In this work, we develop automated decision-making aids utilizing computational trust models to help human operators achieve a more effective and unbiased allocation. Our proposed decision aids resemble the way that humans make an autonomy allocation decision, however, are unbiased and aim to reduce human workload, improve the overall performance, and result in higher acceptance by a human. We consider two types of autonomy control schemes for (semi)autonomous mobile robotic systems. The first type is a two-level control scheme which includes switches between either manual or autonomous control modes. For this type, we propose automated decision aids via a computational trust and self-confidence model. We provide analytical tools to investigate the steady-state effects of the proposed autonomy allocation scheme on robot performance and human workload. We also develop an autonomous decision pattern correction algorithm using a nonlinear model predictive control to help the human gradually adapt to a better allocation pattern. The second type is a mixed-initiative bilateral teleoperation control scheme which requires mixing of autonomous and manual control. For this type, we utilize computational two-way trust models. Here, mixed-initiative is enabled by scaling the manual and autonomous control inputs with a function of computational human-to-robot trust. The haptic force feedback cue sent by the robot is dynamically scaled with a function of computational robot-to-human trust to reduce humans physical workload. Using the proposed control schemes, our human-in-the-loop tests show that the trust-based automated decision aids generally improve the overall robot performance and reduce the operator workload compared to a manual allocation scheme. The proposed decision aids are also generally preferred and trusted by the participants. Finally, the trust-based control schemes are extended to the single-operator-multi-robot applications. A theoretical control framework is developed for these applications and the stability and convergence issues under the switching scheme between different robots are addressed via passivity based measures.

Share

COinS