## The revisiting problem in mobile robot map building: A hierarchical Bayesian approach (2003)

### Cached

### Download Links

- [www.cs.washington.edu]
- [www.cs.washington.edu]
- [www.cs.washington.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | In Proc. of the Conference on Uncertainty in Artificial Intelligence (UAI |

Citations: | 26 - 4 self |

### BibTeX

@INPROCEEDINGS{Stewart03therevisiting,

author = {Benjamin Stewart and Jonathan Ko and Dieter Fox and Kurt Konolige},

title = {The revisiting problem in mobile robot map building: A hierarchical Bayesian approach},

booktitle = {In Proc. of the Conference on Uncertainty in Artificial Intelligence (UAI},

year = {2003},

pages = {551--558},

publisher = {Morgan Kaufmann}

}

### Years of Citing Articles

### OpenURL

### Abstract

We present an application of hierarchical Bayesian estimation to robot map building. The revisiting problem occurs when a robot has to decide whether it is seeing a previously-built portion of a map, or is exploring new territory. This is a difficult decision problem, requiring the probability of being outside of the current known map. To estimate this probability, we model the structure of a ”typical ” environment as a hidden Markov model that generates sequences of views observed by a robot navigating through the environment. A Dirichlet prior over structural models is learned from previously explored environments. Whenever a robot explores a new environment, the posterior over the model is estimated by Dirichlet hyperparameters. Our approach is implemented and tested in the context of multi-robot map merging, a particularly difficult instance of the revisiting problem. Experiments with robot data show that the technique yields strong improvements over alternative methods. 1

### Citations

509 | Factorial Hidden markov models
- Ghahramani, Jordan
- 1997
(Show Context)
Citation Context ...rve discrete views, but rather continuous, noisy versions thereof. In our approach, we determine the frequency counts fi|j using the views that are most likely to have generated the observations. See =-=[1, 16] -=-for approaches dealing with partially observable views. and between the individual Dirichlet priors, we can maximize (3) over the individual priors αj. A rather straightforward derivation similar to ... |

294 | Incremental mapping of large cyclic environments
- Gutmann, Konolige
- 2000
(Show Context)
Citation Context ...termine the likelihood for ”out of map” measurements under the assumption that objects are distributed uniformly, i.e. they assign fixed, identical likelihoods to all observations in unexplored ar=-=eas [13, 4, 12, 8-=-]. However, such approaches ignore valuable information since most environments are structured rather than randomly patched together. The key contribution of this paper is a method for estimatRobot po... |

263 | Rao-Blackwellised Particle Filtering for Dynamic Bayesian Networks
- Doucet, Freitas, et al.
- 2000
(Show Context)
Citation Context ...has to decide whether it came back to a previously explored location, or whether it moves through a similar, unexplored area. Especially mapping approaches based on Rao-Blackwellised particle filters =-=[2, 12, 4]-=- can easily incorporate our structural model. Just like in multi-robot map merging, the model can then be used to assign appropriate probabilities to location hypotheses (particles) in unexplored area... |

195 | A probabilistic online mapping algorithm for teams of mobile robots
- Thrun
- 2001
(Show Context)
Citation Context ...st step, i.e. in deciding whether there is an overlap between the two maps or not. To avoid this decision problem, most existing approaches assume knowledge about the robots’ relative start location=-=s [5, 15, 14]-=-. At the minimum, these techniques require that one robot is known to start in the map already built by the other robot. If we consider the revisiting problem in a Bayesian context, then to make an in... |

191 | Data association in stochastic mapping using the joint compatibility test
- Neira, Tardos
- 2001
(Show Context)
Citation Context ...termine the likelihood for ”out of map” measurements under the assumption that objects are distributed uniformly, i.e. they assign fixed, identical likelihoods to all observations in unexplored ar=-=eas [13, 4, 12, 8-=-]. However, such approaches ignore valuable information since most environments are structured rather than randomly patched together. The key contribution of this paper is a method for estimatRobot po... |

134 | Estimating a dirichlet distribution
- Minka
- 2003
(Show Context)
Citation Context ...he data observed in map l, and ¯ f l |j and ¯αj are the sums over all f l i|j and αij , respectively. The MAP α∗ can be found by maximizing the log of (4) using a conjugate gradients method (se=-=e also [10, 11]-=-). To summarize, the structure of an environment is captured by a collection of multinomial distributions q |j describing the sequence of views observed by a robot as it navigates through the environm... |

104 | Adapting the sample size in particle filters through kldsampling
- Fox
- 2003
(Show Context)
Citation Context ...own to start in the map built by the other robot. In this case, map merging can be solved by localizing one robot in the other robot’s map using a localization approach capable of global localizatio=-=n [6]-=-. To the best of our knowledge, map merging has not been addressed for completely unknown start locations including a chance that the partial maps do not overlap at all. Since the map merging problem ... |

99 | Simultaneous localization and mapping with unknown data association using FastSLAM
- Montemerlo, Thrun
- 2003
(Show Context)
Citation Context ...termine the likelihood for ”out of map” measurements under the assumption that objects are distributed uniformly, i.e. they assign fixed, identical likelihoods to all observations in unexplored ar=-=eas [13, 4, 12, 8-=-]. However, such approaches ignore valuable information since most environments are structured rather than randomly patched together. The key contribution of this paper is a method for estimatRobot po... |

71 | Bayesian Data Analysis. Chapman and Hall/CRC - Gelman, Carlin, et al. - 2004 |

58 | Cooperative concurrent mapping and localization
- Fenwick, Newman, et al.
- 2002
(Show Context)
Citation Context ...st step, i.e. in deciding whether there is an overlap between the two maps or not. To avoid this decision problem, most existing approaches assume knowledge about the robots’ relative start location=-=s [5, 15, 14]-=-. At the minimum, these techniques require that one robot is known to start in the map already built by the other robot. If we consider the revisiting problem in a Bayesian context, then to make an in... |

52 | Dp-slam: Fast, robust simultaneous localization and mapping without predetermined landmarks
- Eliazar, Parr
- 2003
(Show Context)
Citation Context |

52 | A practical, decision-theoretic approach to multi-robot mapping and exploration
- Ko, Stewart, et al.
- 2003
(Show Context)
Citation Context ...tion 4.1 Particle filter for partial map localization The generative model for map merging is implemented using a particle filter [6, 3]. A detailed description of this implementation can be found in =-=[9]. Part-=-icle filters represent posteriors over a robot’s continuous position by sets St = {〈x (i) t , w (i) t 〉 | i = 1, . . . , N} of N weighted samples distributed according to the posterior. Here eac... |

26 |
A hierarchical Dirichlet language model. Natural language engineering
- MacKay, Peto
- 1995
(Show Context)
Citation Context ...i | vt−1 =j, αj, f|j) = p(q|j |αj, f|j) qi|j dq|j � = Dirichlet(q|j | αj + f|j) qi|j dq|j = αij + fi|j � i ′ αi ′ j + fi ′ , (2) |j where (2) follows from the properties of the Dirich=-=let distribution [10]. T-=-hus, the prior and the frequency counts are sufficient statistics for the posterior over the parameters of our structural model. The individual αij ’s are often referred to as prior samples, since ... |

25 | H.: Towards multi-vehicle simultaneous localisation and mapping
- Williams, Dissanayake, et al.
- 2002
(Show Context)
Citation Context ...st step, i.e. in deciding whether there is an overlap between the two maps or not. To avoid this decision problem, most existing approaches assume knowledge about the robots’ relative start location=-=s [5, 15, 14]-=-. At the minimum, these techniques require that one robot is known to start in the map already built by the other robot. If we consider the revisiting problem in a Bayesian context, then to make an in... |

13 | A hierarchical Bayesian Markovian model for motifs in biopolymer sequences
- Xing, Jordan, et al.
- 2002
(Show Context)
Citation Context ...rve discrete views, but rather continuous, noisy versions thereof. In our approach, we determine the frequency counts fi|j using the views that are most likely to have generated the observations. See =-=[1, 16] -=-for approaches dealing with partially observable views. and between the individual Dirichlet priors, we can maximize (3) over the individual priors αj. A rather straightforward derivation similar to ... |