This work lays out our investigation into the noninvasive, automatic measurement of blood pressure. We have investigated the use of a thermal camera to detect changes in surface temperature due to the pulse in the radial and ulner arteries and then moved on to detecting the same pulse using a low-cost thermistor. Our preliminary results presented here indicate that the pulse is indeed detectable using a simple thermistor aparatus and that it should be possible in the future to develop a calibrated system to measure blood pressure using the pulse transit time method.

We consider risk sensitive filtering and smoothing for a dynamical system whose output is a vector process in /spl Ropf//sup 2/. The components of the observation process are a Markov process observed through a Brownian motion and a Markov process observed through a Poisson process. Risk-sensitive filters for the robust estimation of an indirectly observed Markov state processes are given. These filters are stochastic partial differential equations for which robust discretizations are obtained. Computer simulations are given which demonstrate the benefits of risk sensitive filtering.

We study the decoding problem in an uncertain noise environment. If the receiver knows the noise probability density function (PDF) at each time slot or its a priori probability, the standard Viterbi (1967) algorithm (VA) or the a posteriori probability (APP) algorithm can achieve optimal performance. However, if the actual noise distribution differs from the noise model used to design the receiver, there can be significant performance degradation due to the model mismatch. The minimax concept is used to minimize the worst possible error performance over a family of possible channel noise PDFs. We show that the optimal robust scheme is difficult to derive; therefore, alternative, practically feasible, robust decoding schemes are presented and implemented on a VA decoder and two-way APP decoder. The performance analysis and numerical results show our robust decoders have a performance advantage over standard decoders in uncertain noise channels, with no or little computational overhead. Our robust decoding approach can also explain why for turbo decoding overestimating the noise variance gives better results than underestimating it.

We study and develop several robust decoding algorithms based on the minimax rule. The algorithms robust to estimation errors and channel variations. The decoding problem is studied when using a matched and mismatched decoder. Examples using robust the Viterbi algorithm (1967) and MAP algorithm are also presented.

The well-known algorithms for time-optimal trajectory planning are difficult to apply in practice because they rely on imperfect models of the robot dynamics, they take no account of controller dynamics, and they provide no connection to tracking accuracy. In this paper we propose two schemes for planning and implementing time-optimal control to enable robots under computed torque control to track paths to a prescribed tolerance. Experimental results confirm the theories, showing that both schemes track to a prescribed accuracy in a near time-optimal fashion. The second scheme is shown to perform better, achieving complete torque utilization nearly all of the time.

In this paper, we propose an approach to solving the problem of robust time-optimal path tracking control to meet a prescribed tolerance in the presence of plant uncertainty. Our approach involves identifying the modelling errors as disturbances in joint accelerations and using a theory that relates such disturbances to the tracking performance of a robot under computed-torque control. The theory allows us to calculate the torque levels that need to be held in reserve and the controller gains that are required to reject the expected levels of disturbance. Our experiments show that if these levels of torque are held in reserve and if the controllers are tuned in this way, then the robot performs near time-optimal tracking meeting the specified tolerance for tracking accuracy.

In previous work the authors (1996) developed a method for planning robust time-optimal reference trajectories for robots under computed-torque control. That scheme was based on the idea of holding enough torque on reserve during trajectory planning to ensure that actuators would not saturate due to disturbances. This, together with the correct controller gain settings, ensured that tracking was accurate to a specified tolerance. This work extends that approach to a less conservative "closed-loop" architecture for "online trajectory generation" which is similar to Dahl and Neilsen's scheme (1990) for "online trajectory time-scaling". We propose a feedback control law for setting the reference path acceleration such that path tracking is nearly time-optimal, robustly controllable, and accurate to a prescribed tolerance. We show how this law can be determined offline using shooting methods and implemented with little online memory. Experimental results demonstrate that the tracking times are reduced compared to our previous approach, while tracking accuracy and robustness remain approximately the same.

In this paper we propose algorithms for parameter estimation of fast-sampled homogeneous Markov chains observed in white Gaussian noise. Our algorithms are obtained by the robust discretization of stochastic differential equations involved in the estimation of continuous-time hidden Markov models (HMM's) via the EM algorithm. We present two algorithms: the first is based on the robust discretization of continuous-time filters that were recently obtained by Elliott to estimate quantities used in the EM algorithm; the second is based on the discretization of continuous-time smoothers, yielding essentially the well-known Baum-Welch re-estimation equations. The smoothing formulas for continuous-time HMM's are new, and their derivation involves two-sided stochastic integrals. The choice of discretization results in equations which are identical to those obtained by deriving the results directly in discrete time. The filter-based EM algorithm has negligible memory requirements; indeed, independent of the number of observations. In comparison the smoother-based discrete-time EM algorithm requires the use of the forward-backward algorithm, which is a fixed-interval smoothing algorithm and has memory requirements proportional to the number of observations. On the other hand, the computational complexity of the filter-based EM algorithm is greater than that of the smoother-based scheme. However, the filters may be suitable for parallel implementation. Using computer simulations we compare the smoother-based and filter-based EM algorithms for HMM estimation. We provide also estimates for the discretization error.