Fennema, Claude L. "Interweaving Reason, Action and Perception" Technical Report, COINS Department, University of Massachusetts, Amherst MA, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....locales is connected. When the free space in a locale is not connected, the network can be repaired by splitting the locale into sublocales and performing some minor network reorganization. This is the process referred to as restructuring in Figure 10. Details about restructuring can be found in [15]. Once restructuring is complete, planning with A invoked using the new scope locale. Each time A fails to find a solution sketch a plan continues to work its way up the contained by hierarchy until it either finds a solution or exhausts the possibilities. In the latter case it stops and ....

....2.4.5) The first three of these steps are identical for the two servoing operations. The difference between action level and planlevel servoing lies in what is done with the results of the matching operation. A detailed discussion of the complexity for steps 1, 2, and 3 can be found in [13] and [15] but a summary of the results is as follows: 1. The complexity of the landmark selection process is a function of the local complexity of the environment, as measured by the number of faces m and vertices n associated with the locale containing the agent and that locale s offspring. The total ....

[Article contains additional citation context not shown here]

Fennema, Claude L. "Interweaving Reason, Action and Perception" Technical Report, COINS Department, University of Massachusetts, Amherst MA, 1994.

....the Mount Holyoke mobile robot Susan B. b) and the UMass HUMMWV (c) All are equipped with a black and white TV cameras. 4 System (Waxman, LeMoigne, Davis et al. 37] AuRA (Arkin [2] and the Martin Marietta ALV (Turk, Morgenthaler, Gremban, and Marra [36] A close analysis (Fennema [13]) of these systems shows each of them is an instance of the architecture depicted in Figure 4 (a) and that information flows in these systems as shown in Figure 4 (b) It is characteristic of these systems that each invocation of a module makes a large contribution to the ongoing intellectual ....

....2. An RML Implementation The first detailed design of an RML system appeared during the fall of 1988 in Fennema, Riseman and Hanson [10] and early experiments have been described in Fennema and Hanson [11] 12] A description of the first complete implementation can be found in Fennema [13]. This section summarizes the key elements of that system. The implementation is built around a central model that contains information about the environment, the robot s capabilities, and the task to be accomplished. The control structure follows the ideas illustrated in Figure 1. Beginning with ....

[Article contains additional citation context not shown here]

Fennema, Claude L. "Interweaving Reason, Action and Perception." PhD. Dissertation, University of Massachusetts, Amherst, September 1991.

....error in motion. In the latter case, this size limits the amount of image motion that can occur from frame to frame. The basic algorithm for matching templates to image patches is a combination of normalized crosscorrelation and steerable filters. The normalized cross correlation measure [7] assumes that the surfaces in the environment are Lambertian, that they can be locally approximated by a plane, and that the illumination incident on the surfaces can be locally approximated by a constant. Under these assumptions the correlation measure is normalized in that it is independent of ....

....can be locally approximated by a constant. Under these assumptions the correlation measure is normalized in that it is independent of the illumination incident on the surface. However, good experimental results have been obtained with this measure on surfaces that are only weakly Lambertian (see [7] for a derivation and [8] for experiments with this measure) However, normalized cross correlation degrades when there is a relative rotation between the templates and image patches [19] To compensate for 2D rotations it is sufficient to note that linear shift invariance in polar space is ....

Fennema, C. L., "Interweaving Reason, Action and Perception", COINS TR91-56, Dept. of Computer Science, Univ. of Massachusetts, Amherst, 1991.

....attempts to preserve these model image correspondences over time. Thus, visual feature tracking can be used to close a feedback loop around the pose of an object. This enables the construction of controllers that servo a manipulator or mobile robot to a desired pose or to trace a desired trajectory[3, 5]. In this paper we describe a new tracking technique and it s application to visually servo a manipulator with poor kinematics in a peg in hole insertion task described below. The kinematics are poor in the sense that the accuracy of spatial resolution provided by the kinematics is coarser than ....

....corners can be inferred by computing the intersections of the appropriate edges. Rotation compensated normalized correlation is used to localize the templates in appropriate search windows around their hypothesized locations and is discussed below. 3. Template Localization Normalized correlation [3] is a well known match measure and a search for the maximum correlation score of a template over a search window is typically use to localize an image patch in the window. However, this scheme is sensitive to rotations. In order to compensate for rotations, edges and their orientations are The ....

C. L. Fennema, "Interweaving Reason, Action and Perception", COINS TR91-56, Dept. of Computer Science, Univ. of Massachusetts, Amherst, 1991.

.... functions see [Connolly93] and for a thorough introduction to path planning see [Latombe91] For a survey of architectures for mobile robot navigation, including planning and reactivity see [Ravela92] The kernel of the tracking system described in this paper is based on normalized correlation [Fennema91] in polar space. For studies of correlation and SSD see [Wood83]and [Anandan89] respectively. Tracking has been approached from a tokenbased view point and [Deriche91, Koller93] where the tokens are line segments or groups of segments. Direct methods for contour tracking were employed in [Kass88] ....

....to compensate for large instantaneous rotations. An excellent study of visual servoing can be found in [Hashimoto94] which primarily covers a manipulator, hand eye environment. Visual servoing has been used on mobile robots in structured environments [Zhang93] and in outdoor environments [Fennema91, Dickmanns88] 3 Stealth Planning Stealth paths which avoid terrain obstacles and visible to target locations are planned over a region Omega of the terrain map using harmonic functions. Terrain obstacles are identified as areas in Omega which are not traversable by the vehicle and can include steep slopes, ....

[Article contains additional citation context not shown here]

Fennema, C. L., "Interweaving Reason, Action and Perception", COINS TR91-56, Dept. of Computer Science, Univ. of Massachusetts, Amherst, 1991.

....is normalized in that it is independent of the illumination incident on the surface. However, good experimental results have been obtained with this measure on surfaces that do not Figure 2: Comparison of NCC and NCC R algorithms under 2D feature rotation fit the Lambertian assumption(see [6] for a derivation and [7] for experiments with this measure) Normalized cross correlation degrades when there is a relative rotation between the templates and image patches. To compensate for 2D rotations it is sufficient to note that equation 1 is linear shift invariant in cartesian space and ....

Fennema, C. L., "Interweaving Reason, Action and Perception", COINS TR91-56, Dept. of Computer Science, Univ. of Massachusetts, Amherst, 1991.

....approximated by a constant. Under these assumptions the correlation measure is normalized in that it is independent of the illumination incident on the surface. However, good experimental results have been obtained with this measure on surfaces that do not fit the Lambertian assumption(see [Fennema 91] for a derivation. Normalized cross correlation degrades when there is a relative rotation between the templates and image patches. To compensate for 2D rotations it is sufficient to note that equation 1 is linear shift invariant in cartesian space and hence is translation invariant. ....

Fennema, C. L., "Interweaving Reason, Action and Perception", COINS TR91-56, Dept. of Computer Science, Univ. of Massachusetts, Amherst, 1991.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC