Abstract: Building and deploying autonomous systems in the open world has long been a goal of both the artificial intelligence (AI) and robotics communities. From autonomous driving, to health care, to office assistance, these systems have the potential to transform society and alter our everyday lives. The open world, however, presents novel challenges that question the typical assumptions made by the models and frameworks often used in contemporary AI and robotics. To more reliably handle these challenges in a safe capacity, many autonomous systems deployed in human environments already entail some measure of reliance on human assistance. This thesis aims to address the question: how to best perform planning for autonomous systems that are aware of both their strengths and limitations, and have the ability to seek various forms of external assistance to better complete their tasks.
This thesis therefore proposes two complementary approaches to the problem of reliable autonomy in the open world. First, we propose a novel planning framework called a competence-aware system that enables an autonomous system to reason about its own competence in the form of multiple levels of autonomy during planning. We additionally propose a method for enabling a competence-aware system to improve its competence over time through online model updates based on human feedback. Second, we propose a novel planning model called a semi-observable Markov decision process which models problems in which an autonomous system can either fully observe its state or has no observability at all. Finally, we propose to combine these approaches, extending the competence-aware system to the non-fully observable setting, and discuss some of the challenges faced therein.
Advisor: Shlomo Zilberstein