## Hybrid simultaneous localization and map building: closing the loop with multi-hypothesis tracking

### Cached

### Download Links

Venue: | In Proc. 2002 IEEE Intl. Conf. on Robotics & Automation |

Citations: | 28 - 0 self |

### BibTeX

@INPROCEEDINGS{Tomatis_hybridsimultaneous,

author = {Nicola Tomatis and Illah Nourbakhsh},

title = {Hybrid simultaneous localization and map building: closing the loop with multi-hypothesis tracking},

booktitle = {In Proc. 2002 IEEE Intl. Conf. on Robotics & Automation},

year = {}

}

### Years of Citing Articles

### OpenURL

### Abstract

In this paper simultaneous localization and map building is performed with a hybrid, metric- topological, approach. A global topological map connects local metric maps, allowing a compact environment model, which does not require global metric consistency and permits both precision and robustness. However, the most important innovation of the approach is the way how loops in the environment are handled by map building using the information of the multi hypotheses topological localization. The method uses data from a 360 ° laser scanner to extract corners and openings for the topological approach and infinite lines for the metric method. This hybrid approach has been tested in a 50 x 25 m 2 portion of the institute building with a fully autonomous robot. The performances of the whole system are proven empirically by comparing maps generated by independent explorations, testing the localization capabilities,

### Citations

613 | Monte carlo localization for mobile robots
- Dellaert, Fox, et al.
- 1998
(Show Context)
Citation Context ...on and multimodality. However this approach remains computationally intractable for current embedded systems. A more efficient alternative has recently been proposed, but the Monte Carlo localization =-=[8]-=- has not yet been extended for map building. Metric and topological approaches are converging, as in [6], [8] and [9], to hybrid solutions. Going in this direction, in [17] the approach consists in ex... |

386 |
Estimating uncertain spatial relationships in robotics
- Smith, Self, et al.
- 1987
(Show Context)
Citation Context ...r the metric approach an Extended Kalman Filter (EKF) is used. This method has already proven its strength for localization in [1]. Map building can therefore be done with the Stochastic Map approach =-=[16]-=-. Topological navigation uses a Partially Observable Markov Decision Process (POMDP) [4] for state estimation. This permits efficient planning in the large, has an advantageous symbolic representation... |

184 | Acting under uncertainty: Discrete bayesian models for mobile robot navigation
- Cassandra, Kaelbling, et al.
- 1996
(Show Context)
Citation Context ...oven its strength for localization in [1]. Map building can therefore be done with the Stochastic Map approach [16]. Topological navigation uses a Partially Observable Markov Decision Process (POMDP) =-=[4]-=- for state estimation. This permits efficient planning in the large, has an advantageous symbolic representation for manmachine interaction and is robust due to its multi hypotheses tracking. 2. Envir... |

181 |
Dynamic map building for an autonomous mobile robot
- Leonard, Durrant-Whyte, et al.
- 1992
(Show Context)
Citation Context ...m the map too. 3.4 Metric Localization and Map Building This section describes briefly the main characteristics of the stochastic map approach [16], which permits using an extended Kalman filter [7], =-=[14]-=- for localization. System State Vector: With this approach both the robot position xr = ( xyθ , , )'and the features xi = ( α, r)'are represented in the system state vector: x r x x = 1 Cx ( ) = x n C... |

152 | World modeling and position estimation for a mobile robot using ultrasonic ranging
- Crowley
- 1989
(Show Context)
Citation Context ...d from the map too. 3.4 Metric Localization and Map Building This section describes briefly the main characteristics of the stochastic map approach [16], which permits using an extended Kalman filter =-=[7]-=-, [14] for localization. System State Vector: With this approach both the robot position xr = ( xyθ , , )'and the features xi = ( α, r)'are represented in the system state vector: x r x x = 1 Cx ( ) =... |

99 | Integrating grid-based and topological maps for mobile robot navigation
- Thrun, Bucken
- 1996
(Show Context)
Citation Context ...the Monte Carlo localization [8] has not yet been extended for map building. Metric and topological approaches are converging, as in [6], [8] and [9], to hybrid solutions. Going in this direction, in =-=[17]-=- the approach consists in extracting a topological map from a grid map by means of a Voronoi based method, while [18] proposes to use the Baum-Welch algorithm as in [11], but to build a topologically ... |

76 |
Markov Localization: A Probabilistic Framework for Mobile Robot Localization and Navigation
- Fox
- 1998
(Show Context)
Citation Context ...le dynamic environments. All these approaches are robust, but are imprecise because the position is represented by a location without precise metric information. To face this, the Markov localization =-=[9]-=- has been proposed: A fine grained grid guarantees both precision and multimodality. However this approach remains computationally intractable for current embedded systems. A more efficient alternativ... |

68 | Integrating topological and metric maps for mobile robot navigation: A statistical approach
- Thrun, Gutmann, et al.
- 1998
(Show Context)
Citation Context ...t local place would be required in order to allow the robot relocating itself by means of the topological approach. Map Building: Instead of using a complex scheme for model learning like in [11] and =-=[18]-=-, where an extension of the BaumWelch algorithm is adopted, here the characteristics of the observation graph (section 2.1) are used. When the robot feels confident about its position, it can decide i... |

52 |
Building a global map of the environment of a mobile robot: The importance of correlations
- Castellanos, Tardos, et al.
- 1997
(Show Context)
Citation Context ...t in a local metric map, however, due to the fact that these maps are supposed to be small, the drift in odometry does not cause any relevant problem to the local consistency, as it has been shown in =-=[5]-=-. The current method works as follows: The robot does not try to recognize if a single perception of the environment has already been seen somewhere else. However, as soon as the robot creates the map... |

47 |
A Qualitative Approach to Robot Exploration and Map Learning
- Kuipers, Byun
- 1987
(Show Context)
Citation Context ...ible due to the computational complexity of this task. Furthermore, the question of how to deal with this kind of multi hypotheses when planning remains open. On the other hand topological approaches =-=[12]-=- can handle multi hypotheses tracking and have a topological global consistency, which is easier to maintain. Their robustness has been proven firstly with the state set progression [15], whichshas th... |

43 | Multisensor on-the-fly localization::: Precision and reliability for applications. Robotics and Autonomous Systems
- Arras, Tomatis, et al.
- 2001
(Show Context)
Citation Context ...s paper it is shown how loops can be closed within the same framework. For the metric approach an Extended Kalman Filter (EKF) is used. This method has already proven its strength for localization in =-=[1]-=-. Map building can therefore be done with the Stochastic Map approach [16]. Topological navigation uses a Partially Observable Markov Decision Process (POMDP) [4] for state estimation. This permits ef... |

30 |
Feature-Based Multi-Hypothesis Localization and Tracking for Mobile Robots Using Geometric Constraints
- Arras, Castellanos, et al.
- 2002
(Show Context)
Citation Context ...icient, as confirmed by a recent work [6], where a solution is proposed by extending the absolute localization to include a localization relative to local reference frames. Some recent works, such as =-=[2]-=-, propose Multi Hypotheses Tracking (MHT) with EKFs. These approaches are then precise and robust. However, when the robot has to relocalize itself, localization during motion is infeasible due to the... |

24 | Simultaneous localization and map building: A global topological model with local metric maps
- Tomatis, Nourbakhsh, et al.
- 2001
(Show Context)
Citation Context ...ions and edges between nodes. Edges to a place correspond to a switch from the topological to the metric paradigm. The effectiveness of this method for localization has already been shown in [19]. In =-=[20]-=- the extension to simultaneous localization and map building is presented. In this paper it is shown how loops can be closed within the same framework. For the metric approach an Extended Kalman Filte... |

22 |
Robot navigation with markov models: A framework for path planning and learning with limited computational resources
- Koenig, Goodwin, et al.
- 1995
(Show Context)
Citation Context ...he current local place would be required in order to allow the robot relocating itself by means of the topological approach. Map Building: Instead of using a complex scheme for model learning like in =-=[11]-=- and [18], where an extension of the BaumWelch algorithm is adopted, here the characteristics of the observation graph (section 2.1) are used. When the robot feels confident about its position, it can... |

18 | A hybrid approach for robust and precise mobile robot navigation with compact environment modeling
- Tomatis, Nourbakhsh, et al.
- 2001
(Show Context)
Citation Context ...cal locations and edges between nodes. Edges to a place correspond to a switch from the topological to the metric paradigm. The effectiveness of this method for localization has already been shown in =-=[19]-=-. In [20] the extension to simultaneous localization and map building is presented. In this paper it is shown how loops can be closed within the same framework. For the metric approach an Extended Kal... |

16 | The need for autonomy and real-time in mobile robotics: a case study of xo/2 and pygmalion
- Brega, Tomatis, et al.
- 2000
(Show Context)
Citation Context ...For the experiments, the mobile robot Donald Duck has been used. It is a fully autonomous mobile vehicle based on a PowerPC 604e 300MHz running XO/2, a deadline driven hard real-time operating system =-=[3]-=-. Donald uses encoders and SICK LMS200 laser scanners as sensors. It navigates locally by means of a motion control algorithm, which plays the role of both position controller and obstacle avoidance: ... |

13 |
Simultaneous localization and map building for mobile robots: a landmark based approach
- Castellanos, Devy, et al.
- 2000
(Show Context)
Citation Context ...over (lost situation). In [5] it has been shown that by taking into account all the correlations the global consistency is better maintained, but this is not sufficient, as confirmed by a recent work =-=[6]-=-, where a solution is proposed by extending the absolute localization to include a localization relative to local reference frames. Some recent works, such as [2], propose Multi Hypotheses Tracking (M... |

12 |
Lola: Probabilistic navigation for topological maps
- Gutierrez-Osuna, RC
- 1996
(Show Context)
Citation Context ...ector of last step and PoaSEk ( , ( ) ) is a normalizing factor. Heading Estimator: Because the position estimator does not take into account the heading of the robot, this is done separately like in =-=[10]-=-. However in this case the orientation is estimated by a weighted mean of each observed line that is either horizontal or vertical with respect to the environment. This can be viewed as an EKF for hea... |

8 |
Automatic Mapping of Dynamic Office Environments. Autonomous Robots
- Kunz, Willeke, et al.
- 1999
(Show Context)
Citation Context ...n been generalized to the POMDP approach [4], [10]. For automatic mapping in [11] the Baum-Welch algorithm has been used for model learning. In contrast to the above mentioned topological approaches, =-=[13]-=- proposes a topological approach, which heavily rely on odometry in order better to handle dynamic environments. All these approaches are robust, but are imprecise because the position is represented ... |

7 |
Dervish: An Office-Navigating Robot. Artificial Intelligence and Mobile
- Nourbakhsh
- 1998
(Show Context)
Citation Context ...al approaches [12] can handle multi hypotheses tracking and have a topological global consistency, which is easier to maintain. Their robustness has been proven firstly with the state set progression =-=[15]-=-, whichshas then been generalized to the POMDP approach [4], [10]. For automatic mapping in [11] the Baum-Welch algorithm has been used for model learning. In contrast to the above mentioned topologic... |