Dr. Shrihari Vasudevan, Australian Centre for Field Robotics, The University of Sydney.


In applications such as space-exploration, mining or agriculture automation, modeling the underlying resource is a fundamental problem. For such applications, an efficient, flexible and high-fidelity representation of the quantity of interest is critical. The key challenges in realizing this are that of dealing with the problems of uncertainty and incompleteness. Uncertainty and incompleteness are virtually ubiquitous in any sensor based application as sensor capabilities are limited. The problem is magnified in a field automation scenario due to sheer scale of the application. Incompleteness is a major problem in any large scale modeling endeavor as sensors have limited range and applicability. A more significant contributor to this issue is the that of cost - sampling and collecting such data can be expensive. Data is typically collected through various sensors/processes of widely differing characteristics and consequently lead to different kinds of information. Often the resource is characterized by numerous quantities (for example, soil composition in terms of numerous elements). These quantities often are correlated. Given these issues, there is a need for representation and data-fusion methods that can handle spatially correlated, incomplete and uncertain data. Gaussian processes provides a solution for these problems.

This tutorial will address the problems of (1) modeling using Gaussian processes and (2) data fusion using Gaussian processes. The former will introduce Gaussian processes, detail how they can be used to model data and show results of some benchmarking experiments conducted. The latter will elaborate on different approaches to fusing multiple multi-sensor data sets using Gaussian processes. It will demonstrate information fusion using both homogeneous and heterogeneous data and will also provide some interesting benchmarking results that compare the different approaches. Experimental results using large scale (spanning several sqkm and/or having several hundred thousand data) sensor data will be presented. The methods introduced in this tutorial are generic and may be applied in any domain.

About the Speaker

Dr. Shrihari Vasudevan has a BE in Computer Science and Engineering from the University of Madras, India (2002), an MS in Computer Science / Intelligent Robotics from the University of Southern California, USA (2004) and a DSc in Intelligent Robotics from the Swiss Federal Institute of Technology Zurich (2008). He is currently a Research Fellow at the Australian Centre for Field Robotics, The University of Sydney. His research interests span perception and learning for robotics and intelligent systems. Specifically, he is interested in sensor based perception (statistical modeling, mapping, representations, pattern recognition/detection/classification and predictive-modeling/forecasting), data fusion and machine learning towards developing intelligent robots and systems. For more information, see .