Our exhibit offers conference participants a chance to evaluate the ubiquitous Roomba vacuum as an educational resource - particularly, as an inexpensive autonomous robot for validating AI algorithms.
Funded by an NSF DUE award and the commercial venture RoombaDevTools, we will have hardware and software for demos and hands-on trials.
Contact: Zach Dodds, dodds 'at' cs.hmc.edu
Innovative algorithms and software have been developed for anomaly detection in deterministic dynamical time series. These include off-line (training) methods for modeling and on-line (testing) methods for measuring model conformance. Modeling is based on one-class supervised learning of the normal class. Furthermore, models are human comprehensible so they may be manually validated and adjusted. Finally, we require our conformance measurement software to operate in an embedded platform having limited resources. The paper builds on prior algorithmic papers, focusing on the tool kit architecture and case studies. Interface and Control Systems have productized the software.
Contact: Walter Schiefele, walter 'at' interfacecontrol.com