#### DMCA

## MULTI-ROBOT ACTIVE INFORMATION GATHERING USING RANDOM FINITE SETS (2015)

### Citations

12379 |
Elements of Information Theory
- Thomas, Cover
- 1991
(Show Context)
Citation Context ...nce matrix, Σ, max k (λk). (2.49) General Distributions Uncertainty measures for general distributions come from information theory. The most common measures are due to Shannon [91]. Cover and Thomas =-=[20]-=- provide an excellent summary of information theory and provide many useful identities and inequalities. The entropy of a continuous random variable X is H[X] = − ∫ p(x) log p(x) dx, (2.50) where the ... |

10881 |
A Mathematical Theory of Communication
- Shannon
- 1948
(Show Context)
Citation Context ...genvalue of the covariance matrix, Σ, max k (λk). (2.49) General Distributions Uncertainty measures for general distributions come from information theory. The most common measures are due to Shannon =-=[91]-=-. Cover and Thomas [20] provide an excellent summary of information theory and provide many useful identities and inequalities. The entropy of a continuous random variable X is H[X] = − ∫ p(x) log p(x... |

3845 |
A new approach to linear filtering and prediction problems
- Kalman
- 1960
(Show Context)
Citation Context ...ion over all possible target states using this method. In order for the Bayes filter to be computationally tractable, we must make some simplifying assumptions. 6 Kalman Filter The Kalman filter (KF) =-=[53]-=- is an implementation of the Bayes filter for linear Gaussian systems. This means that the target state is represented by a Gaussian distribution, the measurement and motion models are linear, and all... |

1346 | Probabilistic Robotics - Thrun, Burgard, et al. - 2005 |

1010 | Alignment by maximization of mutual information
- Viola, Wells
- 1995
(Show Context)
Citation Context ...rted in the literature despite the proliferation of gradient based methods for maximizing mutual information in various applications ranging from channel coding [78, 79], to medical imaging alignment =-=[100]-=-, to the control of robotic sensor networks [41]. In [79] the authors derive an expression that can be shown to be equivalent to a special case of Theorem 3 in which p(X) is not a function of q. Remar... |

735 | Constrained model predictive control: stability and optimality
- Mayne, Rawlings, et al.
- 2000
(Show Context)
Citation Context ...ion for multi-target tracking builds on the literature on receding horizon control and model predictive control. Mayne and Michalska [72] provide a survey of receding horizon control and Mayne et al. =-=[73]-=- provide a survey of model predictive control, including applications in a variety of domains. Jadbabaie [48] utilizes model 31 predictive control to follow trajectories with UAVs. The work of Ryan [8... |

718 | Gaussian processes for Machine Learning
- Rasmussen, Williams
- 2006
(Show Context)
Citation Context ...tion models is used [62]. We take a data-driven approach to modeling the targets' motion, utilizing real-world datasets that are available [55]. In particular, we use Gaussian Process (GP) regression =-=[83]-=- to learn the function that maps the position coordinates of the targets to velocity vectors, as shown by Joseph et al. [49]. GP regression is a Bayesian approximation technique to learn some function... |

502 | A solution to the simultaneous localization and map building (SLAM) problem
- Dissanayake, Newman, et al.
- 2001
(Show Context)
Citation Context ...lti-target tracking with unknown associations is to use the maximum likelihood association. This has been used successfully in a variety of situations, including simultaneous localization and mapping =-=[31]-=-. In this approach, each measurement is checked against each object. Any measurement that is sufficiently close to the estimated target state is accepted as an association, ensuring that only one meas... |

438 | Using Occupancy Grids for Mobile Robot Perception and Navigation
- Elfes
- 1989
(Show Context)
Citation Context ...r of existing solutions to the problem of active SLAM. Carrillo et al. [10] present simulations and experiments utilizing a new utility function for active SLAM using an occupancy grid representation =-=[32]-=- of the environment. Mullane et al. [76, 77] provide the first solution to the SLAM problem based on FISST, looking at the problem of landmark-based mapping. They utilize the Rao-Blackwellized particl... |

364 |
Optimal design of experiments
- Pukelsheim
- 2006
(Show Context)
Citation Context ...al form of the distribution. Gaussian Distribution With a Gaussian distribution, the covariance matrix fully characterizes the spread of the distribution. From the theory of optimal experiment design =-=[9, 81]-=-, there are several standard optimality criteria that map a covariance matrix to a scalar while retaining useful statistical properties. The three most widely used criteria are: • A-optimality minimiz... |

340 | Near-Optimal sensor placements in Gaussian Processes: Theory, efficient August 23, 2012 DRAFT algorithms and empirical
- KRAUSE, SINGH, et al.
(Show Context)
Citation Context ... Number of length scales information was used by Krause and Guestrin [56], Krause et al. [58] to prove near-optimal static placement. The technique was extended to Gaussian processes by Krause et al. =-=[59]-=-. Different approximations were derived for the static sensor placement problem by Choi and How [14, 15], Choi et al. [16], and an informative trajectory planning algorithm was presented by Choi and H... |

299 |
An Introduction to the Theory
- Daley, Vere-Jones
- 2003
(Show Context)
Citation Context ...interested in functions f(X) that represent probability distributions over RFSs. The derivation of a probability distribution over RFSs has its roots in point process theory. See Daley and Vere-Jones =-=[21]-=- or Stone et al. [95, Chapter 5.1] for an overview of the subject. As is the case with random vectors, it is not possible to maintain an arbitrary distribution over RFSs: we must make some assumptions... |

193 |
Multitarget Bayes Filtering via First-Order Multitarget Moments,
- Mahler
- 2003
(Show Context)
Citation Context ... elements are i.i.d. and the cardinality follows a Poisson distribution. The likelihood of such an RFS is p(X) = e−λ ∏ x∈X v(x), (2.36) where λ = ∫ v(x) dx. The PHD filter was first derived by Mahler =-=[67]-=-. In its most generic form, it allows for arbitrary target motion, including the spawning (birth) of new targets and the disappearance of existing targets. In order to derive the PHD filter equations,... |

178 |
Receding horizon control of nonlinear systems
- Mayne, Michalska
- 1990
(Show Context)
Citation Context ...e number of targets is known. Our control policy for active perception for multi-target tracking builds on the literature on receding horizon control and model predictive control. Mayne and Michalska =-=[72]-=- provide a survey of receding horizon control and Mayne et al. [73] provide a survey of model predictive control, including applications in a variety of domains. Jadbabaie [48] utilizes model 31 predi... |

151 | Near-optimal sensor placements: Maximizing information while minimizing communication cost
- Krause, Guestrin, et al.
- 2006
(Show Context)
Citation Context ... likelihood c(z | q) Clutter PHD µ Expected clutter rate ·t Time index T Time horizon Termination criterion L Number of length scales information was used by Krause and Guestrin [56], Krause et al. =-=[58]-=- to prove near-optimal static placement. The technique was extended to Gaussian processes by Krause et al. [59]. Different approximations were derived for the static sensor placement problem by Choi a... |

144 | Near-optimal nonmyopic value of information in graphical models - Krause, Guestrin - 2005 |

141 | The Gaussian mixture Probability Hypothesis Density filter,”
- Vo, Ma
- 2006
(Show Context)
Citation Context ...ase with single target estimation strategies, it is not possible to maintain a generic density function over the state space of the targets. One approach to get around this limitation, from Vo and Ma =-=[101]-=-, is known as the Gaussian Mixture PHD (GM-PHD) filter and represents the PHD as a weighted mixture of Gaussians. In this, they assume that the target motion model and sensor model are linear Gaussian... |

122 |
Statistical Multisource-Multitarget Information Fusion
- Mahler
- 2007
(Show Context)
Citation Context ...presents several of the estimation algorithms. 2.2 Finite Set Statistics This section provides a summary of a set of tools known as Finite Set Statistics (FISST), which were first developed by Mahler =-=[70]-=-. The key feature of FISST is that collections of objects are represented by sets. This departs from traditional robotics solutions that use stacked vectors to represent collections of objects. Sectio... |

121 | Sequential Monte Carlo methods for multitarget filtering with random finite sets,”
- Vo, Singh, et al.
- 2005
(Show Context)
Citation Context ...01, Table II] for a simple pruning and merging strategy. Sequential Monte Carlo PHD Filter Another common approach is to represent the PHD as a set of weighted particles. This approach from Vo et al. =-=[102]-=- is known as the Sequential Monte Carlo PHD (SMC-PHD) filter. This is essentially equivalent to a standard particle filter, except that the weights of the particles are not normalized to have unit wei... |

111 | Maximizing a submodular set function subject to a matroid constraint, to appear in
- Calinescu, Chekuri, et al.
- 2008
(Show Context)
Citation Context ... X , is the target state. For each robot, let Yi be the candidate measurement sets (induced by the candidate robot poses). The collection of Yi, say Y ′ = {Y1, . . . ,YN}, defines a partition matroid =-=[7]-=-: (Y ′, I) where any Y ⊂ Y ′ ∈ I if and only if |Y ∩ Yi| ≤ 1, for all i. That is, Y is a valid assignment of trajectories if and only if it chooses at most one trajectory corresponding to each robot. ... |

94 | Information based adaptive robotic exploration,”
- Bourgault, Makarenko, et al.
- 2002
(Show Context)
Citation Context ...ion strategy. One common approach to robot control for active estimation is to maximize mutual information between the target locations and the robots' measurements. Grocholsky [41], Bourgault et al. =-=[6]-=-, and Cole [18] consider information-theoretic control of robot teams for exploration and tracking tasks using the Decentralized Data Fusion (DDF) architecture to handle inter-agent communication. In ... |

93 | Gradient of mutual information in linear vector Gaussian channels
- Palomar, Verdú
- 2006
(Show Context)
Citation Context ...of mutual information has not been reported in the literature despite the proliferation of gradient based methods for maximizing mutual information in various applications ranging from channel coding =-=[78, 79]-=-, to medical imaging alignment [100], to the control of robotic sensor networks [41]. In [79] the authors derive an expression that can be shown to be equivalent to a special case of Theorem 3 in whic... |

89 | Near-optimal observation selection using submodular functions
- Krause, Guestrin
- 2007
(Show Context)
Citation Context ...z | x,q) Measurement likelihood c(z | q) Clutter PHD µ Expected clutter rate ·t Time index T Time horizon Termination criterion L Number of length scales information was used by Krause and Guestrin =-=[56]-=-, Krause et al. [58] to prove near-optimal static placement. The technique was extended to Gaussian processes by Krause et al. [59]. Different approximations were derived for the static sensor placeme... |

81 | Information-theoretic control of multiple sensor platforms
- Grocholsky
- 2002
(Show Context)
Citation Context ... simple myopic exploration strategy. One common approach to robot control for active estimation is to maximize mutual information between the target locations and the robots' measurements. Grocholsky =-=[41]-=-, Bourgault et al. [6], and Cole [18] consider information-theoretic control of robot teams for exploration and tracking tasks using the Decentralized Data Fusion (DDF) architecture to handle inter-ag... |

64 |
Cooperative air and ground surveillance
- Grocholsky, Keller, et al.
(Show Context)
Citation Context ...his chapter was originally published in [28]. 5.1 Introduction Target detection, localization, and tracking has many applications including search-andrescue [36], wildlife tracking [98], surveillance =-=[42]-=-, and building smart cities [63]. Consequently, such problems have long been a subject of study in the robotics community. Target tracking typically refers to two types of tasks: estimating the trajec... |

62 | Stabilizing receding horizon control of nonlinear systems: a control Lyapunov function approach,”
- Jadbabaie, Yu, et al.
- 1999
(Show Context)
Citation Context ...ol. Mayne and Michalska [72] provide a survey of receding horizon control and Mayne et al. [73] provide a survey of model predictive control, including applications in a variety of domains. Jadbabaie =-=[48]-=- utilizes model 31 predictive control to follow trajectories with UAVs. The work of Ryan [89] is particularly relevant as it uses model predictive control in an information gathering setting, using a ... |

61 | Bayesian Multiple Target Tracking. Artech House - Stone, Barlow, et al. - 1999 |

56 | Mobile sensor network control using mutual information methods and particle filters,”
- Hoffmann, Tomlin
- 2007
(Show Context)
Citation Context ... has seen a lot of attention in recent years as a way of driving robots to localize and track targets. Mutual information is a common objective to use in target tracking problems. Hoffmann and Tomlin =-=[45]-=- and Julian et al. [51] use mutual information to localize a stationary target and explore unknown environments using a team of robots, assuming limited dependence between robots to achieve scalabilit... |

55 |
Community Resource for Archiving Wireless Data At Dartmouth (CRAWDAD). http://crawdad.cs.dartmouth.edu
- Kotz, Henderson
(Show Context)
Citation Context ...17] to stochastic [61]. Often, a 126 mixture of parametric motion models is used [62]. We take a data-driven approach to modeling the targets' motion, utilizing real-world datasets that are available =-=[55]-=-. In particular, we use Gaussian Process (GP) regression [83] to learn the function that maps the position coordinates of the targets to velocity vectors, as shown by Joseph et al. [49]. GP regression... |

53 |
PHD Filters of Higher Order in Target Number,
- Mahler
- 2007
(Show Context)
Citation Context ...ty estimate has high variance, particularly when tracking a large number of targets, due to the fact that the mean and variance of a Poisson distribution are equal. To get around these issues, Mahler =-=[69]-=- developed the CPHD filter. The CPHD filter makes the same assumptions as the PHD filter, except instead of Poisson RFSs, everything is assumed to be an i.i.d. cluster process. This allows for an arbi... |

48 | Analytic implementations of the Cardinalized Probability Hypothesis Density Filter,”
- Vo, Vo, et al.
- 2007
(Show Context)
Citation Context ...t the maximum number of targets that can be tracked is fixed a priori and the cardinality 26 estimate will be biased if the true number of targets is close to the maximum value, as noted by Vo et al. =-=[103]-=-. Before stating the CPHD prediction and update rules, we must define a few variables. Let pt(n) be the likelihood of n targets at time t, pγ,t(n) be the cardinality distribution of the birth process ... |

46 | Dynamic sensor planning and control for optimally tracking targets,
- Spletzer, Taylor
- 2003
(Show Context)
Citation Context ...iew can be formulated as a visual servoing problem. Gans et al. [37] design a controller which guarantees stability while keeping three or fewer targets in the field-of-view of a single mobile robot. =-=[94]-=- present a general solution for the multi-robot, multi-target case using a particle filter formulation. Tracking multiple targets with multiple robots requires explicit or implicit assignment of targe... |

39 |
Data fusion in decentralized sensor networks
- Grime, Durrant-Whyte
- 1994
(Show Context)
Citation Context ...le environments where not all robots will be able to communicate with one another. One common decentralized architecture is Decentralized Data Fusion (DDF), first described by Grime and Durrant-Whyte =-=[40]-=-, in which each robot manages its own copy of the joint belief and aggregates data from the other robots through channel filters which only admit information that is distinct from their current belief... |

39 | Decentralised coordination of mobile sensors using the Max-Sum algorithm
- Stranders, Farinelli, et al.
- 2009
(Show Context)
Citation Context ...lar, Cole [18] examines the scenario where the number of targets is unknown, deriving equations similar to those of the PHD filter but using a very conservative data fusion approach. Stranders et al. =-=[96]-=- and Delle Fave et al. [30] use the max-sum algorithm for decentralized control computations and DDF to share beliefs about target locations. However, all of these approaches make restrictive assumpti... |

35 |
CRAWDAD data set epfl/mobility (v. 2009-02-24),” Downloaded from http://crawdad.cs.dartmouth.edu/epfl/mobility,
- Piorkowski, Sarafijanovic-Djukic, et al.
- 2009
(Show Context)
Citation Context ... to a number of robot and sensor models, for the purposes of testing we restrict our attention to fixed winged aerial robots with downward facing cameras. We use a real-world taxi motion dataset from =-=[80]-=- for the targets and to verify our models. The simulation results reveal that robot teams using the information-based control objective track a smaller number of targets with higher precision compared... |

34 |
Recursive Bayesian search-and-tracking using coordinated UAVs for lost targets
- Furukawa, Bourgault, et al.
- 2006
(Show Context)
Citation Context ...ing a real-world dataset. The research in this chapter was originally published in [28]. 5.1 Introduction Target detection, localization, and tracking has many applications including search-andrescue =-=[36]-=-, wildlife tracking [98], surveillance [42], and building smart cities [63]. Consequently, such problems have long been a subject of study in the robotics community. Target tracking typically refers t... |

30 |
The cardinality balanced multitarget multi-Bernoulli filter and its implementations,”
- Vo, Vo, et al.
- 2009
(Show Context)
Citation Context ...xtension to Other Estimation Algorithms The FISST community has recently developed new estimation algorithms based on random finite sets. The Cardinalized Multi-target Multi-Bernoulli (MeMBer) filter =-=[104]-=- and labeled MeMBer filter [84] represent targets using the so-called Bernoulli RFSs. A Bernoulli RFS has two components, a probability of existence and a probability density of the state of the targe... |

27 |
An information theoretic approach to observer path design for bearings-only tracking,”
- Logothetis, Isaksson, et al.
- 1997
(Show Context)
Citation Context ...ctive target tracking problems have been studied in the literature under many different settings. Solutions have been presented for radio-based sensors [46], range-only sensors [108], bearing sensors =-=[64]-=-, and range and/or bearing sensors [109], under centralized and decentralized settings. Frew and Rock [34] design optimal trajectories for a single robot to track a single moving target using monocula... |

26 | Representation of mutual information via input estimates,”
- Palomar, Verdu
- 2007
(Show Context)
Citation Context ...of mutual information has not been reported in the literature despite the proliferation of gradient based methods for maximizing mutual information in various applications ranging from channel coding =-=[78, 79]-=-, to medical imaging alignment [100], to the control of robotic sensor networks [41]. In [79] the authors derive an expression that can be shown to be equivalent to a special case of Theorem 3 in whic... |

24 | Sensor Andrew: Large-scale campus-wide sensing and actuation,”
- Rowe, Berges, et al.
- 2011
(Show Context)
Citation Context ...ate suspicious packages in a shopping center, or using wireless pings to locate sensors within a smart building or smart city. Real-world examples of such smart building scenarios include Rowe et al. =-=[88]-=-, which features thermostats, microphones, access points, and bluetooth-enabled actuators within a building, and Fu [35], which describes low-power sensors embedded within construction materials. In e... |

20 |
Active planning for underwater inspection and the benefit of adaptivity,”
- Hollinger, Englot, et al.
- 2013
(Show Context)
Citation Context ...51] use mutual information to localize a stationary target and explore unknown environments using a team of robots, assuming limited dependence between robots to achieve scalability. Hollinger et al. =-=[47]-=- use an information-based objective function to perform autonomous ship inspection with an AUV platform. The robot may also move to maximize the immediate information gain, a strategy sometimes known ... |

20 |
A random-finite-set approach to Bayesian SLAM,”
- Mullane, Vo, et al.
- 2011
(Show Context)
Citation Context ...tainty. This ideas runs counter to the probabilistic approach often used in tracking problems, and means making an incorrect association can have a long-lasting impact on the estimate. Mullane et al. =-=[76]-=- show examples of such errors in the context of feature-based mapping. Another approach is the Multiple Hypothesis Tracker (MHT) [95, Chapter 4.2], which assumes the data association to be an unknown ... |

20 |
W.:Taxonomy of multiple target tracking methods,
- Pulford
- 2005
(Show Context)
Citation Context ...mber of targets is very large, the CPHD filter will be significantly slower. 2.2.4 Literature Review Multi-target tracking has also been addressed extensively in the radar tracking community; Pulford =-=[82]-=- provides a taxonomy of techniques. Recently the use of random finite sets has been adopted in mobile robotics, being used for feature-based mapping by Mullane et al. [76, 77]. Lundquist et al. [65] u... |

19 | A bayesian non- parametric approach to modeling mobility patterns,” Autonomous Robots,
- Joseph, Doshi-Velez, et al.
- 2011
(Show Context)
Citation Context ...o extract the motion model, instead of assuming any parametric form. Specifically, we use Gaussian Process (GP) regression to learn a map of velocity vectors for the targets, similar to Joseph et al. =-=[49]-=-. Additionally, we show how to model the appearance and disappearance of targets within the environment. Next, we present a control policy to assign trajectories for all robots in order to maximize th... |

19 |
Community: An Internet of Things Application,”
- Li
- 2011
(Show Context)
Citation Context ...shed in [28]. 5.1 Introduction Target detection, localization, and tracking has many applications including search-andrescue [36], wildlife tracking [98], surveillance [42], and building smart cities =-=[63]-=-. Consequently, such problems have long been a subject of study in the robotics community. Target tracking typically refers to two types of tasks: estimating the trajectories of the targets from the s... |

18 | Survey of maneuvering target tracking. part ii: Motion models of ballistic and space targets. Aerospace and Electronic Systems,
- Li, Jilkov
- 2010
(Show Context)
Citation Context ...results towards solving the problem. An important consideration for target tracking is the motion model for the targets. A number of parametric motion models have been proposed in the literature (see =-=[61]-=- for a detailed survey). We employ a data-driven technique to extract the motion model, instead of assuming any parametric form. Specifically, we use Gaussian Process (GP) regression to learn a map of... |

17 | The bin-occupancy filter and its connection to the PHD filters
- Erdinc, Willett, et al.
- 2009
(Show Context)
Citation Context ...lter While the PHD filter is attractive due to its low computational complexity and relatively simple implementation, it suffers from two potential drawbacks. Firstly, as pointed out by Erdinc et al. =-=[33]-=-, the PHD filter deals poorly with false negatives, drastically decreasing the likelihood of a target being within a given region if no detection is made. Secondly, the target cardinality estimate has... |

15 |
Hero III. Sensor management using an active sensing approach
- Kreucher, Kastella, et al.
- 2005
(Show Context)
Citation Context ...e y = 0 is the event that the robot receives no measurements to any (true or clutter) objects while y = 1 is the complement of this, i.e., the robot receives at least one measurement. Kreucher et al. =-=[60]-=- take a similar approach, using a binary sensor model and an information-based objective function to schedule sensors to track an unknown number of targets. Theorem 4. The mutual information between t... |

15 |
A multi-robot control policy for information gathering in the presence of unknown hazards
- Schwager, Dames, et al.
- 2011
(Show Context)
Citation Context ... avoid hazardous areas as a failed robot provides no information, naturally merging the objectives of localizing targets and avoiding hazards. The research in this chapter was originally published in =-=[26, 27, 90]-=-. 3.1 Problem Formulation Consider a situation where n robots move in a bounded, planar environment E ⊂ R2. Robot i is at position qit ∈ E at time t, and the positions of all the robots can be written... |

14 | On-line boostingbased car detection from arial images”,
- Grabner, Nguyen, et al.
- 2007
(Show Context)
Citation Context ...cles or quadrotor platforms) and other sensor modalities (e.g., lidars or 3D depth cameras). 5.3.1 Sensor Parameterization The problem of detecting vehicles using aerial imagery has been well studied =-=[39, 107]-=-. We use such studies to inform our selection of the sensor detection, measurement, and clutter models. The approaches presented in [39, 107] are similar, searching for image features over a range of ... |

14 |
Robots with their heads in the clouds
- Guizzo
- 2011
(Show Context)
Citation Context ...additional computational resources, and to cloud services. This idea of robots relying on information from a server has been called cloud robotics and has recently generated quite a bit of excitement =-=[43, 44]-=-. A similar idea was also used for estimation and control of groups of robots by Michael et al. [74] where an Asymmetric Broadcast Control (ABC) was used to synthesize locally derived information and ... |

13 | Approximate representations for multi-robot control policies that maximize mutual information
- Charrow, Kumar, et al.
- 2013
(Show Context)
Citation Context ...wn as information surfing [19]. Julian et al. [52] and Souza et al. [93] utilize mutual information to drive a single robot to explore an unknown environment in order to build a map. Charrow et al. =-=[11, 12]-=- use mutual information to drive a team of robots equipped with range-only sensors to track a single moving target in real time and to detect and localize an unknown number of targets with known data ... |

13 | A scalable information theoretic approach to distributed robot coordination
- Julian, Angermann, et al.
(Show Context)
Citation Context ...to perform autonomous ship inspection with an AUV platform. The robot may also move to maximize the immediate information gain, a strategy sometimes known as information surfing [19]. Julian et al. =-=[50]-=- use the gradient of mutual information to drive multiple robots for state estimation tasks, a strategy sometimes known as information surfing [19]. Julian et al. [52] and Souza et al. [93] utilize ... |

12 |
Improved smc implementation of the phd filter,” in
- Ristic, Clark, et al.
- 2010
(Show Context)
Citation Context ...the sensor. New particles are added to the PHD using the birth PHD described above as well as using the most recent measurement set and inverse measurement model, similar to the idea of Ristic et al. =-=[86]-=-. A fixed number of particles, Pb, are drawn from the birth PHD and an additional Pm particles are drawn from the inverse measurement model for each measurements in the most recent set, Zt. The weight... |

12 |
A Note on the Reward Function for PHD Filters with Sensor Control,”
- Ristic, Vo, et al.
- 2011
(Show Context)
Citation Context ... number of targets. There is a relatively limited body of work on active control for target localization based on the RFS framework, with the exception of work by Ristic and Vo [85] and Ristic et al. =-=[87]-=- to maximize information using Rényi's definition. Ristic and Vo [85] track the target positions using samples from the distribution over RFSs, as in Section 2.2.3. In this work, the measurement model... |

12 | Probabilistic Search with Agile UAVs
- Waharte, Symington, et al.
- 2010
(Show Context)
Citation Context ...zero, while if the robot can see most of the cell then µ will be larger. However, the integration does not take into account the area viewed during previous time steps, as was noted by Waharte et al. =-=[105]-=-. Failures of multiple robots are assumed to be conditionally independent of one another given the positions of the hazards so that, p(f | X,H,Q) = ∏ i p(f i | H,qi), (3.5) where p(f i | H,qi) comes f... |

11 |
Distributed Robotic Sensor Networks: An Information-Theoretic Approach,”
- Julian, Angermann, et al.
- 2012
(Show Context)
Citation Context ...ntion in recent years as a way of driving robots to localize and track targets. Mutual information is a common objective to use in target tracking problems. Hoffmann and Tomlin [45] and Julian et al. =-=[51]-=- use mutual information to localize a stationary target and explore unknown environments using a team of robots, assuming limited dependence between robots to achieve scalability. Hollinger et al. [47... |

11 |
Architecture, abstractions, and algorithms for controlling large teams of robots: Experimental testbed and results
- Michael, Fink, et al.
- 2007
(Show Context)
Citation Context ...om a server has been called cloud robotics and has recently generated quite a bit of excitement [43, 44]. A similar idea was also used for estimation and control of groups of robots by Michael et al. =-=[74]-=- where an Asymmetric Broadcast Control (ABC) was used to synthesize locally derived information and provide low-resolution global information to the group. The asymmetry is in the communication betwee... |

10 | Continuous trajectory planning of mobile sensors for informative forecasting. Automatica
- Choi, How
- 2011
(Show Context)
Citation Context ...ifferent approximations were derived for the static sensor placement problem by Choi and How [14, 15], Choi et al. [16], and an informative trajectory planning algorithm was presented by Choi and How =-=[13]-=-. Unfortunately, our algorithm cannot make use of these near-optimality results because the sequential updating of our distribution destroys the submodularity property of mutual information. Other wor... |

10 |
Objective functions for Bayesian control-theoretic sensor management, I: Multitarget first-moment approximation,”
- Mahler
- 2003
(Show Context)
Citation Context ...event that the robot receives no measurements to any (true or clutter) objects while y = 1 is the complement of this, i.e., the robot receives at least one measurement. Mahler proposes a similar idea =-=[68]-=-, where the objective is to maximize the mutual information between the target set and the empty measurement set, i.e., when p(Z = ∅) = 1, so q∗ = argmax q I[X,Z(q) = ∅]. (4.2) This objective is chose... |

10 |
Disruptive technologies: Advances that will transform life, business, and the global economy,
- MANYIKA, CHU, et al.
- 2013
(Show Context)
Citation Context ... the future measurements of the robots. 6.2.4 Interacting With the Internet of Things According to the McKinsey Global Institute, there are twelve key technologies that will transform the way we live =-=[71]-=-. Among those are advanced robotics, the Internet of Things (IoT), and cloud technology. As the IoT continues to expand, wireless devices and sensors will become increasingly prevalent in the environm... |

8 |
Keeping multiple moving targets in the field of view of a mobile camera
- Gans, Hu, et al.
- 2011
(Show Context)
Citation Context ...tories for a single robot to track a single moving target using monocular vision. The problem of keeping targets in a robot's field-of-view can be formulated as a visual servoing problem. Gans et al. =-=[37]-=- design a controller which guarantees stability while keeping three or fewer targets in the field-of-view of a single mobile robot. [94] present a general solution for the multi-robot, multi-target ca... |

8 |
Cloud robotics: Connected to the cloud, robots get smarter
- Guizzo
- 2011
(Show Context)
Citation Context ...additional computational resources, and to cloud services. This idea of robots relying on information from a server has been called cloud robotics and has recently generated quite a bit of excitement =-=[43, 44]-=-. A similar idea was also used for estimation and control of groups of robots by Michael et al. [74] where an Asymmetric Broadcast Control (ABC) was used to synthesize locally derived information and ... |

8 | Road intensity based mapping using radar measurements with a probability hypothesis density filter
- Lundquist, Hammarstrand, et al.
- 2011
(Show Context)
Citation Context ...d [82] provides a taxonomy of techniques. Recently the use of random finite sets has been adopted in mobile robotics, being used for feature-based mapping by Mullane et al. [76, 77]. Lundquist et al. =-=[65]-=- use a PHD filter for extended objects (i.e., objects that return multiple measurements) to create an obstacle map for a vehicle. Atanasov et al. [4] present an approach to localize a robot in a seman... |

7 | Searching for Multiple Targets Using Probabilistic Quadtrees
- Carpin, Burch, et al.
- 2011
(Show Context)
Citation Context ... quadtree data structure. However, the methodology can be easily extended to work with other decompositions. Quadtrees have been used in other localization tasks, such as in the work of Carpin et al. =-=[8]-=-. What distinguishes this approach is the use of finite set statistics and the ability to remove leaves from the tree. 43 The main idea is that initially a coarse discretization is used, which is refi... |

7 | On the comparison of uncertainty criteria for active SLAM
- Carrillo, Reid, et al.
- 2012
(Show Context)
Citation Context ...al form of the distribution. Gaussian Distribution With a Gaussian distribution, the covariance matrix fully characterizes the spread of the distribution. From the theory of optimal experiment design =-=[9, 81]-=-, there are several standard optimality criteria that map a covariance matrix to a scalar while retaining useful statistical properties. The three most widely used criteria are: • A-optimality minimiz... |

7 | A decentralized control policy for adaptive information gathering in hazardous environments
- Dames, Schwager, et al.
(Show Context)
Citation Context ... avoid hazardous areas as a failed robot provides no information, naturally merging the objectives of localizing targets and avoiding hazards. The research in this chapter was originally published in =-=[26, 27, 90]-=-. 3.1 Problem Formulation Consider a situation where n robots move in a bounded, planar environment E ⊂ R2. Robot i is at position qit ∈ E at time t, and the positions of all the robots can be written... |

7 |
Trajectory generation for constant velocity target motion estimation using monocular vision
- Frew, Rock
- 2003
(Show Context)
Citation Context ...s have been presented for radio-based sensors [46], range-only sensors [108], bearing sensors [64], and range and/or bearing sensors [109], under centralized and decentralized settings. Frew and Rock =-=[34]-=- design optimal trajectories for a single robot to track a single moving target using monocular vision. The problem of keeping targets in a robot's field-of-view can be formulated as a visual servoing... |

6 |
An outer-approximation algorithm for generalized maximum entropy sampling
- Choi, How, et al.
(Show Context)
Citation Context ... placement. The technique was extended to Gaussian processes by Krause et al. [59]. Different approximations were derived for the static sensor placement problem by Choi and How [14, 15], Choi et al. =-=[16]-=-, and an informative trajectory planning algorithm was presented by Choi and How [13]. Unfortunately, our algorithm cannot make use of these near-optimality results because the sequential updating of ... |

6 |
Deploying the max-sum algorithm for decentralised coordination and task allocation of unmanned aerial vehicles for live aerial imagery collection
- Fave, Rogers, et al.
- 2012
(Show Context)
Citation Context ... scenario where the number of targets is unknown, deriving equations similar to those of the PHD filter but using a very conservative data fusion approach. Stranders et al. [96] and Delle Fave et al. =-=[30]-=- use the max-sum algorithm for decentralized control computations and DDF to share beliefs about target locations. However, all of these approaches make restrictive assumptions on the form of the dist... |

5 | Efficient targeting of sensor networks for large-scale systems
- Choi, How
- 2010
(Show Context)
Citation Context ...ve near-optimal static placement. The technique was extended to Gaussian processes by Krause et al. [59]. Different approximations were derived for the static sensor placement problem by Choi and How =-=[14, 15]-=-, Choi et al. [16], and an informative trajectory planning algorithm was presented by Choi and How [13]. Unfortunately, our algorithm cannot make use of these near-optimality results because the seque... |

5 |
Cooperative Multi-Target Localization With Noisy Sensors
- Dames, Kumar
- 2013
(Show Context)
Citation Context ...he environment or by downloading from the server, merging the objectives of gathering and disseminating information into a single control law. The research in this chapter was originally published in =-=[22, 24, 25]-=-. 4.1 Problem Formulation This chapter considers the problem of a team of R robots exploring an environment E in search of targets. The robots are assumed to be able to localize themselves within the ... |

5 |
Target tracking without line of sight using range from radio. Autonomous Robots 32(1
- GA, Djugash, et al.
- 2012
(Show Context)
Citation Context ...d number of detections. 123 5.1.1 Related Work Active target tracking problems have been studied in the literature under many different settings. Solutions have been presented for radio-based sensors =-=[46]-=-, range-only sensors [108], bearing sensors [64], and range and/or bearing sensors [109], under centralized and decentralized settings. Frew and Rock [34] design optimal trajectories for a single robo... |

5 |
Random finite sets for robot mapping and slam
- Mullane, Vo, et al.
- 2011
(Show Context)
Citation Context ... tracking community; Pulford [82] provides a taxonomy of techniques. Recently the use of random finite sets has been adopted in mobile robotics, being used for feature-based mapping by Mullane et al. =-=[76, 77]-=-. Lundquist et al. [65] use a PHD filter for extended objects (i.e., objects that return multiple measurements) to create an obstacle map for a vehicle. Atanasov et al. [4] present an approach to loca... |

5 | Simultaneous localization of multiple unknown and transient radio sources using a mobile robot,” Robotics,
- Song, Kim, et al.
- 2012
(Show Context)
Citation Context ...is work to allow for the case of an unknown and changing number of targets. Recently, there has been some work on actively detecting and/or localizing an unknown number of targets using radio sensors =-=[54, 92]-=-, range-only sensors [12], and arbitrary sensor models [25]. Kim et al. [54], Song et al. [92] studied the problem of detecting and localizing an unknown number of radio sources. Unlike all these work... |

4 |
Tracking Aquatic Invaders: Autonomous Robots for Monitoring Invasive Fish
- Tokekar, Branson, et al.
(Show Context)
Citation Context .... The research in this chapter was originally published in [28]. 5.1 Introduction Target detection, localization, and tracking has many applications including search-andrescue [36], wildlife tracking =-=[98]-=-, surveillance [42], and building smart cities [63]. Consequently, such problems have long been a subject of study in the robotics community. Target tracking typically refers to two types of tasks: es... |

3 | Coordinated targeting of mobile sensor networks for ensemble forecast improvement
- Choi, How
- 2010
(Show Context)
Citation Context ...ve near-optimal static placement. The technique was extended to Gaussian processes by Krause et al. [59]. Different approximations were derived for the static sensor placement problem by Choi and How =-=[14, 15]-=-, Choi et al. [16], and an informative trajectory planning algorithm was presented by Choi and How [13]. Unfortunately, our algorithm cannot make use of these near-optimality results because the seque... |

3 |
A cooperative uas architecture for information-theoretic search and track
- Cole
- 2009
(Show Context)
Citation Context ...One common approach to robot control for active estimation is to maximize mutual information between the target locations and the robots' measurements. Grocholsky [41], Bourgault et al. [6], and Cole =-=[18]-=- consider information-theoretic control of robot teams for exploration and tracking tasks using the Decentralized Data Fusion (DDF) architecture to handle inter-agent communication. In particular, Col... |

3 |
Playing Fetch with Your Robot: The Ability of Robots to Locate and Interact with Objects
- Dames, Thakur, et al.
- 2013
(Show Context)
Citation Context ... avoid hazardous areas as a failed robot provides no information, naturally merging the objectives of localizing targets and avoiding hazards. The research in this chapter was originally published in =-=[26, 27, 90]-=-. 3.1 Problem Formulation Consider a situation where n robots move in a bounded, planar environment E ⊂ R2. Robot i is at position qit ∈ E at time t, and the positions of all the robots can be written... |

3 | Multi-target visual tracking with aerial robots
- Tokekar, Isler, et al.
- 2014
(Show Context)
Citation Context ...y two objective functions using the PHD filter: mutual information and the expected number of detections by the robots. We show that both objective functions are submodular, and use a result based on =-=[99]-=- to prove that our greedy control policy is a 2-approximation. In addition to the theoretical analysis we offer, we evaluate our algorithm using simulated experiments. While our framework may be appli... |

2 | Information Acquisition with Sensing Robots: Algorithms and Error Bounds
- Atanasov, Ny, et al.
- 2014
(Show Context)
Citation Context ...vely chain such actions, there would be a exponentially growing number of possible actions. However, many of these would be redundant, i.e., the robots would traverse the same region. Atanasov et al. =-=[3]-=- present one solution to this problem, pruning the tree of motion primitives to eliminate uninformative and redundant actions. We take an alternative approach to curb the number of actions while maint... |

2 | Decentralized Active Information Acquisition: Theory and Application to Multi-Robot SLAM
- Atanasov, Ny, et al.
- 2015
(Show Context)
Citation Context ...r each coordinate in sequence reaching until a local optimum. In practice, the number of cycles through the team until reaching consensus was typically one, and never more than three. Atanasov et al. =-=[5]-=- show that this coordinate descent will result in a 2-approximation of the objective, meaning that the value of the objective using the approximate approach will be at least half of the value of the... |

2 | Active control strategies for discovering and localizing devices with range-only sensors
- Charrow, Michael, et al.
- 2015
(Show Context)
Citation Context ... identify individual targets and that it is able to do so without error. This is a valid assumption in some settings, e.g., localizing wireless sensors using the MAC address to provide a unique label =-=[12]-=-. However, in many other systems there could easily be errors in the data association, e.g., when tracking people in a crowd using facial recognition software, or association may not be possible, e.g.... |

2 |
A Hollinger, and Volkan Isler. Search and pursuit-evasion in mobile robotics
- Chung, Geoffrey
(Show Context)
Citation Context ...lves, we need models for the motion of individual targets as well as the birth/death processes of the targets. A number of motion models have been proposed in the literature, ranging from adversarial =-=[17]-=- to stochastic [61]. Often, a 126 mixture of parametric motion models is used [62]. We take a data-driven approach to modeling the targets' motion, utilizing real-world datasets that are available [55... |

2 |
On Mutual Informationbased Control of Range Sensing Robots for Mapping Applications
- Julian, Karaman, et al.
(Show Context)
Citation Context ...n surfing [19]. Julian et al. [50] use the gradient of mutual information to drive multiple robots for state estimation tasks, a strategy sometimes known as information surfing [19]. Julian et al. =-=[52]-=- and Souza et al. [93] utilize mutual information to drive a single robot to explore an unknown environment in order to build a map. Charrow et al. [11, 12] use mutual information to drive a team of r... |

2 | Robust multiBernoulli filtering,”
- Vo, Vo, et al.
- 2013
(Show Context)
Citation Context ...orithms The FISST community has recently developed new estimation algorithms based on random finite sets. The Cardinalized Multi-target Multi-Bernoulli (MeMBer) filter [104] and labeled MeMBer filter =-=[84]-=- represent targets using the so-called Bernoulli RFSs. A Bernoulli RFS has two components, a probability of existence and a probability density of the state of the target, and represents the estimate ... |

2 |
Information-theoretic control for mobile sensor teams
- Ryan
- 2008
(Show Context)
Citation Context ...3] provide a survey of model predictive control, including applications in a variety of domains. Jadbabaie [48] utilizes model 31 predictive control to follow trajectories with UAVs. The work of Ryan =-=[89]-=- is particularly relevant as it uses model predictive control in an information gathering setting, using a small team of UAVs to localize and track a moving target. We adapt this work to the multi-tar... |

1 |
robotics and automation magazine
- IEEE
- 2014
(Show Context)
Citation Context ...to localize a robot in a semantic map using an approximation algorithm to solve the data association (2.26). Other applications of FISST in robotic mapping, target tracking, and SLAM are presented in =-=[2]-=-. 2.3 Active Information Gathering Information-based control is a common tool for information gathering tasks. The intuition is to drive the team of robots in a way that minimizes some measure of unce... |

1 |
Autonomous Robotic Exploration Using Occupancy Grid
- Carrillo, Dames, et al.
- 2015
(Show Context)
Citation Context ...tion and mapping (SLAM). SLAM has been an area of active research within the robotics community for decades and there are a number of existing solutions to the problem of active SLAM. Carrillo et al. =-=[10]-=- present simulations and experiments utilizing a new utility function for active SLAM using an occupancy grid representation [32] of the environment. Mullane et al. [76, 77] provide the first solution... |

1 |
Chaouki T Abdallah. Information Surfing for Radiation Map Building
- Cortez, Tanner, et al.
(Show Context)
Citation Context ... objective function to perform autonomous ship inspection with an AUV platform. The robot may also move to maximize the immediate information gain, a strategy sometimes known as information surfing =-=[19]-=-. Julian et al. [50] use the gradient of mutual information to drive multiple robots for state estimation tasks, a strategy sometimes known as information surfing [19]. Julian et al. [52] and Souza ... |

1 |
Experimental Characterization of a Bearing-only Sensor for Use With the PHD Filter, 2015. Available at: arXiv:1502.04661 [cs:RO
- Dames, Kumar
(Show Context)
Citation Context ...s to each of the subsequent clusters form a measurement set Z. 4.4.1 Sensor Models We now develop the detection, measurement, and clutter models necessary to utilize the PHD filter. See Appendix B or =-=[23]-=- for further details on the experimental characterization of the sensors. 105 Figure 34: A pictogram of the laser detection model, where dt is the diameter of the target, θsep is the angular separatio... |

1 |
Multi-Robot Detection and Localization of Targets Using Probability Hypothesis Density Functions
- Dames, Kumar
- 2015
(Show Context)
Citation Context ...he environment or by downloading from the server, merging the objectives of gathering and disseminating information into a single control law. The research in this chapter was originally published in =-=[22, 24, 25]-=-. 4.1 Problem Formulation This chapter considers the problem of a team of R robots exploring an environment E in search of targets. The robots are assumed to be able to localize themselves within the ... |

1 |
Autonomous Localization of an Unknown Number of Targets without Data Association Using Teams of Mobile Sensors
- Dames, Kumar
- 2015
(Show Context)
Citation Context ...he environment or by downloading from the server, merging the objectives of gathering and disseminating information into a single control law. The research in this chapter was originally published in =-=[22, 24, 25]-=-. 4.1 Problem Formulation This chapter considers the problem of a team of R robots exploring an environment E in search of targets. The robots are assumed to be able to localize themselves within the ... |

1 |
Tracking an Unknown Number of Moving Targets Using a Team of Mobile Robots
- Detecting
- 2015
(Show Context)
Citation Context ...robots, and the expected number of targets detected by the robot team. We provide extensive simulation evaluations using a real-world dataset. The research in this chapter was originally published in =-=[28]-=-. 5.1 Introduction Target detection, localization, and tracking has many applications including search-andrescue [36], wildlife tracking [98], surveillance [42], and building smart cities [63]. Conseq... |

1 |
Sujit Kuthirummal, Zsolt Kira, and Mihail Pivtoraiko. R-MASTIF: Robotic Mobile Autonomous System for Threat Interrogation and Object Fetch
- Das, Thakur, et al.
- 2013
(Show Context)
Citation Context ...robot. The system used in these experiments uses a template matching algorithm, using shape and color information, to return the 2D positions of all known objects within the field of view. Das et al. =-=[29]-=- provide details on the vision subsystem and other components of the robot. The sensor returns a list of cells occupied by objects. The likelihood of such a measurement set is given by the expression ... |

1 |
RFID-Scale Devices in Concrete
- Fu
- 2014
(Show Context)
Citation Context ... city. Real-world examples of such smart building scenarios include Rowe et al. [88], which features thermostats, microphones, access points, and bluetooth-enabled actuators within a building, and Fu =-=[35]-=-, which describes low-power sensors embedded within construction materials. In each of these examples, the number of objects is not known a priori, and can potentially be very large. The sensors used ... |

1 | Cooperative search of multiple unknown transient radio sources using multiple paired mobile robots
- Kim, Song, et al.
- 2014
(Show Context)
Citation Context ...is work to allow for the case of an unknown and changing number of targets. Recently, there has been some work on actively detecting and/or localizing an unknown number of targets using radio sensors =-=[54, 92]-=-, range-only sensors [12], and arbitrary sensor models [25]. Kim et al. [54], Song et al. [92] studied the problem of detecting and localizing an unknown number of radio sources. Unlike all these work... |

1 |
Randy A Freeman. Decentralized Environmental Modeling by Mobile Sensor Networks
- Lynch, Schwartz, et al.
- 2008
(Show Context)
Citation Context ...stribution destroys the submodularity property of mutual information. Other works concentrate on specific models of target positions or environmental fields. For example the algorithm by Lynch et al. =-=[66]-=- drives robots to decrease the error variance of a distributed Kalman filter estimate of a Gaussian environmental field. By contrast, our algorithm does not make assumptions about the Gaussianity of t... |

1 |
Collaborative Multi-Vehicle SLAM with Moving Object Tracking
- Moratuwage, Vo, et al.
- 2013
(Show Context)
Citation Context ...ot. They show that this approach leads to superior performance compared to traditional approaches to SLAM that utilize heuristic methods for 149 data association and map management. Moratuwage et al. =-=[75]-=- extend the RB-PHD-SLAM algorithm to use multiple feature classes and to work with multiple robots. Each feature class has a different motion model associated with it, allowing the robot team to track... |

1 |
Bayesian Optimisation for Active Perception and Smooth Navigation
- Souza, Marchant, et al.
- 2014
(Show Context)
Citation Context ...n et al. [50] use the gradient of mutual information to drive multiple robots for state estimation tasks, a strategy sometimes known as information surfing [19]. Julian et al. [52] and Souza et al. =-=[93]-=- utilize mutual information to drive a single robot to explore an unknown environment in order to build a map. Charrow et al. [11, 12] use mutual information to drive a team of robots equipped with ra... |