An online learning algorithm for adapting leg stiffness and stride angle for efficient quadruped robot trotting.

Animals adjust their leg stiffness and stride angle in response to changing ground conditions and gait parameters, resulting in improved stability and reduced energy consumption. This paper presents an online learning algorithm that attempts to mimic such animal behavior by maximizing energy efficiency on the fly or equivalently, minimizing the cost of transport of legged robots by adaptively changing the leg stiffness and stride angle while the robot is traversing on grounds with unknown characteristics. The algorithm employs an approximate stochastic gradient method to change the parameters in real-time, and has the following advantages: (1) the algorithm is computationally efficient and suitable for real-time operation; (2) it does not require training; (3) it is model-free, implying that precise modeling of the robot is not required for good performance; and (4) the algorithm is generally applicable and can be easily incorporated into a variety of legged robots with adaptable parameters and gaits beyond those implemented in this paper. Results of exhaustive performance assessment through numerical simulations and experiments on an under-actuated quadruped robot with compliant legs are included in the paper. The robot platform used a pneumatic piston in each leg as a variable, passive compliant element. Performance evaluation using simulations and experiments indicated that the algorithm was capable of converging to near-optimal values of the cost of transport for given operating conditions, terrain properties, and gait characteristics with no prior knowledge of the terrain and gait conditions. The simplicity of the algorithm and its demonstrably improved performance make the approach of this paper an excellent candidate for adaptively controlling tunable parameters of compliant, legged robots.

Shared Prosthetic Control Based on Multiple Movement Intent Decoders.

SIGNIFICANCE: A number of movement intent decoders exist in the literature that typically differ in the algorithms used and the nature of the outputs generated. Each approach comes with its own advantages and disadvantages. Combining the estimates of multiple algorithms may have better performance than any of the individual methods. OBJECTIVE: This paper presents and evaluates a shared controller framework for prosthetic limbs based on multiple decoders of volitional movement intent. METHODS: An algorithm to combine multiple estimates to control the prosthesis is developed in this paper. The capabilities of the approach are validated using a system that combines a Kalman filter-based decoder with a multilayer perceptron classifier-based decoder. The shared controller's performance is validated in online experiments where a virtual limb is controlled in real-time by amputee and intact-arm subjects. During the testing phase subjects controlled a virtual hand in real time to move digits to instructed positions using either a Kalman filter decoder, a multilayer perceptron decoder, or a linear combination of the two. RESULTS: The shared controller results in statistically significant improvements over the component decoders. Specifically, certain degrees of shared control result in increases in the time-in-target metric and decreases in unintended movements. CONCLUSION: The shared controller of this paper combines the good qualities of component decoders tested in this paper. Herein, combining a Kalman filter decoder with a classifier-based decoder inherits the flexibility of the Kalman filter decoder and the limited unwanted movements from the classifier-based decoder, resulting in a system that may be able to perform the tasks of everyday life more naturally and reliably.

Compensation of Temperature Effects on Lamb Waves Using Mode Decomposition and a Nonlinear Model.

Active Lamb-wave-based structural health monitoring techniques have been widely studied to inspect large structures using permanently installed arrays of sensors and actuators. Most of these methods depend on comparing baseline signals recorded from the structure before going into service and test signals acquired during inspection. Temperature changes affect the propagation of the wave in a nonlinear and mode-dependent manner. As a result, baseline comparison methods fail when the test and baseline signals are acquired at vastly different temperatures. Approximate methods that compensate for the effects of temperature on the waves using signal stretch models have been introduced in the literature. These methods are effective when the temperature changes are small and the propagation distances are short. However, they perform poorly when these conditions are not satisfied. Consequently, there is a need for better temperature compensation algorithms than presently available. This article presents a data-driven approach that separately compensates for the effects of temperature on different mode components of the sensor signals. The performance of the temperature compensation algorithm of this article is compared with that of a commonly used baseline signal stretch (BSS) algorithm using experimental signals measured from an aluminum panel and a unidirectional composite panel. Analysis results indicate that the method of this article outperforms the BSS algorithm for large temperature differences. The usefulness of the temperature compensation algorithm is further validated by demonstrating the ability of compensated signals to accurately reconstruct anomaly maps associated with damaged composite structures.

A Nonlinear Latching Filter to Remove Jitter From Movement Estimates for Prostheses.

Continuous movement intent decoders are critical for precise control of hand and wrist prostheses. Noise in biological signals (e.g., myoelectric or neural signals) can lead to undesirable jitter in the output of these types of decoders. A low-pass filter (LPF) at the output of the decoder effectively reduces jitter, but also substantially slows intended movements. This paper introduces an alternative, the latching filter (LF), a recursive, nonlinear filter that provides smoothing of small-amplitude jitter but allows quick changes to its output in response to large input changes. The performance of a Kalman filter (KF) decoder smoothed with an LF is compared with that of both an KF decoder without an additional smoother and a KF decoder smoothed with a LPF. These three algorithms were tested in real-time on target holding and target reaching tasks using surface electromyographic signals recorded from 5 non-amputee subjects, and intramuscular electromyographic and peripheral neural signals recorded from an amputee subject. When compared with the LPF, the LF provided a statistically significant improvement in amputee and non-amputee subjects' ability to hold the hand steady at requested positions and achieve movement goals faster. The KF decoder with LF provided a statistically significant improvement in all subjects' ability to hold the prosthetic hand steady, with only slightly lower speeds, when compared to the unsmoothed KF.

Deep Learning Movement Intent Decoders Trained With Dataset Aggregation for Prosthetic Limb Control.

SIGNIFICANCE: The performance of traditional approaches to decoding movement intent from electromyograms (EMGs) and other biological signals commonly degrade over time. Furthermore, conventional algorithms for training neural network based decoders may not perform well outside the domain of the state transitions observed during training. The work presented in this paper mitigates both these problems, resulting in an approach that has the potential to substantially improve the quality of life of the people with limb loss. OBJECTIVE: This paper presents and evaluates the performance of four decoding methods for volitional movement intent from intramuscular EMG signals. METHODS: The decoders are trained using the dataset aggregation (DAgger) algorithm, in which the training dataset is augmented during each training iteration based on the decoded estimates from previous iterations. Four competing decoding methods, namely polynomial Kalman filters (KFs), multilayer perceptron (MLP) networks, convolutional neural networks (CNN), and long short-term memory (LSTM) networks, were developed. The performances of the four decoding methods were evaluated using EMG datasets recorded from two human volunteers with transradial amputation. Short-term analyses, in which the training and cross-validation data came from the same dataset, and long-term analyses, in which the training and testing were done in different datasets, were performed. RESULTS: Short-term analyses of the decoders demonstrated that CNN and MLP decoders performed significantly better than KF and LSTM decoders, showing an improvement of up to 60% in the normalized mean-square decoding error in cross-validation tests. Long-term analyses indicated that the CNN, MLP, and LSTM decoders performed significantly better than a KF-based decoder at most analyzed cases of temporal separations (0-150 days) between the acquisition of the training and testing datasets. CONCLUSION: The short-term and long-term performances of MLP- and CNN-based decoders trained with DAgger demonstrated their potential to provide more accurate and naturalistic control of prosthetic hands than alternate approaches.

Lamb Wave Mode Decomposition Based on Cross-Wigner-Ville Distribution and Its Application to Anomaly Imaging for Structural Health Monitoring.

Lamb waves are characterized by their multimodal and dispersive propagation, which often complicates analysis. This paper presents a method for separation of the mode components and reflected components in sensor signals in an active structural health monitoring (SHM) system. The system is trained using linear chirp signals but works for arbitrary excitation signals. The training process employs the cross-Wigner-Ville distribution (xWVD) of the excitation signal and the sensor signal to separate the temporally overlapped modes in the time-frequency domain. The mode decomposition method uses a ridge extraction algorithm to separate each signal component in the time-frequency distribution. Once the individual modes are separated in the time-frequency domain, they are reconstructed in the time domain using the inverse xWVD operation. The propagation impulse response associated with each component can be directly estimated for chirp inputs. The estimated propagation impulse response can be used to separate the modes resulting from arbitrary excitation signals as long as their frequency components fall in the range of the chirp signal. The usefulness of the mode decomposition algorithm is demonstrated on a new health monitoring system for composite structures. This system performs anomaly imaging using the first arriving mode extracted from sensor array signals acquired from the structure. The anomaly maps are computed using a sparse tomographic reconstruction algorithm. The reconstructed map can locate anomalies on the structure and estimate their boundaries. Comparisons with methods that do not employ mode decomposition and/or sparse reconstruction techniques indicate a substantially better performance for the method of this paper.

Control of Dynamic Limb Motion Using Fatigue-Resistant Asynchronous Intrafascicular Multi-Electrode Stimulation.

Asynchronous intrafascicular multi-electrode stimulation (aIFMS) of small independent populations of peripheral nerve motor axons can evoke selective, fatigue-resistant muscle forces. We previously developed a real-time proportional closed-loop control method for aIFMS generation of isometric muscle force and the present work extends and adapts this closed-loop controller to the more demanding task of dynamically controlling joint position in the presence of opposing joint torque. A proportional-integral-velocity controller, with integrator anti-windup strategies, was experimentally validated as a means to evoke motion about the hind-limb ankle joint of an anesthetized feline via aIFMS stimulation of fast-twitch plantar-flexor muscles. The controller was successful in evoking steps in joint position with 2.4% overshoot, 2.3-s rise time, 4.5-s settling time, and near-zero steady-state error. Controlled step responses were consistent across changes in step size, stable against external disturbances, and reliable over time. The controller was able to evoke smooth eccentric motion at joint velocities up to 8 deg./s, as well as sinusoidal trajectories with frequencies up to 0.1 Hz, with time delays less than 1.5 s. These experiments provide important insights toward creating a robust closed-loop aIFMS controller that can evoke precise fatigue-resistant motion in paralyzed individuals, despite the complexities introduced by aIFMS.

Multiple-input single-output closed-loop isometric force control using asynchronous intrafascicular multi-electrode stimulation.

Although asynchronous intrafascicular multi-electrode stimulation (IFMS) can evoke fatigue-resistant muscle force, a priori determination of the necessary stimulation parameters for precise force production is not possible. This paper presents a proportionally-modulated, multiple-input single-output (MISO) controller that was designed and experimentally validated for real-time, closed-loop force-feedback control of asynchronous IFMS. Experiments were conducted on anesthetized felines with a Utah Slanted Electrode Array implanted in the sciatic nerve, either acutely or chronically ( n = 1 for each). Isometric forces were evoked in plantar-flexor muscles, and target forces consisted of up to 7 min of step, sinusoidal, and more complex time-varying trajectories. The controller was successful in evoking steps in force with time-to-peak of less than 0.45 s, steady-state ripple of less than 7% of the mean steady-state force, and near-zero steady-state error even in the presence of muscle fatigue, but with transient overshoot of near 20%. The controller was also successful in evoking target sinusoidal and complex time-varying force trajectories with amplitude error of less than 0.5 N and time delay of approximately 300 ms. This MISO control strategy can potentially be used to develop closed-loop asynchronous IFMS controllers for a wide variety of multi-electrode stimulation applications to restore lost motor function.

Application of the magnitude-squared coherence function between uterine and umbilical flow velocity waveforms for predicting placental dysfunction: a preliminary study.

To examine whether the magnitude-squared coherence between uterine and umbilical blood flow velocity waveforms can, in conjunction with estimated fetal weight, uterine and umbilical pulsatility indices, fetal and maternal heart rates, diastolic notching and the amniotic fluid index, create a sensitive and specific model for the prediction of placental dysfunction. Binary logistic prediction models are created for preeclampsia, pregnancy induced hypertension and intrauterine growth restriction in a study group of 284 unselected midtrimester pregnancies. In each study group, the median value of derived parameters were compared with the uncomplicated pregnancy control group. The magnitude-squared coherence function between the uterine and umbilical flow velocity waveforms was found to be a statistically significant predictor of preeclampsia during the midtrimester of pregnancy. The magnitude-squared coherence did not improve the prediction of intrauterine growth restriction or pregnancy induced hypertension. The inclusion of magnitude-squared coherence as one of the prediction parameters may improve the early identification of pregnancies subsequently complicated by preeclampsia.

A mathematical basis for the application of the modified geometric method to maximum frequency estimation.

The application of ultrasound in assessing the fetal cardiovascular system often requires the accurate estimation of maximum blood flow velocity waveforms using Doppler measurements. The modified geometric method estimates the maximum Doppler frequency as the frequency at which the vertical distance between the integrated spectrum and the reference line that connects the origin to the maximum value of the integrated spectrum is the largest. This paper presents a mathematical formulation for a class of maximum blood flow velocity estimation algorithms that includes the modified geometric method. The analysis provides a rationale for the continued use of the modified geometric method for estimating the maximum frequency envelopes of Doppler signals. This paper also contains experimental results demonstrating the superiority of the modified geometric method over a commonly used threshold crossing method.

Robust estimation of fetal heart rate variability using Doppler ultrasound.

This paper presents a new measure of heart rate variability (HRV) that can be estimated using Doppler ultrasound techniques and is robust to variations in the angle of incidence of the ultrasound beam and the measurement noise. This measure employs the multiple signal characterization (MUSIC) algorithm which is a high-resolution method for estimating the frequencies of sinusoidal signals embedded in white noise from short-duration measurements. We show that the product of the square-root of the estimated signal-to-noise ratio (SNR) and the mean-square error of the frequency estimates is independent of the noise level in the signal. Since varying angles of incidence effectively changes the input SNR, this measure of HRV is robust to the input noise as well as the angle of incidence. This paper includes the results of analyzing synthetic and real Doppler ultrasound data that demonstrates the usefulness of the new measure in HRV analysis.