## Sensor Based Motion Planning: The Hierarchical Generalized Voronoi Graph (1996)

### Cached

### Download Links

Citations: | 78 - 9 self |

### BibTeX

@MISC{Choset96sensorbased,

author = {Howie Choset},

title = {Sensor Based Motion Planning: The Hierarchical Generalized Voronoi Graph},

year = {1996}

}

### Years of Citing Articles

### OpenURL

### Abstract

The hierarchical generalized Voronoi graph (HGVG) is a roadmap that can serve as a basis for sensor based robot motion planning. A key feature of the HGVG is its incremental construction procedure that uses only line of sight distance information. This work describes basic properties of the HGVG and the procedure for its incremental construction using local range sensors. Simulations and experiments verify this approach. 1 Introduction Sensor based motion planning incorporates sensor information, reflecting the current state of the environment, into a robot's planning process, as opposed to classical planning, which assumes full knowledge of the world's geometry prior to planning. Sensor based planning is important for realistic deployment of robots because: (1) the robot often has no a priori knowledge of the world; (2) the robot may have only a coarse knowledge of the world because of limited computer memory; (3) the world model is bound to contain inaccuracies which can be overcom...

### Citations

2671 |
A robust layered control system for a mobile robot
- Brooks
(Show Context)
Citation Context ...he geometry and topology of the problem is simple. One class of heuristic algorithms is a behavioral based approach in which the robot is armed with a simple set of behaviors such as following a wall =-=[5]-=-. A hierarchy of cooperating behaviors forms more complicated behaviors such as exploration. An extension of this type of approach is called sequencing [20, 21, 22]. Since there are strong experimenta... |

1988 |
Robot Motion Planning
- Latombe
- 1991
(Show Context)
Citation Context ...ed as the union of convex obstacles. Three classes of complete classical motion planning schemes have been developed: potential field approaches, roadmap methods, and cellular decomposition (Latombe, =-=[28]-=-). Potential Fields In the potential field approach (Khatib, [25]; Latombe, [28]), the robot is modeled as a particle acting under the influence of a potential function U : R m ! R that encodes inform... |

956 |
Optimization and nonsmooth analysis
- Clarke
- 1990
(Show Context)
Citation Context ...inition 2.1 (Single Object Distance Function) The distance between a point, x and a convex set C i is d i (x) = min c 0 2C i kx \Gamma c 0 k; (2.1) where k \Delta k is the two-norm in R m . In Clarke =-=[19]-=- it is shown that the gradient of d i (x) is rd i (x) = x \Gamma c 0 kx \Gamma c 0 k ; (2.2) 21 d i d i c 0 x C i Figure 2.1 Distance between x and C i is the distance to the closest point on C i . Th... |

909 |
Real-time Obstacle Avoidance for Manipulators and Mobile Robots
- Khatib
- 1986
(Show Context)
Citation Context ...assical motion planning schemes have been developed: potential field approaches, roadmap methods, and cellular decomposition (Latombe, [28]). Potential Fields In the potential field approach (Khatib, =-=[25]-=-; Latombe, [28]), the robot is modeled as a particle acting under the influence of a potential function U : R m ! R that encodes information about the environment such as the goal and obstacle locatio... |

566 | Voronoi Diagrams - A Survey of a Fundamental Geometric Data Structure - Aurenhammer - 1991 |

480 |
The Complexity of Robot Motion Planning
- Canny
- 1988
(Show Context)
Citation Context ...R (accessibility), 2. there exists a path from q goal 2 FS to some q 0 goal 2 R (departibility), and 3. there exists a path in R between q 0 start and q 0 goal (connectivity). Using a roadmap (Canny, =-=[6]-=-; Latombe, [28]), the planner can construct a path between any two points in a connected component of the robot's free space by first finding a collision free path onto the roadmap (accessibility), tr... |

301 | Exact Robot Navigation Using Artificial Potential Functions - Rimon, Koditschek - 1992 |

282 |
An algorithm for planning collision-free paths among polyhedral obstacles
- Lozano-Perez, Wesley
- 1979
(Show Context)
Citation Context ...bility). Three primary types of roadmaps have been developed: visibility graph, retraction, and silhouette. Visibility Graph. The nodes of the visibility graph (Latombe, [28]; LozanoPerez and Wesley, =-=[30]-=-) include the start location, the goal location, and all the vertices of the obstacles (or configuration space obstacles). The edges of the visibility graph are straight line segments that connect two... |

150 |
T.Weymouth, “Topological mapping for mobile robots using a combination of sonar and vision sensing
- Kortenkamp
- 1994
(Show Context)
Citation Context ...s method requires landmarks in constructing its map and is limited to the planar case. Another example of robotic exploration with realistic sensor assumptions can be found in Kortenkamp and Weymouth =-=[27]. In this -=-work, a map called a "topological map" of the world is formed. However, the algorithm in [27] assumes that the environment is populated with rectangular obstacles that meet at nearly ninety ... |

149 | Path-planning strategies for a point mobile automaton moving amidst unknown obstacles of arbitrary shape
- Lumelsky, Stepanov
- 1987
(Show Context)
Citation Context ...curate representation of the world. There are many non-heuristic algorithms for which provably correct solutions exist in the plane (see [35] for an overview). For example, Lumelsky's "bug" =-=algorithm [31]-=- is one of the first provably correct sensor based schemes to work in the plane. In this method, the robot travels along a straight line segment that connects the start and goal. If the robot encounte... |

104 |
Constructing force-closure grasps
- Nguyen
- 1988
(Show Context)
Citation Context ... ; m+ 1g if and only if 1. m of the m+ 1 vectors are linearly independent, and 2. a strictly positive sum of the m+1 vectors is the zero vector, i.e., P i=m+1 i=1si v i = 0 wheresi ? 0 for all i. See =-=[33, 23]-=- for proof. Lemma A.19 A set of n arbitrary vectors fv i 2 R m : i = 1; : : : ; ng positvely spans R m if and only if there exists a v i such that for all w 2 R m , hw; v i i ? 0. Proof: Let w = P n i... |

91 |
A 'retraction' method for planning the motion of a disc
- O'Dunlaing, Yap
- 1985
(Show Context)
Citation Context ...ility graph for the three obstacles represented as filled in polygons. The thick dotted line represents the shortest path between the start and goal. Retraction. A retraction ( ' O'D'unlaing and Yap, =-=[34]-=-) is a continuous mapping of the free space FS onto a one-dimensional subset of FS. The image of the retraction is the roadmap. A popular retraction in motion planning is termed the Voronoi diagram. T... |

85 |
Efficient algorithms for incremental distance computation
- Lin, Canny
- 1991
(Show Context)
Citation Context ...ance function which measures distance between a point and a convex polygon. The distance function used in the simulator is a modified version of Lin and Canny's closest feature algorithm described in =-=[29]-=-. Lin and Canny define a Voronoi region for each feature on the polygon. If a point is in the Voronoi region of a particular feature, then the distance to the polygon is the distance to the feature, w... |

79 | Robot navigation in unknown terrains: Introductory survey of non-heuristic algorithms
- Rao, Hareti, et al.
- 1993
(Show Context)
Citation Context ...ived increased attention lately, as it is requirement for realistic deployment of autonomous robots in unstructured environments. For a review of many sensor-based planning techniques, see Rao et al. =-=[35]-=-. Most current sensor based planning methods are heuristic algorithms which work very well under a variety of conditions, but there are no proofs of correctness that guarantee a path can be found. Mor... |

74 |
tensor analysis, and applications
- Manifolds
- 1988
(Show Context)
Citation Context ...unction. A point c is within line of sight of x if there exists a straight line segment which connects x and c without penetrating any obstacle. That is, c is within line of sight of x if for all t 2 =-=[0; 1]-=-, (x(1 \Gamma t) + ct) lies in FS. Let ~ C i (x) be the set of points on an object C i that are within line of sight of x, i.e., ~ C i (x) = fc 2 C i : x(1 \Gamma t) + ct 2 FS; 8t 2 [0; 1]g: If ~ C i ... |

68 | Convex partitions of a polyhedra: A lower bound and worst-case optimal algorithm - Chazelle - 1984 |

61 |
The kinematics of hyper-redundant robot locomotion
- Chirikjian, Burdick
- 1995
(Show Context)
Citation Context ...rections, described below. 6.2.1 Sensor Based Motion Planning for Set Robots The original problem that motivated this work was sensor based planning for highly articulator robots such as robot snakes =-=[11, 12, 14, 13, 37]-=-. Such robots can be represented as a point in a multi-dimensional configuration space. The analysis done in this work only applies to robots whose configuration spaces are diffeomorphic to R m and is... |

60 |
Elementary classical analysis
- Marsden, Hoffman
- 1993
(Show Context)
Citation Context ...ns and results which are useful to this work. Although the following definitions are standard to point set topology, we used the definitions from Marsden's book entitled Elementary Classical Analysis =-=[32]-=-. A.1.1 Topology Notation and Definitions Definition A.1 (ffl-Neighborhood) ([32], p. 32) An ffl-neighborhood about a point x 2 R m is the set fy 2 R m : kx \Gamma yk ! fflg. Definition A.2 (Open Sets... |

55 | Sensor based planning, part I: The generalized Voronoi graph
- Choset, Burdick
- 1995
(Show Context)
Citation Context ...d i 2 \Gamma d j 2 ) ' + fi ` r(d i 3 \Gamma d j 3 ) " ; w AE = 0; ) ��" r(d i 2 \Gamma d j 2 ) ' + fi ff ` r(d i 3 \Gamma d j 3 ) " ; w AE = 0: Since by Equidistant Surface Transve=-=rsality Assumption [16], for al-=-l w 2 T x S i 1 j 1 : ��` r(d i 2 \Gamma d j 2 )(x) ' ; w AE 6= 0 ��` r(d i 3 \Gamma d j 3 )(x) ' ; w AE 6= 0; we conclude that r(d i 2 \Gamma d j 2 ) = fi ff ` r(d i 3 \Gamma d j 3 ) ' . Howe... |

51 | An opportunistic global path planner
- Canny, Lin
- 1990
(Show Context)
Citation Context ...rough the critical points. The recursion terminates when the slice is one-dimensional. The union of the silhouette curves is a connected roadmap. The Opportunistic Path Planner (OPP) of Canny and Lin =-=[8]-=-, [9] is an example of a silhouette method in which local maxima of a continuous and differentiable potential function, restricted to a slice, are traced out as a slice sweeps through the work space o... |

47 |
A modal approach to hyperredundant manipulator kinematics
- Chirikjian, Burdick
- 1994
(Show Context)
Citation Context ...rections, described below. 6.2.1 Sensor Based Motion Planning for Set Robots The original problem that motivated this work was sensor based planning for highly articulator robots such as robot snakes =-=[11, 12, 14, 13, 37]-=-. Such robots can be represented as a point in a multi-dimensional configuration space. The analysis done in this work only applies to robots whose configuration spaces are diffeomorphic to R m and is... |

46 | Sensor based planning, part II: Incremental contruction of the generalized Voronoi graph
- Choset, Burdick
- 1995
(Show Context)
Citation Context ... Distance Measurement Sensor Model. point from the sensor, without knowing from which obstacle the nearest point came. Therefore, the incremental construction of the GVG described in Chapter 4 and in =-=[17]-=- has to be adapted for sensor based implementation on actual robots. In order to do this, we recall the raw distance function which was defined in Section 4.4. 5.3.1 Raw Distance Function A key featur... |

41 |
On the complexity of d-dimensional voronoi diagrams
- Klee
- 1980
(Show Context)
Citation Context ...: : i m+1 g. The number of m combinations of m+ 1 elements is m+ 1 ( \Gamma m+1 m \Delta = m+ 1). \Xi This result has been previously determined for the special case where each object is a point site =-=[26]-=-. 49 2.3.4 The Multi-object Distance Function is Smooth on a k-Equidistant Face The multi-object distance function is nominally not smooth. However, when constrained to a k-equidistant surjective surf... |

32 | Simplified Voronoi diagrams
- Canny, Donald
- 1988
(Show Context)
Citation Context ...at is, it is the set of points in m dimensions equidistant to m convex obstacles. Though the GVG introduced in this work appears to be new, a disconnected GVG-like structure for SE(3) is described in =-=[7]-=-. The GVG, described in Chapter 2, can be thought of as the natural extension of the GVD into higher dimensions. However, unlike the GVD, the GVG is in general not a roadmap because it is not necessar... |

32 | ALFA: A Language for Programming Reactive Robot Control Systems, Robotics and Automation - Gat - 1991 |

29 |
L<Kinematically Optimal Hyper-redundant Manipulator Configuratiox
- Chirikjian, Burdick
- 1992
(Show Context)
Citation Context ...rections, described below. 6.2.1 Sensor Based Motion Planning for Set Robots The original problem that motivated this work was sensor based planning for highly articulator robots such as robot snakes =-=[11, 12, 14, 13, 37]-=-. Such robots can be represented as a point in a multi-dimensional configuration space. The analysis done in this work only applies to robots whose configuration spaces are diffeomorphic to R m and is... |

29 |
A “retraction” method for learned navigation in unknown terrains for a circular robot
- Rao, Stoltzfus, et al.
- 1991
(Show Context)
Citation Context ...for a disk in the plane. However, the method in [34] requires full knowledge of the world's geometry prior to the planning event, and its retract methodology may not extend to non-planar problems. In =-=[36]-=-, an incremental approach to create a Voronoi diagram-like structure, which is limited to the case of a plane, was introduced. Prior work (e.g., Avis and Bhattacharya [3]) describes the Voronoi graph,... |

27 |
Sensor based planning and nonsmooth analysis
- Choset, Burdick
- 1994
(Show Context)
Citation Context ...xed stepping direction. However, gradient ascent can be a computationally expensive procedure because of its slow convergence. Also, the constant step direction leads to undesirable roadmap artifacts =-=[15]-=-. Our approach borrows some basic ideas and techniques from numerical continuation methods [24]. Continuation methods are used to trace the roots of the expression G 1 (y; ) = 0 as the parametersis va... |

25 |
Robot navigation by conditional sequencing
- Gat, Dorais
- 1994
(Show Context)
Citation Context ...le set of behaviors such as following a wall [5]. A hierarchy of cooperating behaviors forms more complicated behaviors such as exploration. An extension of this type of approach is called sequencing =-=[20, 21, 22]-=-. Since there are strong experimental results indicating the utility of these approaches (such as [22]), some of these algorithms may provide a future basis for provably correct sensor based planners.... |

20 |
Algorithms for computing d-dimensional Voronoi diagrams and their duals
- Avis, Bhattacharya
- 1983
(Show Context)
Citation Context ...d to non-planar problems. In [36], an incremental approach to create a Voronoi diagram-like structure, which is limited to the case of a plane, was introduced. Prior work (e.g., Avis and Bhattacharya =-=[3]-=-) describes the Voronoi graph, an extension of the Voronoi diagram extended into higher dimensions. The Voronoi graph is the locus of points in m dimensions equidistant to m point sites. The generaliz... |

17 |
Integrating Reaction and Planning in a Heterogeneous Asynchronous Architecture for Controlling Real World Mobile Robots
- Gat
- 1992
(Show Context)
Citation Context ...le set of behaviors such as following a wall [5]. A hierarchy of cooperating behaviors forms more complicated behaviors such as exploration. An extension of this type of approach is called sequencing =-=[20, 21, 22]-=-. Since there are strong experimental results indicating the utility of these approaches (such as [22]), some of these algorithms may provide a future basis for provably correct sensor based planners.... |

15 | Sensor based planning for a planar rod robot, in
- Choset, Konuksven, et al.
- 1996
(Show Context)
Citation Context ... to the obstacle and the nearest point on the obstacle to the rod. roadmap is termed the rod hierarchical generalized Voronoi graph (rod-HGVG), the planar version of which has already been defined in =-=[18]-=-. The rod-HGVG is defined in terms of the rod single object distance function which is the shortest distance between a rod R at configuration q and a convex obstacle C i . See Figure 6.2. The rod dist... |

13 |
Construction of C-space roadmaps using local sensory data—What should the sensors look for
- Rimon, Canny
- 1994
(Show Context)
Citation Context ...he complete classical planning methods. An example of a complete roadmap scheme is Canny and Lin's Opportunistic Path Planner (OPP) [9]. Rimon adapted this motion planning scheme for sensor based use =-=[38]-=-. To our knowledge, this is the first sensor based motion planning scheme for workspaces whose dimension is greater than two. However, connectivity of the roadmap in [38] cannot be guaranteed without ... |

13 |
Representing the spatial experience and solving spatial problems in a simulated robot environment
- Rowat
- 1979
(Show Context)
Citation Context ...ii Feodosevich Voronoi who lived from 1868 until 1908, the Voronoi diagram was first studied by DesCartes in 1644. The Voronoi diagram was first applied to robotics in Rowat's work on computer vision =-=[40]-=-. In this and other works, the sites were sets, instead of points. In such cases, in keeping with the Voronoi diagram literature, we term this structure the generalized Voronoi diagram in order to dis... |

11 |
Theory and applications of hyper-redundant robotic manipulators
- Chirikjian
- 1992
(Show Context)
Citation Context |

9 |
Motion planning with uncertainty for highly redundant kinematic structures I
- Reznik, Lumelsky
- 1992
(Show Context)
Citation Context |

8 |
Real-time Onstacle Avoidance for Fast Mobile Robots in Cluttered Environments
- Borenstein, Koren
- 1990
(Show Context)
Citation Context ...se algorithms may provide a future basis for provably correct sensor based planners. Another type of heuristic approach involves the discretization of a planar world into pixels (Borenstein and Koren =-=[4]-=-). In this approach, each pixel is assigned a value indicating the likelihood that it overlaps an obstacle. This method lends itself very nicely to implementation with real sensors, but discretizing t... |

6 |
Vision-Basied Motion Planning and Exploration Algorithms for Mobile Robots
- Taylor, Kriegman
- 1994
(Show Context)
Citation Context ...th does not reflect the topology of the free space (the region of the environment not occupied by obstacles) and thus, it cannot be used to guide future robot excursions. Taylor and Kriegman's method =-=[43]-=- has a provably correct solution, uses realistic sensor assumptions, and need not require prior knowledge of the goal's location. In this method, the robot forms a graph of a bounded free space by cir... |

6 |
Simpli ed voronoi diagrams
- Canny, Donald
- 1988
(Show Context)
Citation Context ...hat is, it is the set of points in m dimensions equidistant tom convex obstacles. Though the GVG introduced in this work appears to be new, a disconnected GVG-like structure for SE(3) is described in =-=[3]-=-. The GVG can be thought of as the natural extension of the GVD into higher dimensions. However, unlike the GVD, the GVG is not necessarily connected in dimensions greater than two, and thus, in gener... |

5 |
Descent Algorithm for a Class of Convex Nondifferentiable Functions
- Scheimberg, Oliveira
- 1992
(Show Context)
Citation Context ...and only if fv i 2 R m : i = 1; : : : ; ng positively span R m . Lemma A.22 [Goldman and Tucker] It requires a minimum of (m + 1) vectors to positively span R m . The results of Scheimber and Olivera =-=[41] state tha-=-t the second derivative of D, termed a generalized hesian, is always positive or contains only positive values. In a sense, the graph of the function D is always "concave up." Therefore, the... |

1 |
Linear Inequalities and Related Systems, chapter Polyhedral Convex Cones
- Golman, Tucker
- 1956
(Show Context)
Citation Context ... ; m+ 1g if and only if 1. m of the m+ 1 vectors are linearly independent, and 2. a strictly positive sum of the m+1 vectors is the zero vector, i.e., P i=m+1 i=1si v i = 0 wheresi ? 0 for all i. See =-=[33, 23]-=- for proof. Lemma A.19 A set of n arbitrary vectors fv i 2 R m : i = 1; : : : ; ng positvely spans R m if and only if there exists a v i such that for all w 2 R m , hw; v i i ? 0. Proof: Let w = P n i... |