This paper presents results of an effort to use the dynamic information present in an on-line handwriting signal. On-line handwriting recognition systems often ignore most of the dynamic information available in the signal. They commonly go to the extent of retaining the order of points being sampled and throwing away all other dynamic (speed) information through a resampling of some sort. The system of [1], developed by our group at IBM Research. In the algorithm used by the system in [1], the writing is re-sampled to produce equidistant points; then a segment of the writing with a fixed number of points (hence the same Euclidean Length) is used to produce the feature vector. This feature extraction technique is used for both training and decoding. With a system of this type, the characters should be formed at a nominal size to get an acceptable comparison. For this reason, a size normalization is very important before training or decoding.

...rganizing associative neural network architecture that can be used to approximate the inverse-dynamics in the form of a Position-and-Direction-to-Action (PDA) map is also described. Similarities between the basal ganglia -- thalamocortical loops and the SDS scheme are discussed and it is argued that the SDS scheme could be viewed as a model of higher order motor functions of these areas.

In this paper a learning method is described which enables a conventional industrial robot to accurately execute the teach-in path in presence of dynamical effects and high speed. After training the system is capable of generating positional commands that in combination with the standard robot controller lead the robot along the desired trajectory. The mean path deviations are reduced to a factor of 20 for our test configuration. For low speed motion the learned controllers' accuracy is in the range of the resolution of the positional encoders. The learned controller does not depend on specific trajectories. It acts as a general controller that can be used for non-recurring tasks as well as for sensor-based planned paths. For repetitive control tasks accuracy can be even increased. Such improvements are caused by a three level structure estimating a simple process model, optimal a posteriori commands, and a suitable feedforward controller, the latter including neural networks for the r...

Humans exploit dynamics---gravity, inertia, joint coupling, elasticity, and so on---as a regular part of skillful, coordinated movements. Such movements comprise everyday activities, like reaching and walking, as well as highly practiced maneuvers as used in athletics and the performing arts. Robots, especially industrial manipulators, instead use control schemes that ordinarily cancel the complex, nonlinear dynamics that humans use to their advantage. Alternative schemes from the machine learning and intelligent control communities offer a number of potential benefits, such as improved efficiency, online skill acquisition, and tracking of nonstationary environments. However, the success of such methods depends a great deal on structure in the form of simplifying assumptions, prior knowledge, solution constraints and other heuristics that bias learning. My premise

We consider the problem of learning how to control a plant with non-linear control characteristics and solving the path-planning problem at the same time. The solution is based on a path-planning model that designates a speed field to be tracked, the speed field being the gradient of the equilibrium solution of a diffusion-like process which is simulated on an artificial neural network by spreading activation. The relaxed diffusion field serves as the input to the interneurons which detect the strength of activity flow in between neighboring discretizing neurons. These neurons then emit the control signals to control neurons which are linear elements. The interneuron to control-neuron connections are trained by a variant of Hebb's rule during control. The proposed method, whose most attractive feature is that it integrates reactive path-planning and continuous motion control in a natural fashion, can be used for learning redundant control problems. 2 Contents 1

The problem of controlling underwater mobile robots in 6 degrees of freedom (DOF) is addressed. Uncertainties in the input matrix due to partly known nonlinear thruster characteristics are modelled as multiplicative input uncertainty. This paper proposes two methods to compensate for the model uncertainties: (1) an adaptive passivitybased control scheme and (2) deriving a hybrid (adaptive and sliding) controller. The hybrid controller consists of a switching term which compensates for uncertainties in the input matrix and an on-line parameter estimation algorithm. Global stability is ensured by applying Barbalat's Lyapunov-like lemma. The hybrid controller is simulated for the horizontal motion of the Norwegian Experimental Remotely Operated Vehicle (NEROV). 1 Introduction Non-destructive testing of underwater structures require high performance manoeuvres of underwater mobile robots within and close to underwater installations. Until recently, remotely operated vehicles (ROVs) have b...

This paper reports comparative experiments with a new model-based adaptive force control algorithm for robot arms. This controller provides simultaneous position and force trajectory tracking of a robot arm whose tool tip is in point contact with a smooth rigid surface. The algorithm is provably stable with respect to the commonly accepted rigid-body nonlinear dynamical model for robot arms. Comparative experiments show the new adaptive model-based controller to provide performance superior to that of both (i) non model-based controllers and (ii) non adaptive controllers over a wide range of operating conditions. I.

This dissertation addresses three related topics in the application of virtual fix-tures to bilateral telemanipulation systems. Bilateral telemanipulation is the direct human control of a remote robot, with force and/or tactile feedback, and virtual fixtures are guidance modes, implemented in software, that assist the user in ac-complishing a telemanipulated task. The first topic addressed in this dissertation is the design of functional and stable forbidden-region virtual fixtures, which prevent robot motion into forbidden-regions of the workspace. Metrics are defined to evalu-ate the effectiveness of forbidden-region virtual fixtures, and a human-factors exper-iment uses these metrics to quantify how users interact with various combinations of forbidden-region virtual fixtures and telemanipulation control system. A method to predict system stability that incorporates an explicit model of the telemanipulator and bounding models of human users is created and experimentally verified. Next, a new condition is presented for the passivity of a virtual wall with sampling, sensor quantization, and friction effects, for an impedance-type robot. This condition is

Abstract—This paper is concerned with the design of a neuroadaptive trajectory tracking controller. The paper presents a new control scheme based on inversion of a feedforward neural model of a robot arm. The proposed control scheme requires two modules. The first module consists of an appropriate feedforward neural model of forward dynamics of the robot arm that continuously accounts for the changes in the robot dynamics. The second module implements an efficient network inversion algorithm that computes the control action by inverting the neural model. In this paper, a new extended Kalman filter (EKF) based network inversion scheme is proposed. The scheme is evaluated through comparison with two other schemes of network inversion: gradient search in input space and Lyapunov function approach. Using these three inversion schemes the proposed controller was implemented for trajectory tracking control of a two-link manipulator. Simulation results in all cases confirm the efficacy of control input prediction using network inversion. Comparison of the inversion algorithms in terms of tracking accuracy showed the superior performance of the EKF based inversion scheme over others. R I.

The development and expression of many higher level cognitive functions, such as imitation, spatial perception, and tool-use relies on a multi-modal representation of the body known as the body schema. Although many studies support the hypothesis that the body schema is adaptive and alterable throughout ontogenetic development, the mechanisms underlying its plasticity have yet to be clarified. Here we argue that the temporal integration of multisensory information is a plausible candidate mechanism to explain how manipulated objects (e.g. tools) can become incorporated into the body schema. To demonstrate the validity of our idea, we introduce a model of body schema adaptation instantiated in a small-sized, table-top, tool-using humanoid. The robot’s task is to learn to reach for and touch a visually salient distant object, first with its “bare ” hand, and then— using the acquired know-how—with a reach-extending tool (a stick). Our experimental results show that in order to successfully causally relate and integrate vision, touch and proprioception, and to learn to use the tool, timing is of crucial relevance. On a more general note, this study also suggests that synthetic modeling might not only be a valid avenue towards getting a better grasp on results provided by neuropsychology and neurophysiology, but also a powerful approach for building advanced tool-using robots.