M. Kawato, "Computational schemes and neural network models for formation and control of multijoint arm trajectory", in Neural Networks for Control, Sutton Miller and Werbos, Eds. MIT Press, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....y # [t] is the desired response) through the forward model. In their work, when they deal with plants that are required to follow a particular trajectory, they use a modified algorithm, equivalent to that of backpropagation through time [17] 18] A similar approach was taken by Kawato et. al [19], 20] 21] and D. Psaltis [22] Generally however, it is not always possible to reach a desired next state, from any other dynamic system state, and therefore their methods are not generally applicable to control problems. In particular such an approach cannot be applied directly (although it ....

M. Kawato, "Computational schemes and neural network models for formation and control of multijoint arm trajectory", in Neural Networks for Control, Sutton Miller and Werbos, Eds. MIT Press, 1990.

....of parameters within the feedback loop in a closed loop control scheme. FEL is a feed forward neural network structure which, when training, learns the inverse dynamics of the controlled plant (Fig. 1) This method is based on contemporary physiological studies of the human cortex [25] [26]. The total control effort applied to the plant is the sum of the feedback control output and network control output. The ideal configuration of the neural network would correspond to the inverse mathematical model of the system s plant. The network is given the information of the desired ....

M. Kawato "Computational schemes and neural network models for formation and control of multi-joint arm trajectory," , W. T. Miller, R. S. Sutton, and P. J. Werbos, in Neural Network for Control. Cambridge, MA: MIT Press, 1990.

....[36] shows that the back propagation process actually computes a factorization of the transpose of the network Jacobian. Jordan also demonstrates the utility of this technique in a variety of control problems [37, 36, 40] Variants of this technique have been applied to control problems by Kawato [41], Miyata [55] Nguyen and Widrow [61] and others. However, as noted by Barto [7] as well as Jordan and Rumelhart [40] a back propagation network model is not essential for applying this technique. Any forward model of the process that can be differentiated can be used. In their study of reaching ....

M. Kawato. Computational schemes and neural network models for formation and control of multijoint arm trajectory. In T. Miller, R. S. Sutton, and P. J. Werbos, editors, Neural Networks for Control. The MIT Press, Cambridge, MA, 1990.

....learn an appropriate controller for a given controlled object. The next subsection will introduce three different approaches that have been proposed in the literature for this particular learning task. A more detailed account of this topic can be found, for example, in Jordan and Rumelhart (1992) Kawato (1990). See also CONTROL THEORY, BIOLOGICAL. 2.2 Forward and Inverse Models A forward model is a representation of the transformation from motor commands to movements, in other words a model of the controlled object. An inverse model is a representation of the transformation from desired movements to ....

....mapping. This would be most useful, for example, in the case of a controller that transforms a sensory stimulus or an internal representation of it into motor commands, as in the late stages of a sensorimotor transformation. A third approach, called feedback error learning and proposed by Kawato (1990), is depicted in Figure 4. In that scheme an error is computed between a desired trajectory and the actual trajectory executed by the plant. A so called feedback controller, which is a linear approximation of the inverse model, transforms the trajectory error into a motor command error, which is ....

Kawato, M., 1990, Computational Schemes and Neural Network Models for Formation and Control of Multijoint Arm Trajectory, in Neural Networks for Control, (W.T. Miller, R.S. Sutton, P.J. Werbos Eds.), Cambridge:MIT Press, pp. 197-228.

.... 1997; Lew Butterworth, 1997; Thelen et al. 1993; von Hofsten Ronnqvist, 1993) This research is complemented by a wide variety of computational models that simulate learning to reach (Berthier, 1996 1997; Berthier, Singh, Barto, Houk, 1993; Bullock Grossberg, 1988; Hinton, 1984; Kawato, 1990; Kettner, Marcario, Port, 1993; Kuperstein, 1988; Rosenbaum, Loukopoulos, Meulenbroek, Vaughan, Engelbrecht, 1995; Sporns Edelman, 1993; Vos Scheepstra, 1993) There are several important differences, though, between how these models simulate motor learning and how infants learn to reach. ....

....movement trajectory is planned. One strategy, often employed in robotics (see Hollerbach, 1990) is to first plan a reaching trajectory in Cartesian coordinates, and then to use an internal (inverse) model to convert these coordinates into joint angle changes or muscle activations (Flash, 1987; Kawato, 1990; Saltzman, 1979) However, as several infant researchers have argued, there is little or no evidence to suggest that infants use a pre planned motor program for reaching (e.g. Thelen, et al. 1993) Learning to Reach 4 Constraining the Movement Search Space A computational model of reaching ....

[Article contains additional citation context not shown here]

Kawato, M. (1990). Computational schemes and neural network models for formation and control of multijoint arm trajectory. In W.T. Miller, R.S. Sutton, & P.J. Werbos (Eds.), Neural networks for control, pp. 197-228. Boston: MIT Press.

....Jordan [64] shows that the back propagation process computes a factorization of the transpose of the network Jacobian. Jordan also demonstrates the utility of this technique in a variety of control problems [65, 64, 68] Variants of this technique have been applied to control problems by Kawato [70], Miyata [97] Nguyen and Widrow [107] and others. However, as noted by Barto [16] as well as Jordan and Rumelhart [68] the model need not be a back propagation network for this technique to apply. 22 Any forward model of the process that can be differentiated can be used. In their study of ....

Kawato, M. Computational schemes and neural network models for formation and control of multijoint arm trajectory. In T. Miller, R. S. Sutton, and P. J. Werbos, editors, Neural Networks for Control. The MIT Press, Cambridge, MA, 1990.

....and difficult to derive analytically. A direct approach to neural control is to use an ANN to learn this mapping based on the motor inputs and the sensorial outputs. This supervised learning scheme, often termed direct inverse modeling, is simple but hardly ever sufficient (e.g. see Kawato et al. [9] for a discussion) It assumes in particular a one to one mapping between inputs and outputs and specific conditions for the learning phase. Without specifically addressing robotic applications, Agarwal s [10] general classification of neural networkbased control approaches illustrates ....

Kawato, M. (1991), Computational Schemes and Neural Network Models for Formation and Control, in: Neural Networks for Control, A Bradford Book, MIT Press, pp. 5-58.

.... artificial systems (i.e. autonomous robots) It is now widely appreciated that this is a difficult problem because the systems are non linear, the kinematics can be unknown, there can be excess degrees of freedom offering nonunique solutions, and the available state information can contain errors [Kawato, 1990; Oyama et al. 1991] This problem is only important to the extent that it is desirable to have an autonomous robot that can see and manipulate objects in its environment. If we wish to replace man in some industries where the costs of labor exceed the benefits, then this and similar problems of ....

Kawato, M. [1990] Computational schemes and neural network models for formation and control of multijoint arm trajectory. In W.T. Miller, III, R.S. Sutton and P.J. Werbos (eds.) Neural Networks for Control, Cambridge, MA: MIT Press, 197-228.

....does take place in some cognitive systems 19 . However, while the premises of the DC argument might be true (that a distinction needs to be 18 (Clark and Grush, 1999) 19 The empirical evidence for the existence of emulator systems in cognitive systems is presented in (Wolpert, 1995) (Kawato, 1990). 8 made between cognitive and adaptive systems, for example) its conclusion that the capacity for emulation is a necessary condition for cognition, is problematic, if not just false. There are several problems with the model and I ll consider each one in turn. Firstly, as a description of our ....

Kawato, M. (1990). Computational Schemes and Neural Network Models for Formation and Control of Multi-joint Arm Trajectory. In W.T Miller III, R. Sutton and P. Werbos (eds.), Neural Networks for Control. Cambridge, MA: MIT Press.

.... evident in these movements One explanation is that the motor control system has evolved to optimize a quantity such as total jerk in the movement, and the velocity profile of the hand in reaching movements or the joint in single joint movements is a consequence of this optimization (see [5] [54]) To some extent, the joint torque profiles could even be simply scaled by the control system to produce different movements with similarly scaled velocity profiles at the hand [5] Or, in order to coordinate the joints of the arm to produce a nearly straight trajectory at the hand, the control ....

M. Kawato. Computational schemes and neural network models for formation and control of multijoint arm trajectory. In W. T. Miller III, R. S. Sutton, and P. J. Werbos, editors, Neural Networks for Control, pages 197--228. MIT Press, Cambridge, MA, 1990.

....from intentions to actions. The idea of using an internal model to augment the capabilities of supervised learning algorithms has also been proposed by Werbos (1987) although his perspective differs in certain respects from our own. There have been a number of further developments of the idea (Kawato, 1990; Miyata, 1988; Munro, 1987; Nguyen Widrow, 1989; Robinson Fallside, 1989; Schmidhuber, 1990) based either on the work of Werbos or our own unpublished work (Jordan, 1983; Rumelhart, 1986) There are also close ties between our approach and techniques in optimal control theory (Kirk, 1970) ....

....where the trajectory in the open interval from 0 to T is unconstrained. 17 One approach to solving such problems is to learn a one step forward model of the arm dynamics and then to use backpropagation in time in a recurrent network that includes the forward model and a controller (Jordan, 1990; Kawato, 1990). 18 In many problems involving delayed temporal consequences, however, it is neither feasible nor desirable to learn a dynamic forward model of the environment, either because the environment is too complex or because solving the task at hand does not require knowledge of the evolution of all ....

Kawato, M. (1990). Computational schemes and neural network models for formation and control of multijoint arm trajectory. In W. T. Miller, III, R. S. Sutton, & P. J. Werbos (Eds.), Neural Networks for Control. Cambridge: MIT Press.

....for robot control. Specifically, we are evolving neural networks to control a robot arm manipulator using visual input. Most neural network applications to this problem learn hand eye coordination through supervised training methods such as backpropagation or conjugate gradient descent (Kawato 1990; Miller 1989; van der Smagt 1995; Werbos 1992) Supervised learning, however, requires training examples that demonstrate correct mappings from input to output. The current approaches for generating training examples for robot arm control are very limited and ineffective in uncertain or ....

Kawato, M. (1990). Computational schemes and neural network models for formation and control of multijoint arm trajectory. In Neural Networks for Control. Cambridge, MA: MIT Press.

....model was learned by our algorithm, even when contaminated with irrelevant and redundant inputs. 2 The Curse of Dimensionality Both in research of biological and robotic motor control, the need for internal models has been emphasized in order to achieve accurate control of fast movements ([8]; 9] For models with only few input dimensions, learning approaches have been quite successful (e.g, 10] 11] 12] However, for higher dimensional learning tasks, it has been unclear whether learning approaches can succeed. For instance, we are interested in learning the inverse dynamics ....

M. Kawato, "Computational schemes and neural network models for formation and control of multijoint arm trajectory, " in Neural Networks for Control, W. T. Miller Iii, R. S. Sutton, and P. J. Werbos, Eds. Cambridge, MA: MIT Press, 1990, pp. 197-228.

....maintenance and retrieval of acquired sequential memory. III. HYPOTHESIS ON MULTIPLE REPRESENTATIONS IN THE BASAL GANGLIA LOOPS In visuomotor tasks such as visually guided reaching, problems in inverse kinematics and inverse dynamics must be solved to reach a target based on visual information [11]. We propose that it is easier to learn a visuomotor sequence in visual coordinates particularly when the sequence is learned by trial and error, whereas it is faster and easier to execute the sequence in body based coordinates once it is acquired. We further propose that it is advantageous to ....

M. Kawato, "Computational schemes and neural network models for formation and control of multijoint arm trajectory," in Neural Networks for Control, Cambridge, MA: MIT Press, 1990.

No context found.

Kawato, M. (1990). Computational schemes and neural network models for formation and control of multijoint arm trajectory. In W. T. Miller, III, R. S. Sutton, & P. J. Werbos (Eds.), Neural Networks for Control. Cambridge: MIT Press.

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