Reinforcement Learning for Intelligent Robotic Systems: Challenges and Current Trends

Add to your list(s) Download to your calendar using vCal

• Alessandro Farinelli, University of Verona, Italy
• Friday 25 June 2021, 10:00-11:00
• https://bham-ac-uk.zoom.us/j/83696622212?pwd=NFJSajF4VVhDL2pKY1oxbXlvNlFEQT09.

If you have a question about this talk, please contact Martin Rudorfer.

Intelligent Robotic Systems are a critical asset in several application scenarios, ranging from agile manufacturing, to environmental monitoring and logistics. A crucial element for Intelligent Robotic Systems to be successfully deployed in such scenarios is the ability to adapt their behaviours to changes in the operational environments, and Reinforcement Learning is a widely used approach to achieve this. In the last years, Reinforcement Learning (and Deep RL) achieved ground-breaking successes in several scenarios (e.g., Games and Video-Games). However, the adoption of RL techniques for robotics is still challenging. In this talk we will consider two key aspects related to RL for Intelligent Robotic Systems. Specifically, we will focus on formal verification approaches for Deep RL, describing a novel approach (i.e., ProVe) that is based on interval algebra and that is designed to verify behavioural properties of a DRL agent. Moreover, we will present an approach for the identification of unexpected decisions in Partially Observable Monte-Carlo Planning. This approach is based on Satisfiability Modulo Theory (SMT) and is designed to analyse POMCP policies by inspecting their traces. For both aspects we will present recent results, current challenges and future directions, highlighting application scenarios where these techniques can have a key impact.

Bio: Alessandro Farinelli is full professor of Computer Science at the University of Verona since December 2019. His research interests focus on developing novel Artificial Intelligence methodologies applied to robotics and cyber-physical systems. In particular, he works on multi-agent coordination, decentralized optimization, reinforcement learning and data analysis for cyber-physical systems. His main scientific achievements are rooted in the development of highly innovative approaches for coordination of intelligent agents. In the last five years, he focused on the development of innovative approaches for Reinforcement Learning (and particularly Deep RL) applied to mobile robots, as well as the design of novel approaches for situation awareness and anomaly detection in Intelligent Systems (including mobile robots and cyber-physical systems). He participated in several national and international research projects in the broad area of Artificial Intelligence and Robotics and in the last five years he managed research grants, as PI, for a total of about 700K Euro. In particular, he has been WP leader (WP4, Smart robotic boats for water quality monitoring) and technical director for the INTCATCH project (H2020, WATER -1-2014/2015, June 2016 — January 2020, http://intcatch.eu/).

This talk is part of the IRLab Seminars: Robotics, Computer Vision & AI series.

This talk is included in these lists:

• IRLab Seminars: Robotics, Computer Vision & AI
• https://bham-ac-uk.zoom.us/j/83696622212?pwd=NFJSajF4VVhDL2pKY1oxbXlvNlFEQT09

Note that ex-directory lists are not shown.