Computer Science Program Presents
Introduction to the CIS Graduate Program and Ongoing Research at the Fordham Robotics and Computer Vision Laboratory
Monday, March 17, 2014
RKC 100
A presentation by Dr. Robert Moniot, Chair, Department of Computer & Information Science, Fordham University and Dr. Damian Lyons, Director, FRCV Lab, Fordham University
The first presentation overviews the Computer and Information Science (CIS) department at Fordham University and introduces the CIS graduate program in Computer Science.
In the second presentation, three pieces of ongoing research at the FRCV Lab will be overviewed: visual homing, multirobot exploration and formal analysis of robot behavior to generate performance guarantees.
Visual homing is a navigation approach first proposed as a model of inspect behavior. Because it requires only visual image comparisons, it is a simple and general approach. However, goal directed motion in the absence of distance information can be error prone. Nirmal & Lyons (2013) proposed a stereocamera based visual homing whose performance improves on that of regular visual homing.
In deploying a team of robots to explore an area for search and rescue or C-WMD missions, it is preferable for the team to spread out and cover the area as quickly as possible. It is difficult to design a simple, decentralized dispersion algorithm that works with a wide range building layouts. Liu and Lyons (2014) developed a simple yet general potential field approach based on the concept of generating a potential in empty space that reflects coverage.
It would be preferable to deploy autonomous teams rather than teleoperated robots to handle C-WMD missions given the potential for widespread and serious damage. However, autonomous robots can behave very unpredictably. Formal verification techniques, such as model-checking, could be applied to this problem, but the requirement parallel activities, time-constrained and probabilistic action, and real-number variables all cause extreme state-space size issues. Lyons and Arkin (2012) propose an approach to verification of behavior-based robot systems based on a process algebra model of recurrence a dynamic Bayesian network for probabilistic filtering. They show that this can be used for efficient verification of performance guarantees and validate the guarantees with extensive experimentation.
For more information, call 845-758-6822, or e-mail [email protected].
Location: RKC 100