G. Hovland and B. McCarragher, "Frequency-domain force measurements for discrete event contact recognition," In Proceedings of the 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Linear discriminant functions are then determined from the classified data. As an alternative to using geometric information, Hovland and McCarragher proposed using fast Fourier transforms to capture the dynamic nature of a contact change and Hidden Markov Models to model the information [29]. Although they achieved a high success rate (98 ) this approach is not easily accomplished in real time. Another method which does not require geometric information was demonstrated by Hara and Yokogawa [23] Fuzzy sets were used to model contact forces and recognize the contact formation. ....

.... Another system based on the discrete events (referenced here as the hybrid dynamic model) is that proposed by Hovland, Sikka, and McCarragher in [30] They use a discrete event controller and assume the existence of a process monitor that detects and identifies changes in contact (such as that in [29]) This reference focuses on the control part of the skill, using a Hidden Markov Model to model the human s motions. The Markov states correspond directly to the discrete states (i.e. contact formations) The goal is to learn the velocity command which allows the robot to achieve the next ....

G. Hovland and B. McCarragher, "Frequency-domain force measurements for discrete event contact recognition," In Proceedings of the 1996.

....configuration of the workpiece is explicitly determined from the force measurement and a correction of motion is generated based on the estimated contact configuration. To better detect discrete contact transitions, dynamic models of the process have been used for analyzing measured forces [11] [13]. Although these explicit force guided controls are a useful concept, particularly for the monitoring of assembly processes, there are inherent difficulties in applying it to real world problems. Force and moment measured by a force sensor are quite noisy and erratic due to friction at the ....

G. E. Hovland and B. J. McCarragher, "Frequency-domain force measurements for discrete event contact recognition," in Proc. IEEE Int. Conf. Robot. Automat., 1996.

....from CAD models. As such, most methods of identifying contact formations use detailed geometric models of the assembly parts, combined with position information and data from a wrist force sensor (e.g. 6] One notable exception is the sensor based method proposed by Hovland and McCarragher [8]. Another is the method proposed by Cervera et al. [1] which also clusters force sensor signals to identify state information. Our work differs in the clustering mechanism and also includes a critical preprocessing step which minimizes sensor ambiguities. In our previous work [12, 9] we ....

G. Hovland and B. McCarragher, "Frequencydomain force measurements for discrete event contact recognition," In Proceedings of the 1996 IEEE International Conference on Robotics and Automation, volume 2, pp. 1166--1171, Minneapolis, MN, April 1996.

....find a method of identifying the contact formation (CF) from the force signals only, without using geometric models of the workpieces. Several approaches have been tried. McCarragher has used the discrete event approach to assembly, incorporating system dynamics and qualitative reasoning [7] In [6], FFT s are used to identify contact states; however, this approach cannot easily be run in realtime. Hara and Yokogawa have used fuzzy logic to recognize the CF from force signals; however, only a small number of states for one case was considered [4] Asada proposed the use of a neural network ....

G.E. Hovland and B.J. McCarragher. Frequencydomain force measurements for discrete event contact recognition. In Proceedings of the 1996 IEEE International Conference on Robotics and Automation, volume 2, pages 1166--1171, Minneapolis, MN, April 1996.

....to identify the discrete contact states required a geometric model of the workpieces [6] a burden we would like to avoid in robot programming by demonstration. McCarragher et al. have used the discrete event approach to assembly, incorporating system dynamics and qualitative reasoning [11] In [8], FFT s are used to identify contact states; however, this approach cannot easily be run in real time. In [9] Hidden Markov Models are used to model skills, which include the contact states and discretized velocity commands representing the transitions between contact states. This provides a ....

G.E. Hovland and B.J. McCarragher. Frequencydomain force measurements for discrete event contact recognition. In Proceedings of the 1996 IEEE International Conference on Robotics and Automation, volume 2, pages 1166--1171, Minneapolis, MN, April 1996.

....characteristics. Observations for improving the convergence and error recovery characteristics are given. As the simulations and experiments demonstrate, a highly successful, hybrid dynamic adaptation system has been presented. Moreover, this adaptation ability along with the state monitoring of [4], and the error detection and recovery of [8] form a highly effective, intelligent motion 14 control system. ....

G. E. Hovland and B. J. McCarragher. Frequency-domain force measurements for discrete event contact recog- nition. In IEEE Intl. Conf. on Robotics and Automation, pages 1166-1171, April 1996.

....controller. The second aspect of a skill is the process monitor, which monitors the plant variables and provides the discrete events at the discrete level. Any model of skill must take these two aspects into account. An HMM based process monitor using the observed force signal is described in [11]. Each event is modeled as an HMM. The sensory data corresponding to an event is scored by all the HMMs and the recognized event corresponds to the HMM that predicts the sensory data with maximum probability. In this paper, we use the process monitor described in [11] and focus on the ....

....force signal is described in [11] Each event is modeled as an HMM. The sensory data corresponding to an event is scored by all the HMMs and the recognized event corresponds to the HMM that predicts the sensory data with maximum probability. In this paper, we use the process monitor described in [11], and focus on the representation of the discrete event controller using an HMM. An HMM can be used naturally to represent a discrete event controller for an assembly skill. An HMM consists of two stochastic processes. The underlying stochastic process is a Markov process represented by the ....

G. Hovland and B. J. McCarragher, Frequency-Domain Force Measurements for Discrete Event Contact Recognition, submitted to the 1996 IEEE International Conference on Robotics and Automation.

....and M z are the planar force torque measurements. P x , P y and ff are the planar position measurements. The origin of the x 0 , y 0 axes is located at the base of the manipulator. A significant amount of information can be extracted from the force signals by using frequency domain measurements [3, 5]. In section 3.1 we describe the transform from time domain to frequency domain on the force measurements. In section 3.2 the planar position measurements P x , P y and ff are used to calculate distance functions between selected surfaces and edges of the workpiece and the environment. When the ....

G.E. Hovland and B.J. McCarragher, FrequencyDomain Force Measurements for Discrete Event Contact Recognition. Proc. of the 1996 IEEE Int. Conf. on Robotics and Automation, Minneapolis, pp. 1166-1171.

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