Safe and reliable decision-making is critical for long-term deployment of autonomous systems. Despite the recent advances in artificial intelligence and robotics, ensuring safe and reliable operation of human-aligned autonomous systems in open-world environments remains a challenge because these systems often operate based on incomplete information. In this talk, I will present an overview of some of our recent efforts in mitigating the undesirable impacts arising due to model incompleteness. First, I will present techniques to overcome Markovian and non-Markovian negative side effects. Second, I will present an approach for reward alignment using explanations. Finally, I will present a technique to maintain and restore safety of autonomous systems using meta-reasoning.
Sandhya Saisubramanian is an Assistant Professor in EECS at ¾«¶«Ó°ÊÓ State University. Her research focus is on reliable decision making in single and multiple agents that operate in fully and partially observable open-world environments. She is a recipient of the Outstanding Program Committee member award at the ICAPS 2022 and a Distinguished Paper award at IJCAI 2020. She received her Phd from the University of Massachusetts Amherst.