## XTAG 7'ec&/cal Report. I)ep~rtrnent o[' Computer a.ml hdbrmation Sciences, University or lhmnsylwmia, l)hihuh!lldli~t, PA (2005)

Venue: | In In'ogress Ilindle, D |

Citations: | 10 - 1 self |

### BibTeX

@INPROCEEDINGS{Ranganathan05xtag7'ec&/cal,

author = {Ananth Ranganathan and Frank Dellaert},

title = {XTAG 7'ec&/cal Report. I)ep~rtrnent o[' Computer a.ml hdbrmation Sciences, University or lhmnsylwmia, l)hihuh!lldli~t, PA},

booktitle = {In In'ogress Ilindle, D},

year = {2005}

}

### OpenURL

### Abstract

Probabilistic techniques have become the mainstay of robotic mapping, particularly for generating metric maps. In previous work, we have presented a hitherto nonexistent general purpose probabilistic framework for dealing with topological mapping. This involves the creation of Probabilistic Topological Maps (PTMs), a sample-based representation that approximates the posterior distribution over topologies given available sensor measurements. The PTM is inferred using Markov Chain Monte Carlo (MCMC) that overcomes the combinatorial nature of the problem. In this paper, we address the problem of integrating appearance measurements into the PTM framework. Specifically, we consider appearance measurements in the form of panoramic images obtained from a camera rig mounted on a robot. We also propose improvements to the efficiency of the MCMC algorithm through the use of an intelligent data-driven proposal distribution. We present experiments that illustrate the robustness and wide applicability of our algorithm. 1

### Citations

2014 |
Principal Component Analysis
- Jolliffe
- 2002
(Show Context)
Citation Context ... of each row of the panoramic image and storing only the first few coefficients corresponding to the lower spatial frequencies [12]. While more popular dimensionality reduction techniques such as PCA =-=[9]-=- exist, the drawback of such systems is the need to further preprocess the measurement images in order to obtain rotational invariance. In contrast, the magnitudes of Fourier coefficients in a Fourier... |

448 | FastSLAM: A Factored Solution to the Simultaneous Localization
- Montemerlo, Thrun, et al.
- 2002
(Show Context)
Citation Context ...robustness and wide applicability of our algorithm. 1 Introduction Mapping an unknown and uninstrumented environment is one of the foremost problems in robotics. For this purpose, both metric maps [4]=-=[13]-=- and topological maps [16][2] have been explored in depth as viable representations of the environment. In both cases, probabilistic approaches have had great success in dealing with the inherent unce... |

405 | A robot exploration and mapping strategy based on a semantic hierarchy of spatial representations". Robotics urrd .-l~utunurrruu.s Sgstems
- Kuipers, Byun
- 1991
(Show Context)
Citation Context ...patial connectivity of the environment by representing it as a graph with arcs connecting the nodes that designate significant places in the environment, such as corridor junctions and room entrances =-=[11]-=-. Arguably the hardest problem in topological mapping is the perceptual aliasing problem, which is an instance of 1sFigure 1: Camera rig mounted on the robot to obtain panoramic images the data associ... |

253 | Probabilistic robot navigation in partially observable environments
- Simmons, Koenig
- 1995
(Show Context)
Citation Context ...olution space. [1] uses a second order HMM to model the environment. The use of a limited, multiple-hypothesis space over correspondences through the use of POMDPs is prevalent also in the literature =-=[19]-=-[22]. Others use a non-probabilistic approach to the perceptual aliasing problem by applying a clustering algorithm to the measurements to identify distinctive places, an instance being [10]. Numerous... |

222 | Image segmentation by data-driven Markov chain monte carlo
- Tu, Zhu
- 2002
(Show Context)
Citation Context ...y are prone to failing silently in environments with severe perceptual aliasing. Data-driven proposals have previously been used various fields - for example in Computer Vision for image segmentation =-=[23]-=-, and in Statistics to analyze mixture models [8]. In general, data-driven proposals cause a significant speed-up in the sampling algorithm in cases where the state space being considered is enormous.... |

185 | Topological simultaneous localization and mapping (SLAM): Toward exact localization without explicit localization
- Choset, Nagatani
- 2001
(Show Context)
Citation Context ...ity of our algorithm. 1 Introduction Mapping an unknown and uninstrumented environment is one of the foremost problems in robotics. For this purpose, both metric maps [4][13] and topological maps [16]=-=[2]-=- have been explored in depth as viable representations of the environment. In both cases, probabilistic approaches have had great success in dealing with the inherent uncertainties associated with rob... |

140 |
Occupancy grids : a probabilistic framework for robot percpetion and navigation
- Elfes
- 1989
(Show Context)
Citation Context ...he robustness and wide applicability of our algorithm. 1 Introduction Mapping an unknown and uninstrumented environment is one of the foremost problems in robotics. For this purpose, both metric maps =-=[4]-=-[13] and topological maps [16][2] have been explored in depth as viable representations of the environment. In both cases, probabilistic approaches have had great success in dealing with the inherent ... |

137 |
Combinatorial algorithms
- Nijenhuis, Wilf
- 1979
(Show Context)
Citation Context ...ies is thus equal to the number of set partitions of the set of measurements (and hence, also to the set of all possible clusterings of the set of measurements). This number is called the Bell number =-=[14]-=-, and grows hyper-exponentially with the number of measurements. 3.2 Inferring PTMs using Markov chain Monte Carlo The aim of inference in the space of topologies is to obtain the posterior probabilit... |

133 | Learning topological maps with weak local odometric information
- Shatkay, Kaelbling
- 1997
(Show Context)
Citation Context ...ap representations do not deal with general topological maps, but with the use of Markov decision processes to learn a policy that the robot follows to navigate the environment. Shatkay and Kaelbling =-=[18]-=- use the Baum-Welch algorithm, a variant of the EM algorithm used in the context of HMMs, to solve the aliasing problem for topological mapping. However, this approach is well-known to be prone to loc... |

91 | A split-merge Markov chain Monte Carlo procedure for the Dirichlet Process mixture model
- Jain, Neal
(Show Context)
Citation Context ...th severe perceptual aliasing. Data-driven proposals have previously been used various fields - for example in Computer Vision for image segmentation [23], and in Statistics to analyze mixture models =-=[8]-=-. In general, data-driven proposals cause a significant speed-up in the sampling algorithm in cases where the state space being considered is enormous. In such cases, a normal proposal would provide a... |

87 | Bootstrap learning for place recognition
- Kuipers, Beeson
- 2002
(Show Context)
Citation Context ...iterature [19][22]. Others use a non-probabilistic approach to the perceptual aliasing problem by applying a clustering algorithm to the measurements to identify distinctive places, an instance being =-=[10]-=-. Numerous approaches also exist for the use of local appearance in place recognition, for example [21][3]. However, all these methods are inherently brittle in the sense that they are prone to failin... |

67 | Towards a general theory of topological maps - Remolina, Kuipers |

50 |
Some location problems for robot navigation using a single camera
- Sugihara
- 1988
(Show Context)
Citation Context ...ing a clustering algorithm to the measurements to identify distinctive places, an instance being [10]. Numerous approaches also exist for the use of local appearance in place recognition, for example =-=[21]-=-[3]. However, all these methods are inherently brittle in the sense that they are prone to failing silently in environments with severe perceptual aliasing. Data-driven proposals have previously been ... |

29 | Robust place recognition using local appearance based methods
- Dudek, Jegessur
- 2002
(Show Context)
Citation Context ...a clustering algorithm to the measurements to identify distinctive places, an instance being [10]. Numerous approaches also exist for the use of local appearance in place recognition, for example [21]=-=[3]-=-. However, all these methods are inherently brittle in the sense that they are prone to failing silently in environments with severe perceptual aliasing. Data-driven proposals have previously been use... |

29 |
editors. Markov chain Monte Carlo in practice
- Gilks, Richardson, et al.
- 1996
(Show Context)
Citation Context ... a sample-based approximation to the posterior distribution is used for this purpose. The sample-based approximation, in turn, is computed using the Markov Chain Monte Carlo (MCMC) sampling algorithm =-=[5]-=- that overcomes the combinatorial nature of the state space. The intuitive reason for computing the posterior is to solve the aliasing problem for topologies in a systematic manner. The set of all pos... |

28 | Hybrid simultaneous localization and map building: closing the loop with multi-hypothesis tracking
- Tomatis, Nourbakhsh, et al.
- 2002
(Show Context)
Citation Context ...ion space. [1] uses a second order HMM to model the environment. The use of a limited, multiple-hypothesis space over correspondences through the use of POMDPs is prevalent also in the literature [19]=-=[22]-=-. Others use a non-probabilistic approach to the perceptual aliasing problem by applying a clustering algorithm to the measurements to identify distinctive places, an instance being [10]. Numerous app... |

26 | The revisiting problem in mobile robot map building: A hierarchical Bayesian approach
- Stewart, Ko, et al.
- 2003
(Show Context)
Citation Context ...which is an instance of 1sFigure 1: Camera rig mounted on the robot to obtain panoramic images the data association problem, also variously known as “closing the loop” [6] or “the revisiting problem” =-=[20]-=-. It is the problem of determining whether sensor measurements taken at different points in time correspond to the same physical location. When a robot receives a new measurement, it has to decide whe... |

19 | Loop-closing and planarity in topological map building
- Savelli, Kuipers
- 2004
(Show Context)
Citation Context ...e exist that are capable of dealing with the inference of the posterior distribution over the space of topologies. A recent approach by Remolina and Kuipers [16], improved upon by Savelli and Kuipers =-=[17]-=-, gives an algorithm to build a tree of all possible topological maps that conform to the measurements, but in a non-probabilistic manner . Most instances of previous work extant in the literature tha... |

14 |
A highly efficient fastslam algorithm for generating cyclic maps of large-scale environments from raw laser range measurements
- Haehnel, Burgard, et al.
- 2003
(Show Context)
Citation Context ...he perceptual aliasing problem, which is an instance of 1sFigure 1: Camera rig mounted on the robot to obtain panoramic images the data association problem, also variously known as “closing the loop” =-=[6]-=- or “the revisiting problem” [20]. It is the problem of determining whether sensor measurements taken at different points in time correspond to the same physical location. When a robot receives a new ... |

12 | Inference in the Space of Topological Maps: An MCMC-Based Approach
- Ranganathan, Dellaert
- 2004
(Show Context)
Citation Context ...it has visited previously, or to a completely new location. The aliasing problem is hard because the number of possible choices grows combinatorially with the number of measurements. In previous work =-=[15]-=-, we presented the concept of Probabilistic Topological Maps (PTMs) that deal with the perceptual aliasing problem in a systematic probabilistic manner. A PTM is a probability distribution over the di... |

11 |
Image-based Monte Carlo localisation with omnidirectional images
- Menegatti, Zoccarato, et al.
(Show Context)
Citation Context ...ature [7] of a panoramic image as the appearance measurements in our appearance model. Fourier signatures, which have previously been used in the context of localization using omni-directional vision =-=[12]-=-, are a low-dimensional representation of images using Fourier coefficients. They allow inexpensive matching of images to determine correspondence to physical locations. Further, due to the periodicit... |

6 | Omnidirectional image-based modeling: three approaches to approximated plenoptic representations
- Ishiguro, Ng, et al.
- 2003
(Show Context)
Citation Context ... with the space of all possible clusterings. This is true since the space of topologies is exactly the same as that of all possible clusterings of available measurements. We use the Fourier signature =-=[7]-=- of a panoramic image as the appearance measurements in our appearance model. Fourier signatures, which have previously been used in the context of localization using omni-directional vision [12], are... |

2 |
Place learning and recognition using hidden markov models
- Mari
- 1997
(Show Context)
Citation Context ...variant of the EM algorithm used in the context of HMMs, to solve the aliasing problem for topological mapping. However, this approach is well-known to be prone to local minima in the solution space. =-=[1]-=- uses a second order HMM to model the environment. The use of a limited, multiple-hypothesis space over correspondences through the use of POMDPs is prevalent also in the literature [19][22]. Others u... |