Most visual servoing approaches explore the requirements for tracking and eventually grasping of a moving object based on the task at hand. On the other hand, active vision relies on the properties of the environment and the task to define a strategy of measurements by one or more cameras. The focus of our work is to achieve a high level of interaction between the two approaches. In fact the proposed approach addresses at the same time the optimization of the observation process and the achievement of the task at hand. Simulation results are presented for the case of an uncertain pendulum movement. An integrated sensing and actuation system that can operate in a dynamic as well as static environment is presented. It is a multi-sensor system that coordinates observations control, robotic arm command and intercepting the moving object.

A harmonic drive is a compact, light-weight and high-ratio torque transmission device which has almost zero backlash. Its unique performance features captures the attention of designers in many industrial applications, especially in robotics. However, the torque control of harmonic drive systems is still a challenging problem for researchers. In this paper the torque control of an harmonic drive system for constrained motion is examined in detail. A nominal model for the system is obtained from experimental frequency responses of the system, and the deviation of the system from the model is encapsulated by multiplicative uncertainty. A robust torque controller is designed using this information in an H/sub /spl infin//-framework, and implemented on two different setups. It is illustrated that the performance features of the closed-loop system is exceptionally good, both in time and frequency domains.

A harmonic drive is a compact, light-weight and high-ratio torque transmission device which is used in many electrically actuated robot manipulators. In many robotic control strategies it is assumed that the actuator is an ideal torque source. However, converting harmonic drive systems to ideal torque sources is still a challenging control problem for researchers. In this paper the torque control of harmonic drive system under free motion is examined in detail. A built-in torque sensor is developed in order to measure the torque, and by employing a Kalman filter the undesired torque signatures like torque ripples and misalignment torque are filtered out. An empirical nominal model for the system is obtained through experimental frequency response estimates, and the deviation of the system from the model is encapsulated by multiplicative uncertainty. A robust torque controller is subsequently designed in an H/sub /spl infin//-framework and implemented employing Kalman filtered torque estimates. From time and frequency domain experiments, it is shown that the closed-loop system retains robust stability, while improving the tracking performance exceptionally well.

A novel 3-D ultrasound position sensor provides low-cost, high-precision measurements of the Cartesian position of a robot wrist within a the robot's workspace. These measurements can, for instance, be used in a taskspace feedback loop to reduce the kinematic error. The basic idea is to mount ultrasound senders on the manipulator wrist and compute the 3-D Cartesian position of these sensors using 1-D range measurements between the senders and an array of ultrasound receivers at fixed, known locations. The main obstacle to precise range measurements is the presence of air turbulences and convection currents. This gives rise to noise which is spatially and temporally correlated. A stochastic model of this noise is presented and backed by experiments. It is proposed that the changing 3-D Cartesian position of the moving wrist is tracked by the sensor using a modified extended Kalman filter which incorporates the stochastic noise model and information derived from a static reference sender.

The output of a shaft encoder is basically a quantized measurement of the shaft angle. One method of angular velocity and acceleration measurement is by finite difference of angle measurements (net pulse counts) at regular time samples. It is shown that these estimates become significantly degraded as sampling rates increase. Kalman filtering of the angle measurements, based on some model of signal generation, is proposed. Asymptotic analysis of constant sampling rate as the intersample time tends to zero shows that the error variances tend to zero, but at decreasing rates for the angle, velocity, and acceleration errors. The analysis also suggests the use of multiple-integrator systems to model the angle signal.>

A new approach to the modeling of the evolution in time of disease treatment is presented. The resulting model is that of a white stochastic incidence sequence generating a prevalence sequence as the output of a linear, stochastically time-varying system. The average time behavior of the episode is identified through the prevalence autocorrelation function; the latter is measured from data generated by the Quebec health-care system. The distribution of the episode duration is obtained and used to derive other statistical quantities.