AASS Seminar - Enabling Robust Robot Autonomy with Decision-Making under Uncertainty

The research centre AASS arranges a seminar with Bruno Lacerda, University of Oxford, UK.

Remote attendance: https://oru-se.zoom.us/j/65096623293

Abstract

In this presentation, I will argue that the synergy between three factors is critical for creating reliable mission planning algorithms for autonomous robots operating in uncontrolled environments. These factors are (i) utilising historical data gathered online to enhance decision-making under uncertainty models, (ii) implementing principled planning techniques that explicitly reason about the epistemic uncertainty inherent to these data-driven models, and (iii) incorporating rich specifications that go beyond typical expected reward maximisation problems. Developing frameworks that unify these three factors is an open problem in the field of robotics, which is heavily dependent on the specific application domain. I will offer an overview of various works undertaken at the GOALS lab at the Oxford Robotics Institute that consider these three factors in distinct ways. Moreover, I will describe how these works provide a foundation for creating integrated approaches that enable long-term deployment of robots capable of acquiring parametrised environmental models from historical data, plan considering the epistemic uncertainty of such models, and effectively adapt their behaviour to in-mission observations, continuously refining their estimate over the epistemic uncertainty online.

Bio

Bruno Lacerda is a Senior Researcher at the Oxford Robotics Institute, University of Oxford, UK. He received his Ph.D. in Electrical and Computing Engineering from the Instituto Superior Técnico, University of Lisbon, Portugal, in 2013. Between 2013 and 2017, he was a Research Fellow at the School of Computer Science, University of Birmingham, UK. His research focuses on the intersection of decision-making under uncertainty, formal methods, and mobile robotics. In particular, he is interested in using a combination of techniques from learning, planning and model checking to synthesise intelligent, robust and verifiable behaviour, for both single and multi-robot systems.