BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/186
DTSTAMP:20230914T125913Z
SUMMARY:Gaussian Process Modeling of Large Scale Terrain
DESCRIPTION:Speaker: Srihari Vasudevan\nThe University of Sydney\nAustralia
 n Centre for Field Robotics\nRose Street Building J04\nNSW\n\nAbstract: \n
 This talk will focus on the problem of large scale multi-sensor multi-data
 -set sensor based perception and sensor fusion. The application context of
  this study is large scale terrain modeling for mining automation.\n\nBuil
 ding a model of large scale terrain (5 sq km) that can adequately handle u
 ncertainty and incompleteness of sensor data in a statistically sound way 
 is a challenging problem. Most contemporary representations are not equipp
 ed to model spatially correlated data and typically treat data as being st
 atistically independent. To obtain a comprehensive model of such terrain\,
  typically\, multiple sensory modalities as well as multiple data sets are
  required. This necessitates sensor fusion.\n\nIn order to address these i
 ssues\, this work proposed the use of Gaussian processes (GP's) as models 
 of large scale terrain. The model naturally provided a multi-resolution re
 presentation of space\, incorporated and handled uncertainties aptly and c
 oped with incompleteness of sensory information. Gaussian process regressi
 on techniques were applied to estimate and interpolate (to fill gaps in oc
 cluded areas) elevation information across the field. A single non-station
 ary (neural network) Gaussian process was shown to be powerful enough to m
 odel large and complex terrain\, effectively handling issues relating to d
 iscontinuous data.\n\nExperiments were performed on large scale 3D data se
 ts taken using GPS and laser scanners from a mining scenario. Extensive st
 atistical performance evaluation of the technique was performed through cr
 oss validation experiments on the aforementioned data sets. These experime
 nts also compared the proposed modeling approach with most other well know
 n interpolation and representation methods. The outcome of these compariso
 n and benchmarking experiments was that the proposed approach will perform
  as well as grid based techniques or triangulated irregular networks (TIN'
 s) for dense\, relatively flat laser scanner data sets\, however\, for com
 plex and/or sparse data sets\, the proposed Gaussian process modeling appr
 oach will significantly outperform grid based approaches using most standa
 rd interpolation techniques as well as TIN's using triangle based interpol
 ation techniques. \n\nThis work then proposed two approaches to data fusio
 n using Gaussian processes - one based on Heteroscedastic GP's and the oth
 er based on Dependent GP's. The approach based on heteroscedastic GP's mod
 eled the different data sets as different noisy samples of a common underl
 ying terrain. Dependent GP based data fusion modeled each data set using a
  separate GP and learnt spatial correlation information between different 
 GP's through auto and cross correlations. A key novelty of this work was t
 he derivation and use of non-stationary kernels for multi-task problems wi
 th dependent Gaussian processes. The work based on dependent GP's has also
  successfully demonstrated the simultaneous modeling/prediction of multipl
 e properties of the terrain (terrain elevation and color).\n\nTo enable th
 e approach to cope with multiple large-scale data sets\, GP approximations
  were developed for both the learning and inference stages. A local approx
 imation method based on a  moving window  methodology and implemented usin
 g KD-Trees was proposed for both GP learning and inference. Further\, a bl
 ock learning approach to GP learning was proposed which guaranteed the suc
 cessful use of this approach in resource constrained systems. These approx
 imation methods enabled the approach to handle large data sets\, thereby a
 ddressing its scalability issues. \n\nABOUT THE SPEAKER: \n\nDr Shrihari V
 asudevan has a BE in Computer Science and Engineering from the University 
 of Madras (2002)\, an MS in Computer Science / Intelligent Robotics from t
 he 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 / lecturer at the Australian Centre for Fi
 eld Robotics\, The University of Sydney. His research interests may be sum
 marized as the modeling and mining of sensor data. Specifically\, he is in
 terested in sensor based perception\, sensor fusion\, machine learning and
  pattern recognition towards developing intelligent robots and systems.\n
URL:https://www.tcs.tifr.res.in/web/events/186
DTSTART;VALUE=DATE:20110419
LOCATION:A-212 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR