## Probabilistic Algorithms and the Interactive Museum Tour-Guide Robot Minerva (2000)

### Cached

### Download Links

- [www9.in.tum.de]
- [www.informatik.uni-freiburg.de]
- [www.ri.cmu.edu]
- [www.cs.cmu.edu]
- [www-2.cs.cmu.edu]
- [www.informatik.uni-freiburg.de]
- [www.informatik.uni-freiburg.de]
- [www.informatik.uni-freiburg.de]
- [ais.informatik.uni-freiburg.de]
- [hrl.informatik.uni-freiburg.de]
- [ias.in.tum.de]
- [hrl.informatik.uni-freiburg.de]
- [www.ri.cmu.edu]
- [www-2.cs.cmu.edu]
- [robots.stanford.edu]
- [www.cs.cmu.edu]
- [robotics.ee.uwa.edu.au]
- [ijr.sagepub.com]
- DBLP

### Other Repositories/Bibliography

Citations: | 153 - 42 self |

### BibTeX

@MISC{Thrun00probabilisticalgorithms,

author = {S. Thrun and M. Beetz and M. Bennewitz and W. Burgard and A.B. Cremers and F. Dellaert and D. Fox and D. Hähnel and C. Rosenberg and N. Roy and J. Schulte and D. Schulz},

title = {Probabilistic Algorithms and the Interactive Museum Tour-Guide Robot Minerva},

year = {2000}

}

### Years of Citing Articles

### OpenURL

### Abstract

This paper describes Minerva, an interactive tour-guide robot that was successfully deployed in a Smithsonian museum. Minerva's software is pervasively probabilistic, relying on explicit representations of uncertainty in perception and control. This article describes

### Citations

8134 | Maximum likelihood from incomplete data via the em algorithm
- Dempster, Laird, et al.
- 1977
(Show Context)
Citation Context ...es" in the optimization problem; knowing them indeed simplifies the problem greatly. The statistical literature has developed a range of algorithms for such problems, one of which is the EM algor=-=ithm [37, 94]-=-. This algorithm computes a sequence of maps, denoted m [0] , m [1] , . . . , which successively increasing likelihood. The superscript [\Delta] is not to be confused with the time index t or the inde... |

4286 | A tutorial on hidden Markov models and selected applications in speech recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...he one described in this paper has been developed by Shatkay and Kaelbling [126, 127]. Their algorithm generates topological maps using a version of the EM algorithm known as the Baum-Welsh algorithm =-=[107]-=-. To do so, it requires that appropriate landmarks can be found in the sensor data. Minerva's mapping algorithm, in contrast, utilizes all sensor data and generates fine-grained, metric maps. 8.4 Moti... |

3803 | Reinforcement learning: an introduction
- Sutton, Barto
- 1998
(Show Context)
Citation Context ...fter. Thus, the value V (b) of the belief state is the best possible cumulative costs one can expect for being in b. This is expressed as V (b) = Z t+T X =t+1 E[C(ss)js t ] b(s t ) ds: (17) Following =-=[9, 137]-=-, the value function can be computed by recursively adjusting the value of individual belief states b according to V (b) /\Gamma min a Z [V (b 0 ) +C(b 0 )] p(b 0 ja; b; m) db 0 ; (18) which assigns t... |

3711 |
Artificial Intelligence: A Modern Approach
- Russell, Norvig
- 2003
(Show Context)
Citation Context ...o as Markov localization [20, 50, 68, 75, 132], but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models [108], and dynamic belief networks =-=[33, 118]-=-. Kalman filters [72], which are historically the most popular approach for position tracking, represent beliefs by Gaussians. The vanilla Kalman filter also assumes Gaussian noise and linear motion e... |

2621 |
Dynamic Programming
- Bellman
- 1957
(Show Context)
Citation Context ...1, in which case cost is often discounted over time by an exponential factor. The basic idea of POMDPs is to construct a value function in belief space, using a generalized version of value iteration =-=[9, 63]-=-. A value function, denoted by V , measures the expected cumulative cost if one starts in a state s drawn according to the belief distribution 19 b, and acts optimally thereafter. Thus, the value V (b... |

2131 |
A new approach to linear filtering and prediction problems. Transactions of the ASMEJournal of basic Engineering, 82(Series D):35–45
- Kalman
- 1960
(Show Context)
Citation Context ...) b t\Gamma1 (s t\Gamma1 ) ds t\Gamma1 (4) is often referred to as Markov localization [20, 50, 68, 75, 132], but it equally represents a generalization of the basic update equation in Kalman filters =-=[72]-=-, Hidden Markov models [108], and dynamic belief networks [33, 118]. Kalman filters [72], which are historically the most popular approach for position tracking, represent beliefs by Gaussians. The va... |

1995 |
Robot Motion Planning
- Latombe
- 1991
(Show Context)
Citation Context ...bots. Early work on model-based robotics often assumed the availability of a complete and accurate model of the robot and its environment, relying on planners (or theorem provers) to generate actions =-=[21, 82, 125]-=-. Such approaches are often inapplicable to robotics due to the difficulty of generating models that are sufficiently accurate and complete. Recognizing this limitation, some researchers have advocate... |

1307 | Reinforcement learning: A survey
- Kaelbling, Littman, et al.
- 1996
(Show Context)
Citation Context ...AI, the issue of uncertainty in planning has been studied extensively [118]. As discussed above, one of the most popular frameworks is known as partially observable Markov decision processes (POMDPs) =-=[66, 70, 86]-=-. Exact POMDPs are inapplicable to the robot motion planning problem due to their enormous computational complexity. Our approach is a crude approximation to POMDPs which considers uncertainty, but is... |

1136 | CONDENSATION - Conditional density propagation for visual tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ...er integrating a second sensor scan. Now the robot knows its pose with high certainty/accuracy. a version of particle filters [40, 41, 88, 106]. Similar algorithms are known as condensation algorithm =-=[64, 65]-=- in computer vision, and survival of the fittest in artificial intelligence [73]. The basic idea of MCL is to approximate the belief distribution b(s) with a weighted set of samples, also called parti... |

948 |
The EM Algorithm and Extensions
- McLachlan, Krishnan
- 1997
(Show Context)
Citation Context ...es" in the optimization problem; knowing them indeed simplifies the problem greatly. The statistical literature has developed a range of algorithms for such problems, one of which is the EM algor=-=ithm [37, 94]-=-. This algorithm computes a sequence of maps, denoted m [0] , m [1] , . . . , which successively increasing likelihood. The superscript [\Delta] is not to be confused with the time index t or the inde... |

944 |
Numerical methods for unconstrained optimization and nonlinear equations
- Dennis, Schnabel
- 1983
(Show Context)
Citation Context ...mage was taken, the height of ceiling segments, and two additional parameters per image specifying variations in lighting conditions. Our approach employs the well-known Levenberg-Marquardt algorithm =-=[38]-=- for optimization. As a result, the images are brought into local alignment, the ceiling height is estimated, and a global mosaic is constructed. Figure 10b shows the ceiling mosaic of the robot's ope... |

907 |
Behavior-Based Robotics
- Arkin
- 1999
(Show Context)
Citation Context ... competence, but not all modules are required to run the robot. The idea of decentralized, distributed decision making has been at the core of research on behavior-based robotics over the last decade =-=[1, 17, 112]-=-, but there modules are typically much lower in complexity (e.g., finite state machines). 3 Mobile Robot Localization 3.1 The Localization Problem A prime example of probabilistic computing in Minerva... |

834 | An Introduction to Hidden Markov Models
- Rabiner, Juang
- 1986
(Show Context)
Citation Context ...ds t\Gamma1 (4) is often referred to as Markov localization [20, 50, 68, 75, 132], but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models =-=[108]-=-, and dynamic belief networks [33, 118]. Kalman filters [72], which are historically the most popular approach for position tracking, represent beliefs by Gaussians. The vanilla Kalman filter also ass... |

832 | Planning and acting in partially observable stochastic domains
- Kaelbling, Littman
- 1998
(Show Context)
Citation Context ...Possibly the most general framework for probabilistic planning is known as partially observable Markov decision processes, or in short POMDP [96, 133, 135]. Recently, POMDPs have become popular in AI =-=[69, 87]-=-. POMDPs address the problem of choosing actions so as to minimize a scalar cost function, denoted C(s). In robot motion planning, we use C(s) = 0 for goal locations s, and \Gamma1 elsewhere. Since re... |

759 | Intelligence Without Reason
- Brooks
- 1991
(Show Context)
Citation Context ... competence, but not all modules are required to run the robot. The idea of decentralized, distributed decision making has been at the core of research on behavior-based robotics over the last decade =-=[1, 17, 112]-=-, but there modules are typically much lower in complexity (e.g., finite state machines). 3 Mobile Robot Localization 3.1 The Localization Problem A prime example of probabilistic computing in Minerva... |

610 | Robust monte carlo localization for mobile robots
- Dellaert, Thrun, et al.
- 2001
(Show Context)
Citation Context ...cient than grids and more accurate. Therefore, we will describe it here. The Monte Carlo localization algorithm (MCL) is a version of Markov localization that uses samples to approximate the belief b =-=[34, 35, 39, 48, 83]-=-. It is based on the SIR algorithm (SIR stands for sampling/importance resampling) originally proposed in [117], and is 8 (a) (b) robot - robot - Figure 5: Global localization: (a) Pose posterior b t ... |

562 | Contour tracking by stochastic propagation of conditional density
- Isard, Blake
- 1996
(Show Context)
Citation Context ...er integrating a second sensor scan. Now the robot knows its pose with high certainty/accuracy. a version of particle filters [40, 41, 88, 106]. Similar algorithms are known as condensation algorithm =-=[64, 65]-=- in computer vision, and survival of the fittest in artificial intelligence [73]. The basic idea of MCL is to approximate the belief distribution b(s) with a weighted set of samples, also called parti... |

519 |
Dynamic Programming and and Markov Processes
- Howard
- 1960
(Show Context)
Citation Context ...1, in which case cost is often discounted over time by an exponential factor. The basic idea of POMDPs is to construct a value function in belief space, using a generalized version of value iteration =-=[9, 63]-=-. A value function, denoted by V , measures the expected cumulative cost if one starts in a state s drawn according to the belief distribution 19 b, and acts optimally thereafter. Thus, the value V (b... |

519 | Filtering via Simulation: Auxiliary Particle Filters
- Pitt, Shephard
- 1999
(Show Context)
Citation Context ...to 2D). The darker a pose, the more likely it is. (b) shows b t (s t ) after integrating a second sensor scan. Now the robot knows its pose with high certainty/accuracy. a version of particle filters =-=[40, 41, 88, 106]-=-. Similar algorithms are known as condensation algorithm [64, 65] in computer vision, and survival of the fittest in artificial intelligence [73]. The basic idea of MCL is to approximate the belief di... |

498 |
An analysis of time-dependent planning
- Dean, Boddy
- 1988
(Show Context)
Citation Context ...re processing cycles are available, however, the more accurate the result. In Minerva's software, resource flexibility is achieved by two mechanisms: selective data processing and any-time algorithms =-=[32, 151]-=-. Selective data processing is achieved by considering only a subset of the available data, which for example is the case in the localization routine. Other 4 high-level control and learning (mission ... |

481 |
The Complexity of Robot Motion Planning
- Canny
- 1988
(Show Context)
Citation Context ...bots. Early work on model-based robotics often assumed the availability of a complete and accurate model of the robot and its environment, relying on planners (or theorem provers) to generate actions =-=[21, 82, 125]-=-. Such approaches are often inapplicable to robotics due to the difficulty of generating models that are sufficiently accurate and complete. Recognizing this limitation, some researchers have advocate... |

460 | Sequential monte carlo methods for dynamic systems
- Liu, Chen
- 1998
(Show Context)
Citation Context ...to 2D). The darker a pose, the more likely it is. (b) shows b t (s t ) after integrating a second sensor scan. Now the robot knows its pose with high certainty/accuracy. a version of particle filters =-=[40, 41, 88, 106]-=-. Similar algorithms are known as condensation algorithm [64, 65] in computer vision, and survival of the fittest in artificial intelligence [73]. The basic idea of MCL is to approximate the belief di... |

459 |
A model for reasoning about persistence and causation
- Dean, Kanazawa
- 1989
(Show Context)
Citation Context ...o as Markov localization [20, 50, 68, 75, 132], but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models [108], and dynamic belief networks =-=[33, 118]-=-. Kalman filters [72], which are historically the most popular approach for position tracking, represent beliefs by Gaussians. The vanilla Kalman filter also assumes Gaussian noise and linear motion e... |

430 | Globally consistent range scan alignment for environment mapping
- Lu, Milios
- 1997
(Show Context)
Citation Context ... information. Some recent approaches overcome this assumption by "guessing" the correspondence between measurements at different points in time, but they tend to be brittle if those guesses =-=are wrong [56, 89]-=-. In the Smithsonian museum, we know of no set of uniquely identifiable features that would give maps of the target resolution required for accurate localization. 12 4.1 EM Mapping Minerva uses an alt... |

423 | The role of emotion is believable agents
- Bates
- 1994
(Show Context)
Citation Context ... to people through facial gestures [14, 103, 104, 111]. More broadly, Minerva can be seen as a believable robotic agent, and hence is related to literature on believability in software agent research =-=[4]-=- and in robotics [31]. 8.6 Web Interfaces Web interfaces have received serious attention in robotics throughout the last years, since they allow people to tele-operate a robot at a distant site. Three... |

406 | A robot exploration and mapping strategy based on a semantic hierarchy of spatial representations
- Kuipers, Byun
- 1991
(Show Context)
Citation Context ...ed topological. Topological methods represent maps as graphs. Usually, nodes in these graphs correspond to distinct places in the environment, and arcs to actions for moving from one place to another =-=[24, 25, 26, 27, 79, 80, 91, 98, 146, 152]-=-. Probably the most related mapping algorithm to the one described in this paper has been developed by Shatkay and Kaelbling [126, 127]. Their algorithm generates topological maps using a version of t... |

405 | A probabilistic approach to concurrent mapping and localization for mobile robots, Machine Learning 31
- Thrun, Burgard, et al.
- 1998
(Show Context)
Citation Context ...esolution required for accurate localization. 12 4.1 EM Mapping Minerva uses an alternative approach for mapping, which is based on the same mathematical framework as the Kalman filter approach above =-=[144]-=-. In particular, our approach seeks to estimate the mode of the posterior, denotedsm = argmax m p(mjd), instead of the full posterior p(mjd). This might appear quite modest a goal compared to the full... |

384 |
Estimating uncertain spatial relationships
- Smith, Self, et al.
- 1988
(Show Context)
Citation Context ... motion equations; however, extensions exist that relax some of these assumptions [67, 92]. Kalman filters have been applied with great success to a range of tracking and mapping problems in robotics =-=[85, 134]-=-; though they tend not to work well for global localization or the kidnapped robot problem (see [55] for an experimental comparison). Markov localization using discrete, topological representations fo... |

344 |
fusion in certainty grids for mobile robots
- Moravec
- 1988
(Show Context)
Citation Context ...ls of the mathematical derivation and the implementation of this algorithm can be found in [144]. 4.2 Occupancy Grid Maps In a final mapping step, Minerva transforms its maps into occupancy grid maps =-=[43, 97]-=-. Occupancy grids are widely used in mobile robotics; see e.g., [11, 42, 58, 142, 149]. Most state-of-the-art algorithms for generating occupancy grid maps are probabilistic. Occupancy grid mapping ad... |

338 |
The Optimal Control of Partially Observable Markov Decision Processes
- Sondik
- 1971
(Show Context)
Citation Context ...o known obstacles, to minimize the danger of getting lost. Possibly the most general framework for probabilistic planning is known as partially observable Markov decision processes, or in short POMDP =-=[96, 133, 135]-=-. Recently, POMDPs have become popular in AI [69, 87]. POMDPs address the problem of choosing actions so as to minimize a scalar cost function, denoted C(s). In robot motion planning, we use C(s) = 0 ... |

330 | The Vector Field Histogram–Fast Obstacle Avoidance for Mobile Robots
- Borenstein, Koren
- 1991
(Show Context)
Citation Context ... robot to avoid collisions with obstacles---people and exhibits alike. Many collision avoidance methods for mobile robots consider only the kinematics of a robot, without taking dynamics into account =-=[13]-=-. This is legitimate at speeds where robots can stop almost instantaneously. However, at velocities of up to 163 cm/sec, inertia and torque limits impose constraints on robot motion, which may not be ... |

317 | Elephants don’t play chess
- Brooks
- 1990
(Show Context)
Citation Context ...inimum of internal state. Some advocates of this approach went as far as refusing the need for internal models and internal state altogether [15, 28]. Observing that "the world is its own best mo=-=del" [16]-=-, behavior-based approaches usually rely on immediate sensor feedback for determining a robot's action. Ob28 vious limits in perception (e.g., robots cannot see through walls) pose clear boundaries on... |

296 |
The Optimal Control of Partially Observable Markov Decision Processes Over A Finite Horizon
- Smallwood, Sondik
- 1971
(Show Context)
Citation Context ...o known obstacles, to minimize the danger of getting lost. Possibly the most general framework for probabilistic planning is known as partially observable Markov decision processes, or in short POMDP =-=[96, 133, 135]-=-. Recently, POMDPs have become popular in AI [69, 87]. POMDPs address the problem of choosing actions so as to minimize a scalar cost function, denoted C(s). In robot motion planning, we use C(s) = 0 ... |

290 |
Sonar-based real-world mapping and navigation
- Elfes
- 1987
(Show Context)
Citation Context ...hm can be found in [144]. 4.2 Occupancy Grid Maps In a final mapping step, Minerva transforms its maps into occupancy grid maps [43, 97]. Occupancy grids are widely used in mobile robotics; see e.g., =-=[11, 42, 58, 142, 149]-=-. Most state-of-the-art algorithms for generating occupancy grid maps are probabilistic. Occupancy grid mapping addresses a much simpler problem than the one above, namely the problem of estimating a ... |

284 | Markov localization for mobile robots in dynamic environments
- Fox, Burgard, et al.
- 1999
(Show Context)
Citation Context ...lization, the result of this transformation b t (s t ) = j t p(o t js t ; m) Z p(s t ja t\Gamma1 ; s t\Gamma1 ; m) b t\Gamma1 (s t\Gamma1 ) ds t\Gamma1 (4) is often referred to as Markov localization =-=[20, 50, 68, 75, 132]-=-, but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models [108], and dynamic belief networks [33, 118]. Kalman filters [72], which are hist... |

277 | Monte Carlo Localization: Efficient position estimation for mobile robots
- Fox, Burgard, et al.
- 1999
(Show Context)
Citation Context ...cient than grids and more accurate. Therefore, we will describe it here. The Monte Carlo localization algorithm (MCL) is a version of Markov localization that uses samples to approximate the belief b =-=[34, 35, 39, 48, 83]-=-. It is based on the SIR algorithm (SIR stands for sampling/importance resampling) originally proposed in [117], and is 8 (a) (b) robot - robot - Figure 5: Global localization: (a) Pose posterior b t ... |

271 | S.: Experiences with an interactive museum tour-guide robot
- Burgard, Cremers, et al.
- 1999
(Show Context)
Citation Context ...red tour-length regardless of how crowded the museum is. Minerva is a second generation tour-guide robot, following the successful example of the robot Rhino developed by the same team of researchers =-=[19]-=-. Rhino was deployed in the Deutsches Museum in Bonn in 1997, with many of the same probabilistic navigation algorithms. Minerva, however, went beyond Rhino in various ways, from using new probabilist... |

258 |
Learning metric-topological maps for indoor mobile robot navigation
- Thrun
- 1999
(Show Context)
Citation Context ...hm can be found in [144]. 4.2 Occupancy Grid Maps In a final mapping step, Minerva transforms its maps into occupancy grid maps [43, 97]. Occupancy grids are widely used in mobile robotics; see e.g., =-=[11, 42, 58, 142, 149]-=-. Most state-of-the-art algorithms for generating occupancy grid maps are probabilistic. Occupancy grid mapping addresses a much simpler problem than the one above, namely the problem of estimating a ... |

254 | Probabilistic robot navigation in partially observable environments
- Simmons, Koenig
- 1995
(Show Context)
Citation Context ...lization, the result of this transformation b t (s t ) = j t p(o t js t ; m) Z p(s t ja t\Gamma1 ; s t\Gamma1 ; m) b t\Gamma1 (s t\Gamma1 ) ds t\Gamma1 (4) is often referred to as Markov localization =-=[20, 50, 68, 75, 132]-=-, but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models [108], and dynamic belief networks [33, 118]. Kalman filters [72], which are hist... |

241 | Directed Sonar Sensing for Mobile Robot Navigation
- Leonard, Durrant-Whyte
- 1992
(Show Context)
Citation Context ...therefore an essential component of Minerva's and Rhino's software architecture. The reader should notice that localization is a key component in many other successful mobile robot systems (see e.g., =-=[12, 84, 76]). Occasio-=-nally, the localization problem has been referred to as "the most fundamental problem to providing a mobile robot with autonomous capabilities" [29]. The literature distinguishes three types... |

236 |
Complexity of the mover’s problem and generalizations
- Reif
- 1979
(Show Context)
Citation Context ...nsor data and generates fine-grained, metric maps. 8.4 Motion Planning Robot motion planning has been subject to intense research, as documented by a large body of literature on this topic (see e.g., =-=[21, 82, 109, 125]-=-). The majority of work addresses more complicated problems than the one addressed in this article, such as motion planning in higher-dimensional and continuous spaces. Latombe [82] pioneered the use ... |

233 | L.P.: Learning policies for partially observable environments: Scaling up
- Littman, Cassandra, et al.
- 1995
(Show Context)
Citation Context ...AI, the issue of uncertainty in planning has been studied extensively [118]. As discussed above, one of the most popular frameworks is known as partially observable Markov decision processes (POMDPs) =-=[66, 70, 86]-=-. Exact POMDPs are inapplicable to the robot motion planning problem due to their enormous computational complexity. Our approach is a crude approximation to POMDPs which considers uncertainty, but is... |

227 |
A robot that walks: Emergent behaviors from a carefully evolved network
- Brooks
- 1989
(Show Context)
Citation Context ...l-free reactive approaches. Instead of relying on planning, these approaches require programmers to program controllers directly. A popular example of this approach is the "subsumption architectu=-=re " [15]-=-, where controllers are composed of small finite state automata that map sensor readings directly into control while retaining a minimum of internal state. Some advocates of this approach went as far ... |

208 | On sequential simulation-based methods for bayesian filtering
- Doucet
- 1998
(Show Context)
Citation Context ...to 2D). The darker a pose, the more likely it is. (b) shows b t (s t ) after integrating a second sensor scan. Now the robot knows its pose with high certainty/accuracy. a version of particle filters =-=[40, 41, 88, 106]-=-. Similar algorithms are known as condensation algorithm [64, 65] in computer vision, and survival of the fittest in artificial intelligence [73]. The basic idea of MCL is to approximate the belief di... |

193 |
A survey of partially observable Markov decision processes: theory models and algorithms,” Mgmt
- Monahan
- 1982
(Show Context)
Citation Context ...o known obstacles, to minimize the danger of getting lost. Possibly the most general framework for probabilistic planning is known as partially observable Markov decision processes, or in short POMDP =-=[96, 133, 135]-=-. Recently, POMDPs have become popular in AI [69, 87]. POMDPs address the problem of choosing actions so as to minimize a scalar cost function, denoted C(s). In robot motion planning, we use C(s) = 0 ... |

183 | Acting under uncertainty: discrete bayesian models for mobile-robotnavigation
- Cassandra, Kaelbling, et al.
- 1996
(Show Context)
Citation Context ...lization, the result of this transformation b t (s t ) = j t p(o t js t ; m) Z p(s t ja t\Gamma1 ; s t\Gamma1 ; m) b t\Gamma1 (s t\Gamma1 ) ds t\Gamma1 (4) is often referred to as Markov localization =-=[20, 50, 68, 75, 132]-=-, but it equally represents a generalization of the basic update equation in Kalman filters [72], Hidden Markov models [108], and dynamic belief networks [33, 118]. Kalman filters [72], which are hist... |

181 |
Dynamic map building for an autonomous mobile robot
- Leonard, Durrant-Whyte, et al.
- 1992
(Show Context)
Citation Context ... motion equations; however, extensions exist that relax some of these assumptions [67, 92]. Kalman filters have been applied with great success to a range of tracking and mapping problems in robotics =-=[85, 134]-=-; though they tend not to work well for global localization or the kidnapped robot problem (see [55] for an experimental comparison). Markov localization using discrete, topological representations fo... |

180 |
Blanche—an experiment in guidance and navigation of an autonomous robot vehicle
- Cox
- 1991
(Show Context)
Citation Context ...l mobile robot systems (see e.g., [12, 84, 76]). Occasionally, the localization problem has been referred to as "the most fundamental problem to providing a mobile robot with autonomous capabilit=-=ies" [29]-=-. The literature distinguishes three types of localization problems, in increasing order of difficulty: 1. Position tracking. Here the initial robot pose is known, and the goal of localization is to c... |

175 | Estimating the absolute position of a mobile robot using position probability grids
- Burgard, Fox, et al.
- 1996
(Show Context)
Citation Context ...ing small odometric errors [56, 57, 89, 120, 134]. Recently, several researchers proposed the Markov localization used by Minerva, which enables robots to localize themselves under global uncertainty =-=[20, 35, 48, 68, 101, 132]-=-. Minerva's and Rhino's localization algorithm goes beyond previous approaches in that it can cope with invisible hazards and highly dynamic environments. 8.3 Mapping Several major approaches to concu... |

174 | Damn: a distributed architecture for mobile navigation
- Rosenblatt
- 1997
(Show Context)
Citation Context ... competence, but not all modules are required to run the robot. The idea of decentralized, distributed decision making has been at the core of research on behavior-based robotics over the last decade =-=[1, 17, 112]-=-, but there modules are typically much lower in complexity (e.g., finite state machines). 3 Mobile Robot Localization 3.1 The Localization Problem A prime example of probabilistic computing in Minerva... |