## Mapping partially observable features from multiple uncertain vantage points (2002)

### Cached

### Download Links

- [groups.csail.mit.edu]
- [cs.krisbeevers.com]
- [oe.mit.edu]
- [cml.mit.edu]
- [albacore.mit.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | The International Journal of Robotics Research |

Citations: | 40 - 9 self |

### BibTeX

@ARTICLE{Leonard02mappingpartially,

author = {John J. Leonard and Richard J. Rikoski and Paul M. Newman and Michael Bosse},

title = {Mapping partially observable features from multiple uncertain vantage points},

journal = {The International Journal of Robotics Research},

year = {2002}

}

### Years of Citing Articles

### OpenURL

### Abstract

In this paper we present a technique for mapping partially observable features from multiple uncertain vantage points. The problem of concurrent mapping and localization (CML) is stated as follows. Starting from an initial known position, a mobile robot travels through a sequence of positions, obtaining a set of sensor measurements at each position. The goal is to process the sensor data to produce an estimate of the trajectory of the robot while concurrently building a map of the environment. In this paper, we describe a generalized framework for CML that incorporates temporal as well as spatial correlations. The representation is expanded to incorporate past vehicle positions in the state vector. Estimates of the correlations between current and previous vehicle states are explicitly maintained. This enables the consistent initialization of map features using data from multiple time steps. Updates to the map and the vehicle trajectory can also be performed in batches of data acquired from multiple vantage points. The method is illustrated with sonar data from a testing tank and via experiments with a B21 land mobile robot, demonstrating the ability to perform CML with sparse and ambiguous data. KEY WORDS—mapping, navigation, mobile robots 1.

### Citations

3348 |
Multiple View Geometry in Computer Vision
- Hartley, Zisserman
- 2000
(Show Context)
Citation Context ... computer vision community, the analogous problem to CML is Structure from Motion (SFM) (Faugeras 1993; Taylor and Kriegman 1995; Chiuso, Favaro, Jin, and Soatto 2000; Yagi, Shouya, and Yachida 2000; =-=Hartley and Zisserman 2001-=-). An early approach to SFM that used the EKF was developed by Ayache and Faugeras (1989). This work was one of the first to apply geometric constraints in the state estimation process, such as the fu... |

2606 |
Random sample consensus: A paradigm for model fitting
- Fischler, Bolles
- 1981
(Show Context)
Citation Context ...ntly estimate scene structure and camera motion (Triggs, McLauchlan, Hartley, and Fitzgibbon 2000). These techniques typically solve the association problem by using random sample consensus (RANSAC) (=-=Fischler and Bolles 1981-=-) or Least Medians (Faugeras, Luong, and Papadopoulo 2001) to find groups of measurements that originate from the same point in the scene across multiple images. While most work has considered the bat... |

1301 |
Three-Dimensional Computer Vision: a Geometric Viewpoint
- Faugeras
- 1993
(Show Context)
Citation Context ...ceptual grouping routines to be consistently applied for mapping from multiple uncertain vantage points. In the computer vision community, the analogous problem to CML is Structure from Motion (SFM) (=-=Faugeras 1993-=-; Taylor and Kriegman 1995; Chiuso, Favaro, Jin, and Soatto 2000; Yagi, Shouya, and Yachida 2000; Hartley and Zisserman 2001). An early approach to SFM that used the EKF was developed by Ayache and Fa... |

961 | Probabilistic roadmaps for path planning in high dimensional configuration spaces - Kavraki, ˇSvestka, et al. - 1994 |

947 | Sequential Monte Carlo methods in practice - Doucet, Freitas, et al. - 2001 |

852 |
Computer Vision
- Ballard, Brown
- 1982
(Show Context)
Citation Context ...res (point, plane, etc.) or could be spurious. For each type of feature, there is a limited range of locations for a potential feature that are possible. Given these constraints, the Hough transform (=-=Ballard and Brown 1982-=-) can be used as a voting scheme identify point and planar features. More detail on this technique is contained in Tardós et al. (2001). The method is similar in spirit to the TBF method Wijk and Chri... |

749 |
Tracking and Data Association
- Bar-Shalom, Fortmann
- 1988
(Show Context)
Citation Context ...ould be chosen instead. Discrete state estimation — making decisions about the origins of measurements — is usually performed in CML with maximum likelihood methods such as “nearest-neighbor gating” (=-=Bar-Shalom and Fortmann 1988-=-). Such methods encounter difficulties when the distance between features in the environment is smaller than the uncertainty in the robot position. Unfortunately, this situation can arise frequently i... |

742 |
Optimal Filtering
- Anderson, Moore
- 1979
(Show Context)
Citation Context ...cellent performance by keeping the number of trajectory states restricted to a fixed size of 40. With a fixed window size, the process of adding past states is similar to a fixed-lag Kalman smoother (=-=Anderson and Moore 1979-=-). 3.1 Perceptual Grouping The next step in the framework is to apply a perceptual grouping algorithm to examine collectively the entire set of data that came from the current and past vehicle positio... |

441 | An algorithm for tracking multiple targets
- Reid
- 1979
(Show Context)
Citation Context ... estimation problem in which tracking and data association are intertwined (BarShalom and Fortmann 1988). General theoretical models for hybrid state estimation, such as multiple hypothesis tracking (=-=Reid 1979-=-; Mori, Chong, Tse, and Wishner 1986), present a staggeringly large computational burden when applied to CML. The navigation error of the platform prevents one from splitting the solution into separat... |

373 | A solution to the simultaneous localization and map building (SLAM) problem - Dissanayake, Newman, et al. |

356 |
Object Recognition By Computer: The Role of Geometric Constraints
- Grimson
- 1990
(Show Context)
Citation Context ...ptions made by Reid’s algorithm, the method is exponentially complex due to the combinatorics of discrete decision making. The problem bears some resemblance to object recognition in computer vision (=-=Grimson 1990-=-). It is unclear how to incorporate variable-dimensionality (initialization of new features based on state estimates for the robot and other features in the map) into the Mori et al. (1986) algorithm.... |

328 | The Geometry of Multiple Images - Faugeras, Luong, et al. - 2001 |

293 | Incremental mapping of large cyclic environments - Gutmann, Konolige - 2000 |

257 | Directed Sonar Sensing for Mobile Robot Navigation
- Leonard, Durrant-Whyte
- 1992
(Show Context)
Citation Context ... based on looking for several consecutive unexplained measurements that are close to one another, and far from any previously matched features. This policy is referred to as delayed track initiation (=-=Leonard and Durrant-Whyte 1992-=-; Feder, Leonard, and Smith 1999; Dissanayake et al. 2001). In general, there is a trade-off between being more likely to assign a measurement to an old feature, versus using it to initialize a new fe... |

256 | The Spatial semantic hierarchy
- Kuipers
- 2000
(Show Context)
Citation Context ...esent a challenging data interpretation problem. The CML problem can be addressed using a variety of different representations, such as evidence grids (Schultz and Adams 1998) and topological models (=-=Kuipers 2000-=-). We advocate the use of a feature-based, probabilistic representation. Smith, Self, and Cheeseman (1987) were the first researchers to cast the problem of feature-based CML using a variable-dimensio... |

195 | A probabilistic online mapping algorithm for teams of mobile robots
- Thrun
- 2001
(Show Context)
Citation Context ...0). This is the method used in this paper, but it is not the only state estimation algorithm which could be used. For instance, sequential Monte Carlo algorithms (Doucet, de Freitas, and Gordan 2001; =-=Thrun 2001-=-) could be chosen instead. Discrete state estimation — making decisions about the origins of measurements — is usually performed in CML with maximum likelihood methods such as “nearest-neighbor gating... |

190 | Data association in stochastic mapping using the joint compatibility test, Robotics and Automation - Neira, Tardos - 2001 |

189 | Topological simultaneous localization and mapping (slam): Toward exact localization without explicit localization - Choset, Nagatani |

188 | Dynamic map building for an autonomous mobile robot - Leonard, Durrant-Whyte, et al. - 1992 |

186 | Optimization of the simultaneous localization and map-building algorithm for real-time implementation
- Guviant, Nebot
- 2001
(Show Context)
Citation Context ...ete (decision-making) and continuous (filtering) parts of the problem. The extended Kalman filter (EKF) is typically used for continuous state estimation (Smith, Self, and Cheeseman 1987; Feder 1999; =-=Guivant and Nebot 2001-=-; Castellanos and Tardos 2000). This is the method used in this paper, but it is not the only state estimation algorithm which could be used. For instance, sequential Monte Carlo algorithms (Doucet, d... |

153 |
Faugeras, “Maintaining representations of the environment of a mobile robot
- Ayache, D
- 1989
(Show Context)
Citation Context ...the locations of previously observed features in the environment. The robot also has to maintain its map, which can incorporate the fusing of two features that are hypothesized to be the same object (=-=Ayache and Faugeras 1989-=-; Chong and Kleeman 1997a) and the deletion of features that are hypothesized to no longer be present (Leonard, Cox, and Durrant-Whyte 1992). In this manner, the number of elements in the stochastic m... |

150 | An Atlas framework for scalable mapping - Bosse, Newman, et al. - 2003 |

146 | Robust Mapping and Localization in Indoor Environments using Sonar Data - Tardos, Neira, et al. - 2002 |

142 | A New Method for Nonlinear Transformation of Means and Covariances in Filters and Estimators - Julier, Uhlmann, et al. - 2000 |

137 | A stochastic map for uncertain spatial relationships - Smith, Self, et al. - 1988 |

126 |
Mobile Robot Localization and Map Building: A Multisensor Fusion Approach
- Castellanos, Tardós
- 1999
(Show Context)
Citation Context ...d continuous (filtering) parts of the problem. The extended Kalman filter (EKF) is typically used for continuous state estimation (Smith, Self, and Cheeseman 1987; Feder 1999; Guivant and Nebot 2001; =-=Castellanos and Tardós 2000-=-). This is the method used in this paper, but it is not the only state estimation algorithm which could be used. For instance, sequential Monte Carlo algorithms (Doucet, de Freitas, and Gordan 2001; T... |

107 | Structure and motion from line segments in multiple images
- Taylor, Kriegman
- 1995
(Show Context)
Citation Context ...g routines to be consistently applied for mapping from multiple uncertain vantage points. In the computer vision community, the analogous problem to CML is Structure from Motion (SFM) (Faugeras 1993; =-=Taylor and Kriegman 1995-=-; Chiuso, Favaro, Jin, and Soatto 2000; Yagi, Shouya, and Yachida 2000; Hartley and Zisserman 2001). An early approach to SFM that used the EKF was developed by Ayache and Faugeras (1989). This work w... |

101 |
Stochastic multisensory data fusion for mobile robot location and environment modeling
- Moutarlier, Chatila
- 1989
(Show Context)
Citation Context ...ated by Gaussian distributions. State updates are performed with the EKF. With these approximations, and assuming that data association is known, the ¢ ¦ computational complexity � ��� is reduced to (=-=Moutarlier and Chatila 1989-=-b). The method recursively computes � a state estimate time � step , � ��� where ¢ � � � ¦�� � ��� and ¢ � ¢ ¦���� ���¥� � � � ¢ � � � � ¦�������� � ¢ ¦�� � � ��� ¦������ ¢ ����� � � � ¢ ¦������ ¦�� �... |

94 | Adaptive Mobile Robot Navigation and Mapping - Feder, Leonard, et al. - 1999 |

86 |
Modeling a dynamic environment using a bayesian multiple hypothesis approach
- Cox, Leonard
- 1994
(Show Context)
Citation Context ...or of the platform prevents one from splitting the solution into separate subgroups (known as “clusters” in the tracking literature (Kurien 1990)) corresponding to different parts of the environment (=-=Cox and Leonard 1994-=-). Most implementations of stochastic mapping decouple the discrete (decision-making) and continuous (filtering) parts of the problem. The extended Kalman filter (EKF) is typically used for continuous... |

78 | Experimental comparison of techniques for localisation and mapping using a bearing-only sensor - Deans, Hebert - 2000 |

76 | A unifying framework for structure and motion recovery from image sequences
- McLauchlan, Murray
- 1995
(Show Context)
Citation Context ...he variable state dimension filter (VSDF), which is a hybrid batch/recursive technique that combines the characteristics of the EKF and bundle adjustment (McLauchlan 2000; McLauchlan and Murray 1996; =-=McLauchlan and Murray 1995-=-). The VSDF maintains a dynamic time-window of observations and camera motion parameters. We employ a similar idea in this paper. Deans and Hebert (2000) have developed a related method for CML with b... |

63 | Mobile Robot Navigation Using Active Vision
- Davison
- 1999
(Show Context)
Citation Context ... cost (Moutarlier and Chatila 1989b), where � is the number of features in the environment. This has motivated techniques to address the map scaling problem through spatial and temporal partitioning (=-=Davison 1998-=-; Leonard and Feder 2001; Guivant and Nebot 2001). The current paper has not addressed the map scaling problem, however it provides a framework for increasing the reliability of local map building. Th... |

60 |
The Sonar of Dolphins
- Au
- 1993
(Show Context)
Citation Context ...eam sonar measurements. 4.1 Tank Experiment An experiment was conducted using a robotic gantry to emulate the motion and sensing of an underwater vehicle. A 500 KHz binaural sonar was used (Kuc 1996; =-=Au 1993-=-). To show mapping of partially observable features, bearing information was discarded. Two objects were placed in the tank, a metal triangle and a point like object (a fishing bobber). The gantry was... |

57 | Cooperative concurrent mapping and localization - Fenwick, Newman, et al. |

54 |
An experimental system for incremental environment modeling by an autonomous mobile robot
- Moutarlier, Chatila
- 1989
(Show Context)
Citation Context ...ated by Gaussian distributions. State updates are performed with the EKF. With these approximations, and assuming that data association is known, the ¢ ¦ computational complexity � ��� is reduced to (=-=Moutarlier and Chatila 1989-=-b). The method recursively computes � a state estimate time � step , � ��� where ¢ � � � ¦�� � ��� and ¢ � ¢ ¦���� ���¥� � � � ¢ � � � � ¦�������� � ¢ ¦�� � � ��� ¦������ ¢ ����� � � � ¢ ¦������ ¦�� �... |

53 | Decoupled stochastic mapping
- Leonard, Feder
- 2001
(Show Context)
Citation Context ...ier and Chatila 1989b), where � is the number of features in the environment. This has motivated techniques to address the map scaling problem through spatial and temporal partitioning (Davison 1998; =-=Leonard and Feder 2001-=-; Guivant and Nebot 2001). The current paper has not addressed the map scaling problem, however it provides a framework for increasing the reliability of local map building. This is anticipated to gre... |

53 | Explore and return: Experimental validation of real-time concurrent mapping and localization - Newman, Leonard, et al. - 2002 |

49 | Mobile robot sonar for target localization and classification
- Kleeman, Kuc
- 1995
(Show Context)
Citation Context ...alize geometric primitives from a single vantage point. The state-of-the-art in 41sthis area is exhibited by the work of Kleeman et al. (Kleeman 2001; Heale and Kleeman 2001; Chong and Kleeman 1997a; =-=Kleeman and Kuc 1993-=-). For example, Chong and Kleeman (1997b) have used custom advanced sonar arrays to very good effect in testing large-scale CML algorithms. Since this was a scanning sonar, the robot has to stop and s... |

47 | Continuous localization using evidence grids
- Schultz, Adams
- 1998
(Show Context)
Citation Context ..., and data association ambiguity – combine to present a challenging data interpretation problem. The CML problem can be addressed using a variety of different representations, such as evidence grids (=-=Schultz and Adams 1998-=-) and topological models (Kuipers 2000). We advocate the use of a feature-based, probabilistic representation. Smith, Self, and Cheeseman (1987) were the first researchers to cast the problem of featu... |

45 |
Issues in the design of practical multitarget tracking algorithms,” Multitarget-Multisensor Tracking: Advanced Applications
- Kurien
- 1990
(Show Context)
Citation Context ... large computational burden when applied to CML. The navigation error of the platform prevents one from splitting the solution into separate subgroups (known as “clusters” in the tracking literature (=-=Kurien 1990-=-)) corresponding to different parts of the environment (Cox and Leonard 1994). Most implementations of stochastic mapping decouple the discrete (decision-making) and continuous (filtering) parts of th... |

42 |
An Experimental and Theoretical Investigation into Simultaneous Localization and Map Building
- Dissanayake, Newman, et al.
(Show Context)
Citation Context ...e-based CML has shown the importance of maintaining spatial correlations between the state estimates for different features, in order to maintain consistent error bounds (Castellanos and Tardós 2000; =-=Dissanayake et al. 1999-=-). The representation of spatial correlations results in an O(n 2 ) growth in computational cost (Moutarlier and Chatila 1989), where n is the number of features in the environment. This has motivated... |

39 | A batch/recursive algorithm for 3D scene reconstruction
- McLauchlan
- 2000
(Show Context)
Citation Context ...good quality data. McLauchlan has developed the variable state dimension filter (VSDF), which is a hybrid batch/recursive technique that combines the characteristics of the EKF and bundle adjustment (=-=McLauchlan 2000-=-; McLauchlan and Murray 1996; McLauchlan and Murray 1995). The VSDF maintains a dynamic time-window of observations and camera motion parameters. We employ a similar idea in this paper. Deans and Hebe... |

39 | Active Camera Calibration for a HeadEye Platform Using the Variable State-Dimension Filter
- McLauchlan, Murray
- 1996
(Show Context)
Citation Context .... McLauchlan has developed the variable state dimension filter (VSDF), which is a hybrid batch/recursive technique that combines the characteristics of the EKF and bundle adjustment (McLauchlan 2000; =-=McLauchlan and Murray 1996-=-; McLauchlan and Murray 1995). The VSDF maintains a dynamic time-window of observations and camera motion parameters. We employ a similar idea in this paper. Deans and Hebert (2000) have developed a r... |

36 |
I.Christensen, “Triangulation based fusion of sonar data with application in robot pose tracking
- Wijk, Henrik
(Show Context)
Citation Context ... computing circle intersection points and applying angle constraints. The method runs in real-time and has been successfully used for occupancy grid mapping, model-based localization, and relocation (=-=Wijk and Christensen 2000-=-). CML has also been implemented using the TBF for points features (Zunino and Christensen 2001). In this paper, we use manual data association in Section 4 to illustrate various new types of feature ... |

35 |
Digital synthesis of Nonlinear Filters
- Bucy, Senne
- 1971
(Show Context)
Citation Context ... all features, we are left with a “conventional”, fixed¢ ¦ � dimension state estimation problem. The general recursive solution applicable for fully non-linear and non-Gaussian systems is well-known (=-=Bucy and Senne 1971-=-; Sorenson 1988) and given by the following two equations: and ¢ � � ¢ � ¦�� ����� � ¢ � � � ¢ � ¦���� � ¢ � ����� � ¦�� � � � ¦���� ¢ � ¦�� � � ¢ � � ¢ ¢ ¦�� ¦�¦ ����� � ¢ � � ����� ¢ � ¦�� � ¢ ¦�¦ �... |

33 | Tracking and classifying multiple targets without a priori identification - Mori, Chong, et al. - 1986 |

25 |
Probabilistic models and stochastic algorithms for visual tracking
- MacCormick
- 2000
(Show Context)
Citation Context ...for many chal7slenging nonlinear, non-Gaussian state estimation problems; difficulties are encountered, however, in the application of sequential Monte Carlo methods in high-dimensional state spaces (=-=MacCormick 2000-=-). Equations 2 and 3 assume that the correspondence problem is known. When data association uncertainty (the correspondence problem) is added to the formulation, one is left with a hybrid (mixed conti... |

24 | S.: Vanishing points and 3d lines from omnidirectional video - Bosse, Rikoski, et al. |

23 | Approches to mobile robot localization in indoor environment - Jensfelt - 2001 |