## Using the Topological Skeleton for Scalable Global Metrical Map-Building (2004)

Citations: | 31 - 11 self |

### BibTeX

@MISC{Modayil04usingthe,

author = {Joseph Modayil and Patrick Beeson and Benjamin Kuipers},

title = {Using the Topological Skeleton for Scalable Global Metrical Map-Building},

year = {2004}

}

### Years of Citing Articles

### OpenURL

### Abstract

Most simultaneous localization and mapping (SLAM) approaches focus on purely metrical approaches to map-building. We present a method for computing the global metrical map that builds on the structure provided by a topological map. This allows us to factor the uncertainty in the map into local metrical uncertainty (which is handled well by existing SLAM methods), global topological uncertainty (which is handled well by recently developed topological maplearning methods), and global metrical uncertainty (which can be handled effectively once the other types of uncertainty are factored out). We believe that this method for building the global metrical map will be scalable to very large environments.

### Citations

459 | Globally consistent range scan alignment for environment mapping. Autonomous Robots 4:333–349
- Lu, Milios
- 1997
(Show Context)
Citation Context ...detailed global metrical map m using the mode of H(x). A common approach is to use a local optimization technique to align the pose positions with map estimates to converge upon a locally optimal map =-=[26]-=-, [27]. A more principled approach is to run a RaoBlackwellized particle-filtering algorithm, using H(x) as the proposal: p(x, m|z, u) = p(m|x, z, u)·H(x). However, in Section V, we show that in pract... |

429 | A robot exploration and mapping strategy based on a semantic hierarchy of spatial representations
- Kuipers, Byun
- 1991
(Show Context)
Citation Context ...ct topological map. If each place has a reliably perceived label, even with substantial ambiguity (i.e., perceptual aliasing), continued exploration can reliably determine the correct topological map =-=[17]-=-–[19], except in pathologically symmetrical environments. This can be expressed as a topological version of Markov localization [12]. In realistic environments, with rich sensors, the label set is lar... |

413 |
Estimating uncertain spatial relationships in robotics
- Smith, Self, et al.
- 1988
(Show Context)
Citation Context ... the χp as landmarks which are observed one at a time, by a robot taking actions λi between observing the landmarks pi−1 and pi. This is essentially the classic approach of Smith, Self, and Cheeseman =-=[21]-=-. We can also evaluate G(χ) for an arbitrary distribution of F (λ). Note that for a particular value of χ, P (χ|λ, τ) will only be non-zero for a single value of λ, namely when each λi = (⊖χ place(ti−... |

360 |
Sensor fusion in certainty grids for mobile robots
- Moravec
- 1988
(Show Context)
Citation Context ...l uncertainty (which can be handled effectively once the other types of uncertainty are factored out). Our approach builds on many existing methods. To represent metrical maps, we use occupancy grids =-=[13]-=- rather than landmark-based maps [14] in order to use all of the information in the range sensor signal, instead of abstracting the environment to a pre-specified set of landmark types. To build local... |

304 | Incremental mapping of large cyclic environments
- Gutmann, Konolige
- 2000
(Show Context)
Citation Context ...ery large environments. I. INTRODUCTION Many simultaneous localization and mapping (SLAM) methods have the goal of building a global metrical map of the environment within a single frame of reference =-=[1]-=-–[4]. Although these methods have improved dramatically over the past few years, they are still afflicted with the problem of cumulative distortions in the map and robot position, along with the relat... |

277 | D.: A Real-Time Algorithm for Mobile Robot Mapping With Applications to Multi-robot and 3D
- Thrun, Burgard, et al.
- 2000
(Show Context)
Citation Context ...racting the environment to a pre-specified set of landmark types. To build local metrical maps online, we use existing SLAM methods based on Markov localization and implemented using particle filters =-=[15]-=- to represent the pose action model distribution P (xt|ut, xt−1) (used in Figure 1). Local metrical maps avoid difficult localization problems such as closing large loops, so we can use efficient meth... |

274 | The Spatial Semantic Hierarchy - Kuipers - 2000 |

258 | Directed Sonar Sensing for Mobile Robot Navigation
- Leonard, Durrant-Whyte
- 1992
(Show Context)
Citation Context ...ffectively once the other types of uncertainty are factored out). Our approach builds on many existing methods. To represent metrical maps, we use occupancy grids [13] rather than landmark-based maps =-=[14]-=- in order to use all of the information in the range sensor signal, instead of abstracting the environment to a pre-specified set of landmark types. To build local metrical maps online, we use existin... |

204 |
Simultaneous Localization and Mapping with Sparse Extended
- Thrun, Liu, et al.
- 2004
(Show Context)
Citation Context ...ue such as Metropolis-Hastings sampling [22] could be employed. Alternative Gaussian estimation techniques exist that make different tradeoffs among efficiency and soundness guarantees (CPE [23],SEIF =-=[24]-=-, TJTF [25], and CTS [9]). E. Estimating H(x) Given G(χ), we can estimate H(x) using an EKF. While this approach is sound, it requires more computation than necessary to produce a functional estimate ... |

190 | Topological simultaneous localization and mapping (slam): Toward exact localization without explicit localization
- Choset, Nagatani
- 2001
(Show Context)
Citation Context ...hborhoods. IEEE/RSJ CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS), SENDAI, JAPAN, 2004�� �� �� �� �� �� �� To identify and characterize place neighborhoods, we build Voronoi graph descriptions =-=[16]-=- of the scrolling map, looking for nodes in the Voronoi graph whose edges lead off the scrolling map. This trims small terminal edges from the Voronoi graph, and allows us to describe the “local topol... |

176 | FastSLAM 2.0: An improved particle filtering algorithm for simultaneous localization and mapping that provably converges - Montemerlo, Koller, et al. - 2003 |

163 | The unscented particle filter
- Merwe, Freitas, et al.
- 2001
(Show Context)
Citation Context ...n erroneous map. On the other hand, attempting to maintain a distribution over the space of maps is expensive in space and time, and is vulnerable to approximation failures such as particle depletion =-=[5]-=-. A topological map abstracts the environment to a discrete graph in which nodes represent places and edges represent path segments linking the places. The topological map representation provides a ve... |

152 | An atlas framework for scalable mapping
- Bosse, Newman, et al.
- 2002
(Show Context)
Citation Context ...o the current local map and do not accumulate over travel. In [6], we describe our hybrid approach that is based on the Spatial Semantic Hierarchy [7]. The Atlas and CTS frameworks for hybrid mapping =-=[8]-=-, [9] take a similar approach. In this paper, we show how to use an accurate hybrid topological map as a skeleton for building an accurate global metrical map. Our method solves the loop-closing probl... |

142 | Bayesian map learning in dynamic environments
- Murphy
(Show Context)
Citation Context ...of the chain rule for probabilities. This decomposition is valuable since P (m|x, z, u) (map-building given accurate localization) can be computed analytically and incrementally for popular map types =-=[20]-=-. This means that only P (x|z, u) (localization) must be handled carefully. We wish to find a proposal distribution over x which matches the posterior distribution well. First, to include the effect o... |

125 | Thin junction tree filters for simultaneous localization and mapping
- Paskin
- 2002
(Show Context)
Citation Context ...Metropolis-Hastings sampling [22] could be employed. Alternative Gaussian estimation techniques exist that make different tradeoffs among efficiency and soundness guarantees (CPE [23],SEIF [24], TJTF =-=[25]-=-, and CTS [9]). E. Estimating H(x) Given G(χ), we can estimate H(x) using an EKF. While this approach is sound, it requires more computation than necessary to produce a functional estimate of H(x). Gi... |

116 | Robotic exploration as graph construction - Dudek, Jenkin, et al. |

98 | An efficient FastSLAM algorithm for generating maps of large-scale cyclic environments from raw laser range measurements
- Hähnel, Burgard, et al.
- 2003
(Show Context)
Citation Context ...large environments. I. INTRODUCTION Many simultaneous localization and mapping (SLAM) methods have the goal of building a global metrical map of the environment within a single frame of reference [1]–=-=[4]-=-. Although these methods have improved dramatically over the past few years, they are still afflicted with the problem of cumulative distortions in the map and robot position, along with the related p... |

89 | Bootstrap learning for place recognition
- Kuipers, Beeson
(Show Context)
Citation Context ...istinct alternative maps can be concisely represented, and where an efficient topological version of Markov localization, along with topological axioms, can be used to refute incorrect maps [6], [10]–=-=[12]-=-. Metrical uncertainty is factored into independent components rather than being allowed to accumulate along the robot’s travel trajectory. This approach to managing metrical uncertainty means our met... |

73 | Integrating topological and metric maps for mobile robot navigation: a statistical approach
- Thrun, Gutmann, et al.
- 1998
(Show Context)
Citation Context ...ed global metrical map m using the mode of H(x). A common approach is to use a local optimization technique to align the pose positions with map estimates to converge upon a locally optimal map [26], =-=[27]-=-. A more principled approach is to run a RaoBlackwellized particle-filtering algorithm, using H(x) as the proposal: p(x, m|z, u) = p(m|x, z, u)·H(x). However, in Section V, we show that in practice th... |

71 |
Large-scale map-making
- Konolige
- 2004
(Show Context)
Citation Context ...on technique such as Metropolis-Hastings sampling [22] could be employed. Alternative Gaussian estimation techniques exist that make different tradeoffs among efficiency and soundness guarantees (CPE =-=[23]-=-,SEIF [24], TJTF [25], and CTS [9]). E. Estimating H(x) Given G(χ), we can estimate H(x) using an EKF. While this approach is sound, it requires more computation than necessary to produce a functional... |

70 | Towards a general theory of topological maps
- Remolina, Kuipers
(Show Context)
Citation Context ...lly distinct alternative maps can be concisely represented, and where an efficient topological version of Markov localization, along with topological axioms, can be used to refute incorrect maps [6], =-=[10]-=-–[12]. Metrical uncertainty is factored into independent components rather than being allowed to accumulate along the robot’s travel trajectory. This approach to managing metrical uncertainty means ou... |

54 | DP-SLAM: Fast, Robust Simultaneous Localization and Mapping Without Predetermined Landmarks," presented at - Eliazar, Parr - 2003 |

52 | Local metrical and global topological maps in the hybrid Spatial Semantic Hierarchy
- Kuipers, Modayil, et al.
- 2004
(Show Context)
Citation Context ...th segments, incremental localization is highly reliable and current metrical SLAM methods perform extremely well. Errors are restricted to the current local map and do not accumulate over travel. In =-=[6]-=-, we describe our hybrid approach that is based on the Spatial Semantic Hierarchy [7]. The Atlas and CTS frameworks for hybrid mapping [8], [9] take a similar approach. In this paper, we show how to u... |

31 | Using local information in a non-local way for mapping graph-like worlds
- Dudek, Freedman, et al.
- 1993
(Show Context)
Citation Context ...pological map. If each place has a reliably perceived label, even with substantial ambiguity (i.e., perceptual aliasing), continued exploration can reliably determine the correct topological map [17]–=-=[19]-=-, except in pathologically symmetrical environments. This can be expressed as a topological version of Markov localization [12]. In realistic environments, with rich sensors, the label set is larger, ... |

26 |
The Robotics Data Set Repository (Radish), 2003. Available at: http://radish.sourceforge.net
- Howard, Roy
(Show Context)
Citation Context ...s) with 4 distinct local topologies. Considering the 14 travel actions that linked the 15 possible places, we actually close 4 loops of various sizes. The data for this experiment is available online =-=[28]-=-. The robot used in this experiment was an iRobot Magellan Pro with a SICK PLS laser range finder used for perception. Figure 5(a) shows the occupancy grid created when raw odometry is used. Along pat... |

5 |
Graphical Models
- MacKay, Davey
(Show Context)
Citation Context ... a defines the origin. Then, [c]a = [b]a ⊕ [c]b. We also define an inverse operator, [b]a = ⊖[a]b.In a more complicated environment, a sound estimation technique such as Metropolis-Hastings sampling =-=[22]-=- could be employed. Alternative Gaussian estimation techniques exist that make different tradeoffs among efficiency and soundness guarantees (CPE [23],SEIF [24], TJTF [25], and CTS [9]). E. Estimating... |

4 |
constant-time
- Leonard, Newman, et al.
- 2003
(Show Context)
Citation Context ... current local map and do not accumulate over travel. In [6], we describe our hybrid approach that is based on the Spatial Semantic Hierarchy [7]. The Atlas and CTS frameworks for hybrid mapping [8], =-=[9]-=- take a similar approach. In this paper, we show how to use an accurate hybrid topological map as a skeleton for building an accurate global metrical map. Our method solves the loop-closing problem in... |

2 |
Loop-closing and planarity in topological map-buiding
- Savelli, Kuipers
(Show Context)
Citation Context ...p, we use “local topologies” to build a tree of all possible topological maps consistent with exploration experience [6], pruning the tree when the maps are inconsistent with topological axioms [10], =-=[11]-=-. The axioms are expressed in a nonmonotonic logic, so prioritized circumscription provides a simplicity-based preference ordering over remaining consistent maps [10]. In future work, we plan to incor... |