Robotics Seminar: Top-Down vs. Bottom-Up Intelligence: A Robotics Perspective

Abstract: Over the past couple of decades, robotics research has mainly focused on broadening the capabilities of classical automation-style robots and transforming them into intelligent machines which can operate in complex and unstructured environments with minimal human intervention. With recent advances in machine learning, the development of such intelligent robot systems is closer to reality than ever. These advances have been primarily driven by supervised methods that led to tremendous successes in natural language processing, and computer vision. At the moment, there is major research activity on extending the successes of this "top-down" approach to the design and deployment of intelligent robots.

In this talk, I will first give an overview of robot systems we have built over the years, including: ground, surface, and aerial robot teams for environmental and agricultural data gathering; autonomous robots for picking fruit, mowing and weeding pastures and corn fields; and robots for tidying and cleaning our homes. I will then make the case for "bottom-up" approaches needed to make such robots reactive. I will conclude the talk by presenting our work on combining top-down and bottom-up approaches toward building intelligent robots that can operate robustly and safely in everyday environments.

Bio: Volkan Isler is a professor of computer science and engineering at the University of Minnesota. From 2021 to 2023, he was the head of Samsung's AI Center in New York City. He is primarily interested in coupling perception and planning in robotics. Volkan has worked on fundamental algorithmic problems in this domain (pursuit-evasion and sensor planning), and developed field systems for environmental sensing, agricultural and home automation. He is the recipient of the NSF CAREER award, UMN McKnight Land-grant Professorship, and the Institute on the Environment Resident Fellowship. He has served as a member of the IEEE RAS Conference Editorial Board, and as an associate editor for both the IEEE Transactions on Robotics, and IEEE Transactions on Automation Science & Engineering. He is co-founder of Farm-Vision Technologies -- a UMN start-up based on his lab's work on yield mapping for specialty farms.

A pizza lunch for attendees will be available at 12:00 p.m. in Computer Science Building, Room 150/151.