Home > Research > Publications & Outputs > Evolving Fuzzy Classifier for Real-time Novelty...
View graph of relations

Evolving Fuzzy Classifier for Real-time Novelty Detection and Landmark Recognition by a Mobile Robot

Research output: Contribution in Book/Report/ProceedingsChapter


Publication date2007
Host publicationMobile Robots: The Evolutionary Approach
EditorsN Nedja, L Coelho, L Mourelle
Place of PublicationBerlin/Heidelberg
Number of pages30
ISBN (Print)978-3-540-49719-6
<mark>Original language</mark>English

Publication series

NameStudies in Computational Intelligence


In this chapter, an approach to real-time landmark recognition and simultaneous classi¯er design for mobile robotics is introduced. The approach is based on the recently developed evolving fuzzy systems (EFS) method [1], which is based on sub- tractive clustering method [2] and its on-line evolving extension called eClustering [1]. When the robot travels in an unknown environment, the landmarks are auto- matically deteced and labelled by the EFS-based self-organizing classi¯er (eClass) in real-time. It makes fully autonomous and unsupervised joint landmark detec- tion and recognition without using the absolute coordinates (altitude or longitude), without a communication link or any pre-training. The proposed algorithm is re- cursive, non-iterative, incremental and thus computationally light and suitable for real-time applications. Experiments carried out in an indoor environment (an o±ce located at InfoLab21, Lancaster University, Lancaster, UK) using Pioneer3 DX mo- bile robotic platform equipped with sonar and motion sensors are introduced as a case study. Several ways to use the algorithm are suggested. Further investigations will be directed towards development of a cooperative scheme, tests in a realistic outdoor environment, and the presence of moving obstacles. (c) Springer