Geir Hovland and Brenan J. McCarragher. Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research, 17(2):153--168, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the contact is stable again, an estimator is constructed for each probable CF and initialized with the current state estimate and uncertainty. The real CF (real lter) is determined by comparing the probability of the pose, twist and wrench measurements for each of the lters (See e.g. 1] 16] [17]) Once the probability of one of the lters exceeds the other probabilities by a user de ned factor, its corresponding CF, its geometrical parameter estimates and their uncertainty are selected for further estimation. D. Propagation of estimates through sequences of contact formations During ....

G. Hovland and B. J. McCarragher, \Hidden Markov models as a process monitor in robotic assembly," The International Journal of Robotics Research, vol. 17, no. 2, pp. 153-168, February 1998.

....into account at modeling time: learned models can handle deviations around nominal parameter values for which learning data was available at training time. A deterministic, Bayesian, Fuzzy [19] or Neural Network model [1] performs the CF recognition or transition detection. Hidden Markov Models [12, 10] are a popular tool for both the detection of transitions and the recognition of the current cf. Simultaneous CF recognition and geometrical parameter estimation. The di#erent CF models are a function of the uncertain parameters. During task execution, the uncertainty reduces due to the parameter ....

G. E. Hovland and Brenan J. McCarragher. Hidden markov models as a process monitor in robotic assembly. Int. Jnl. of Rob. Res., 17(2):153--168, Feb 1998.

....2. the robot does not know the exact geometrical parameters of the contacting parts, this is a continuous type of uncertainty. Based on position, velocity and or force measurements, different research groups use different sensor processing techniques to deal with these uncertainties: e.g. [1] presents a stochastic filter based on a trained Hidden Markov Model to recognize contact transitions; 2] recognizes CFs with a trained fuzzy or neural network classifier; 3, 4] recognize CFs with deterministic classifiers based on polyhedral convex cones; 5, 6] describe a stochastic estimator ....

G. Hovland and B. J. McCarragher, "Hidden Markov models as a process monitor in robotic assembly," The International Journal of Robotics Research, vol. 17, no. 2, pp. 153--168, February 1998.

....error recovering Adaptation, monitoring Task specification Control Robot, environment Figure 1: The classical three level hierarchical control pattern for intelligent robot systems. time, that component monitors the estimated parameter values, to detect when to switch to the next subtask [7, 8]. 3. The highest ( deliberative ) level is responsible for decision making, i.e. re)planning and on line error recovery. Making intelligent decisions depends heavily on being able to interpret al..l sensor data (raw data, as well as processed data from the two lower levels) within the context of ....

....in a given CF , and matching of the sensor data to a set of possible models and recognize the CF that fits best. What has already been achieved The most successful and advanced approaches rely on tools from Bayesian probability theory, e.g. the Kalman Filter [7] or Hidden Markov Models [8]. Bayesian probability is particularly suited for these tasks, because it is very straightforward to incorporate a priori knowledge (such as models) into the sensor processing, even for one of tasks. Figure 3: Robot placing a cube in a corner. What is the future The medium term achievement that ....

G. Hovland and B. J. McCarragher, "Hidden Markov models as a process monitor in robotic assembly," Int. J. Robotics Research, vol. 17, no. 2, pp. 153--168, 1998.

....parameters. So, in order to compensate for the uncertainty and variation in these parameters, the controller relies on on line estimates from the estimation component (Sect. 4) at the same time, that component monitors the estimated parameter values, to detect when to switch to the next subtask [7, 8]. 3. The highest ( deliberative ) level is responsible for decision making, i.e. re)planning and on line error recovery. Making intelligent decisions depends heavily on being able to interpret al..l sensor data (raw data, as well as processed data from the two lower levels) within the context of ....

....in a given ##, and matching of the sensor data to a set of possible models and recognize the ## that fits best. What has already been achieved The most successful and advanced approaches rely on tools from Bayesian probability theory, e.g. the Kalman Filter [7] or Hidden Markov Models [8]. Bayesian probability is particularly suited for these tasks, because it is very straightforward to incorporate a priori knowledge (such as models) into the sensor processing, even for one of tasks. Figure 3: Robot placing a cube in a corner. What is the future The medium term achievement that ....

G. Hovland and B. J. McCarragher, "Hidden Markov models as a process monitor in robotic assembly," Int. J. Robotics Research, vol. 17, no. 2, pp. 153--168, 1998.

....contact is stable again, an estimator is constructed for each probable CF and initialized with the current state estimate and uncertainty. The real CF (real filter) is determined by comparing the probability of the pose, twist and wrench measurements for each of the filters (See e.g. 1] 16] [17]) Once the probability of one of the filters exceeds the other probabilities by a user defined factor, its corresponding CF, its geometrical parameter estimates and their uncertainty are selected for further estimation. D. Propagation of estimates through sequences of contact formations During ....

G. Hovland and B. J. McCarragher, "Hidden Markov models as a process monitor in robotic assembly," The International Journal of Robotics Research, vol. 17, no. 2, pp. 153--168, February 1998.

....requires very structured and hence expensive setups if one requires absolute certainty about the task sequence. Loosening the requirements for the set up introduces increased uncertainty about the sequence of contact situations. This problem has been tackled using a Hidden Markov Model approach, [8, 12, 18]: the di#erent contact situations correspond to states in the network (each with a typical set of force motion 0 100 100 10 100 0 1.0 Figure 5: Example of a template to be used in a Monte Carlo localization algorithm for finding the planar top of a cylindrical barrel, with a ....

G. Hovland and B. J. McCarragher. Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research, 17(2):153--168, 1998.

....identi cation can be learned. What has to be learned is: 1) what sensor signals belong to each of the possible contact situations (and vice versa) and 2) when does the robot move from one contact type to another contact type (both contact types have, in general, di erent topology and geometry) Hovland and McCarragher (1998) and Eberman (1997) use Hidden Markov Models, Rabiner 1989) Nuttin and Van Brussel (1996) and Nuttin (1998) apply neural networks. Finally, at the highest controller level, the estimated contact situation is used for 1) error recovery and on line replanning, and or 2) the creation of the models ....

Hovland, G. and B. J. McCarragher (1998). Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research 17 (2), 153-168.

....Class of Single Ended Contact Formations to precalculate the templates. Recent refinements use linear discriminant functions instead of thresholds [9] Hovland and McCarragher proposed FFTs for capturing the dynamics of contact changes and Hidden Markov Models to model the resultant information [10]. While they achieved a 97 success rate, they reported times of 0.5 0.6 seconds. Also, significant training is required for eachcontact transition. Hara and Yokogawa [11] used fuzzy sets to recognize CFs# however, only a small number of CFs were considered and no general method was shown. ....

G.E. Hovland and B.J. McCarragher, "Hidden markov models as a process monitor in robotic assembly," International Journal of Robotics Research,vol. 17, no. 2, pp. 153--168, Feb. 1998.

....estimation. I. Introduction H IDDEN Markov Models (HMMs) also known as probabilistic functions of Markov chains [1] have a most successful tradition in the eld of speech recognition [2] 3] Recently, however, applications in new elds, which range from communications [4] and control problems [5] to image sequence classi cation [6] and data analysis in biology [7] 8] 9] are explored with convincing results. Especially for these new applications models often have to be inferred from very long data sets, produced e.g. by on going processes. In these cases it is desirable to update the ....

G.E. Hovland, B.J. McCarragher, " Hidden Markov Models as a Process Monitor in Robotic Assembly," Int. J. Robotics Research, vol. 17, pp. 153-168, 1998.

....requires very structured and hence expensive setups if one requires absolute certainty about the task sequence. Loosening the requirements for the set up introduces increased uncertainty about the sequence of contact situations. This problem has been tackled using a Hidden Markov Model approach, [8, 12, 18]: the di erent contact situations correspond to states in the network (each with a typical set of force motion 0 100 100 10 100 0 1.0 Figure 5: Example of a template to be used in a Monte Carlo localization algorithm for nding the planar top of a cylindrical barrel, with a Gaussian like ....

G. Hovland and B. J. McCarragher. Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research, 17(2):153-168, 1998.

....involved, helping to decide on line among conflicting alternatives. This was a research subject for one of the authors in the past [10] and will be applied to this work. The detection of events signaling environment state changes has been pursued by other researchers in recent years (e.g. [8, 9]) and will also be the subject of further research by our group, due to its importance at the organizational level. ....

G. E. Hovland and B. J. McCarragher. Hidden Markov Models as a Process Monitor in Robotic Assembly. The International Journal of Robotics Research, 17(1), 1998.

....of edge edge and surface vertex contacts. The process monitor is defined as the map b: b (x(t) 4) where is the k th discrete state of the system. The focus of this paper is on the development of control strategies and so we will assume the existence of a perfect process monitor [3]. The discrete event controller is the part of the system which determines the appropriate command to issue, based upon the discrete state of the system. For the purposes of this paper, the control command u(t) will consist of a force and a velocity command u(t) v F T ] where v is the ....

G. Hovland and B. J. McCarragher. Hidden markov models as a process monitor in robotic assembly. Intl. Journal of Robotics Research, 17(2):153-168, January 1998.

....the discrete event with best match is chosen. 1. Training Each individual discrete event k is described by a HMM Ak. The model parameters of A have to be estimated from several training examples of the force signals. The parameters are estimated using the Baum Welch re estimation formula, see [16] for the details. The formula is an iterative solution and at each iteration the probability of the HMM matching the force signals in the training set is increased. The training is performed only once and off line. Evaluation When a discrete event occurs, every HMM corresponding to an admissible ....

G.E. Hovland and B.J. McCarragher, "Hidden Markov Models as a Process Monitor in Robotic As- sembly", The International Journal of Robotics Research, to be published.

....implemented for the real time control of a planar robotic assembly task in a discrete event control framework. One of the monitoring methods used is based on Hidden Markov Models, where the average recognition rate was 87 . Larger recognition rates for the HMM method have been demonstrated by [4]. The rate of 87 was chosen to show the effectiveness of the dynamic sensor selection method. The experiments show that the method performs better than any individual process monitor. A successful event recognition rate of 97 with an average CPU time of 0:38 seconds is achieved when two force ....

....Monitor 3 uses force measurements also, but is fundamentally different from the other two methods. Instead of using only one set of measurements, the forces are logged for a short time period. Process monitor 3 uses Hidden Markov Models (HMMs) and the Fourier transform of the logged force signals, [4]. Each discrete event is modeled by a HMM, where the HMM parameters are estimated from training sets. In realtime operation the HMM which most nearly matches the measured forces F x , F y and M z is chosen. The confidence level is calculated by comparing the probabilities of each HMM given the ....

G.E. Hovland and B.J. McCarragher, Hidden Markov Models as a Process Monitor in Robotic Assembly. International Journal of Robotics Research, to be published.

....Information from different sensors are combined into one single outcome, with (hopefully) better resolution and reliability than could be produced by the single sensors individually. 4. Classification, also known as (template pattern) matching, labelling or event object pattern recognition, [10, 22, 21, 20, 35, 49, 74, 83, 79]. A (spatial or temporal) sequence of sensor signals is compared to logs of previous executions, or to models of the task, in order to find the best correspondence. 5. Sequence planning, or decision making, 2, 1, 14, 17] If the task controller has the option to choose between different ....

....with the best match is chosen. 1. Training Each individual discrete transition k is described by a HMM k . The model parameters of k have to be estimated from several training examples of the force signals. The parameters are estimated using the Baum Welch reestimation formula, see [49] for the details. The formula is an iterative solution and at each iteration the probability of the HMM matching the force signals in the training set is increased. The training is performed only once and off line. Real Time Operation Evaluation of Admissible Events Recognition Based on Model ....

G. Hovland and B. J. McCarragher. Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research, 17, 1998.

....discrete event with best match is chosen. 1. Training Each individual discrete event k is described by a HMM k . The model parameters of k have to be estimated from several training examples of the force signals. The parameters are estimated using the Baum Welch re estimation formula, see [16] for the details. The formula is an iterative solution and at each iteration the probability of the HMM matching the force signals in the training set is increased. The training is performed only once and off line. 2. Evaluation When a discrete event occurs, every HMM corresponding to an ....

G.E. Hovland and B.J. McCarragher, "Hidden Markov Models as a Process Monitor in Robotic Assembly ", The International Journal of Robotics Research, to be published.

No context found.

Geir Hovland and Brenan J. McCarragher. Hidden Markov models as a process monitor in robotic assembly. Int. J. Robotics Research, 17(2):153--168, 1998.

No context found.

G. Hovland and B. J. McCarragher, \Hidden Markov models as a process monitor in robotic assembly," The International Journal of Robotics Research, vol. 17, no. 2, pp. 153-168, February 1998.

No context found.

G. E. Hovland and Brenan J. McCarragher. Hidden markov models as a process monitor in robotic assembly. Int. Jnl. of Rob. Res., 17(2):153--168, Feb 1998.

No context found.

G. E. Hovland and Brenan J. McCarragher. Hidden markov models as a process monitor in robotic assembly. Int. Jnl. of Rob. Res., 17(2):153--168, Feb 1998.

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