Robot triage and variable autonomy for Human-Robot teaming

Ramesh, Aniketh ORCID: 0000-0002-0469-0024 (2025). Robot triage and variable autonomy for Human-Robot teaming. University of Birmingham. Ph.D.

Ramesh2025PhD.pdf
Text - Accepted Version
Available under License Creative Commons Attribution.
Download (22MB)

Abstract

This thesis addresses the problem of systematically detecting, quantifying and mitigating runtime performance degradation in robots during task execution, through interaction with a human operator who can service robots through level of autonomy switching. Robots are increasingly popular today for executing a wide variety of tasks autonomously. However, robots often face runtime performance degradation (e.g. poor navigation, getting stuck, noisy sensor readings), which can cause them to fail task execution or perform sub-optimally. To continue task execution, it is crucial to detect situations where robots face issues so that operator assistance can be enabled through Variable Autonomy paradigms (i.e., switching between different levels of autonomy on demand). Although operators can constantly monitor a robot’s task execution to detect situations where they need assistance, this is cognitively demanding. Furthermore, as the scale and complexity of robot missions increase, multiple robots will be required to execute tasks simultaneously and work together. This may result in multiple robots simultaneously requiring assistance with task execution, with varying degrees of relevance to the overall mission. In such situations, it is important to optimise the interaction paradigm so as not to overload the limited cognitive capacity of the human operator while not compromising the overall mission performance. AI agents can help with such situations, as they can effectively triage robots needing operator assistance and trigger autonomous recovery behaviours for robots if necessary. To this end, this thesis addresses the problem of human-multi-robot teaming by a) proposing a framework for detecting and quantifying performance degradation during task execution, b) proposing several methods to deal with the detected performance degradation via variable autonomy paradigms and human interactions, and c) systematically evaluating the proposed solutions in disaster response and remote inspection inspired scenarios via systematic HRI experiments with human participants.

Specifically, the first major contribution of this thesis is the`Robot Vitals and Robot Health' framework, which serves as a foundation for Human-Multi-Robot Teaming. Evidence is presented that this framework can be used to estimate the robot's state and quantify its online performance degradation. The second contribution is using this framework to realise different variable autonomy architectures to enable a robot to overcome performance degradation by requesting the operators’ help, self-regulating its autonomy level or blending varying degrees of operator input to improve task performance. Evidence is provided to demonstrate that conveying information about robot performance degradation to the operator through visual cues can significantly influence their approach to servicing the robot. This promotes a more risk-aware strategy, reducing the aggregate risk of robot failure during mission runtime. Lastly, insights from previous experiments are utilized to propose a novel interaction paradigm for Variable Autonomy Multi-Robot Systems, aimed at facilitating robot triage through multi-modal interface sensory cues. Experimental evidence shows that our design can significantly improve task performance and reduce aggregate risk of robot failure, which enhances the ease of use, transparency, and trust in the system without increasing the cognitive workload of the operator. An analysis of the experimental results is then presented, highlighting how human factors such as personality, trust, and user understanding of the system are crucial in improving the performance of Variable Autonomy Multi-Robot Systems. By carefully examining human interaction with robot teams and understanding key components responsible for improving task performance and minimising operator cognitive demand, this thesis serves as a solid methodological foundation for future work in the field of human-multi-robot teaming.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):