Citation
Oh Songhwai. "Robust Navigation and Tracking in Dynamic Environments". Talk or presentation, 8, September, 2015.

Abstract
With the recent development in robotics, we can expect that more service robots will be assisting us in the near future in places, such as offices, malls, and homes. But, for a robot to coexist with humans and operate successfully in crowded and dynamic environments, a robot must be able to act safely and harmoniously with human participants in the environment. In this talk, I will describe our recent work on robust navigation using Gaussian process regression and robust target tracking using chance-constrained optimization. An autoregressive Gaussian process (AR-GP) motion model is developed to predict the future trajectories of pedestrians using measurements from the partially observable egocentric view of a robot. A robot is controlled in real-time based on predicted pedestrian trajectories using the proposed AR-GP motion controller. In order to make the AR-GP motion model robust against outliers and noises, a structured low-rank matrix approximation method using nuclear-norm regularized l1-norm minimization is developed to approximate kernel matrices. A leveraged non-stationary kernel function is proposed to incorporate both positive and negative training samples, in order to speed up the learning process. Using the AR-GP motion model, we have developed a robust target tracking algorithm based on chance-constrained optimization for a mobile sensor with bounded fan-shaped sensing regions, such that the tracking success probability is guaranteed and the travel distance is minimized. I will also describe experimental results from the proposed algorithms. This talk is based on joint work with Sungjoon Choi, Eunwoo Kim, and Yoonseon Oh.

Electronic downloads

• seminar.2015-09-08.Songhwai_Oh.pdf · application/pdf · 0 kbytes

• HTML

  Oh Songhwai. <a
  href="http://chess.eecs.berkeley.edu/pubs/1110.html"
  ><i>Robust Navigation and Tracking in Dynamic
  Environments</i></a>, Talk or presentation,  8,
  September, 2015.

• Plain text

  Oh Songhwai. "Robust Navigation and Tracking in Dynamic
  Environments". Talk or presentation,  8, September,
  2015.

• BibTeX

  @presentation{Songhwai15_RobustNavigationTrackingInDynamicEnvironments,
      author = {Oh Songhwai},
      title = {Robust Navigation and Tracking in Dynamic
                Environments},
      day = {8},
      month = {September},
      year = {2015},
      abstract = {With the recent development in robotics, we can
                expect that more service robots will be assisting
                us in the near future in places, such as offices,
                malls, and homes. But, for a robot to coexist with
                humans and operate successfully in crowded and
                dynamic environments, a robot must be able to act
                safely and harmoniously with human participants in
                the environment. In this talk, I will describe our
                recent work on robust navigation using Gaussian
                process regression and robust target tracking
                using chance-constrained optimization. An
                autoregressive Gaussian process (AR-GP) motion
                model is developed to predict the future
                trajectories of pedestrians using measurements
                from the partially observable egocentric view of a
                robot. A robot is controlled in real-time based on
                predicted pedestrian trajectories using the
                proposed AR-GP motion controller. In order to make
                the AR-GP motion model robust against outliers and
                noises, a structured low-rank matrix approximation
                method using nuclear-norm regularized l1-norm
                minimization is developed to approximate kernel
                matrices. A leveraged non-stationary kernel
                function is proposed to incorporate both positive
                and negative training samples, in order to speed
                up the learning process. Using the AR-GP motion
                model, we have developed a robust target tracking
                algorithm based on chance-constrained optimization
                for a mobile sensor with bounded fan-shaped
                sensing regions, such that the tracking success
                probability is guaranteed and the travel distance
                is minimized. I will also describe experimental
                results from the proposed algorithms. This talk is
                based on joint work with Sungjoon Choi, Eunwoo
                Kim, and Yoonseon Oh.},
      URL = {http://chess.eecs.berkeley.edu/pubs/1110.html}
  }
  

Posted by Sadigh Dorsa on 9 Sep 2015.
For additional information, see the Publications FAQ or contact webmaster at chess eecs berkeley edu.