This study presents and experimentally validates an adaptive control method for human-exoskeleton interaction through online adaptation of desired joint trajectories. Leveraging gait phase and human-exoskeleton interaction torque estimators, our approach enables seamless assistance adaptation to varying walking patterns and speeds. Specifically, a pre-trained neural network approximates the exoskeleton's dynamics, enabling real-time interaction torque estimation from kinematic measurements and commanded motor torques alone. These estimates drive a gradient-descent update of the joint reference trajectories, minimizing a cost function that penalizes both interaction torques and trajectory modification, ensuring bounded convergence and stability without user-specific parameter tuning. We compared our adaptive controller with a fixed-trajectory gait-phase-based controller during overground and treadmill walking at three self-selected speeds ranging from 0.4 to 0.8 m/s. In 16 participants, the adaptive controller significantly reduced the hip and knee interaction torques by 51.2%$\pm$11.1 and 63.9%$\pm$29.7, respectively, during overground walking. Muscular effort significantly decreased in Bicep Femoris (21.0%$\pm$34.5) and Rectus Femoris (28.1%$\pm$34.6), while remaining unchanged in other muscles. Cadence and gait speed increased by 7.6%$\pm$5.2 and 10.7%$\pm$8.3, respectively, indicating that participants could walk faster with less effort due to trajectory adaptation. Post-adaptation trajectories more closely resembled those of walking without the exoskeleton, and exoskeleton-torques aligned more closely with human biological torques. Our proposed adaptive controller, which requires only exoskeleton kinematics, also maintained performance during treadmill walking across speeds, demonstrating speed-invariant behaviour compared to the non-adaptive controller.

Intermittent pneumatic compression (IPC) systems apply external pressure to the lower limbs and enhance peripheral blood flow. We previously introduced a cardiac-gated compression system that enhanced arterial blood velocity (BV) in the lower limb compared to fixed compression timing (CT) for seated and standing subjects. However, these pilot studies found that the CT that maximized BV was not constant across individuals and could change over time. Current CT modelling methods for IPC are limited to predictions for a single day and one heartbeat ahead. However, IPC therapy for may span weeks or longer, the BV response to compression can vary with physiological state, and the best CT for eliciting the desired physiological outcome may change, even for the same individual. We propose that a deep reinforcement learning (DRL) algorithm can learn and adaptively modify CT to achieve a selected outcome using IPC. Herein, we target maximizing lower limb arterial BV as the desired outcome and build participant-specific simulated lower limb environments for 6 participants. We show that DRL can adaptively learn the CT for IPC that maximized arterial BV. Compared to previous work, the DRL agent achieves 98% ± 2 of the resultant blood flow and is faster at maximizing BV; the DRL agent can learn an "optimal" policy in 15 minutes ± 2 on average and can adapt on the fly. Given a desired objective, we posit that the proposed DRL agent can be implemented in IPC systems to rapidly learn the (potentially time-varying) "optimal" CT with a human-in-the-loop.

Intermittent pneumatic compression (IPC) systems are employed to treat vascular diseases. It has been shown that applying cardiac-gated compression effectively enhances femoral blood velocity (BV), but the optimal compression timing likely varies between individuals and may vary over time. While a previous work has shown the usability of one heartbeat ahead BV estimation to optimize the compression timing, that study was limited to a single treatment session and the BV estimator performance may deteriorate for the next sessions. Therefore, the goal of this study is to develop BV estimators and evaluate their accuracy over a longer time-scale. Six participants wore a custom IPC system and experienced random cardiac-gated compression timings for 1.5 hours per day for two days. A data- driven model was trained on electrocardiogram and applied pressure data to predict femoral BV one heartbeat ahead in a closed loop manner. The mean R2for this model across participants on the second session was 0.74 ± 0.09 and the mean absolute error was approximately 3%, which is a reduction of only 11% compared to the first sessions, for both metrics. This study is the first to show that BV across IPC sessions can be predicted using a pre-trained model. This work may lead to a significant improvement in IPC performance with only an initial model training session.

OBJECTIVE: To develop and evaluate an accurate method for cuffless blood pressure (BP) estimation during moderate- and heavy-intensity exercise. METHODS: Twelve participants performed three cycling exercises: a ramp-incremental exercise to exhaustion, and moderate and heavy pseudorandom binary sequence exercises on an electronically braked cycle ergometer over the course of 21 minutes. Subject-specific and population-based nonlinear autoregressive models with exogenous inputs (NARX) were compared with feedforward artificial neural network (ANN) models and pulse arrival time (PAT) models. RESULTS: Population-based NARX models, (applying leave-one-subject-out cross-validation), performed better than the other models and showed good capability for estimating large changes in mean arterial pressure (MAP). The models were unable to track consistent decreases in BP during prolonged exercise caused by reduction in peripheral vascular resistance, since this information is apparently not encoded in the employed proxy physiological signals (electrocardiography and forehead PPG) used for BP estimation. Nevertheless, the population-based NARX model had an error standard deviation of 11.0 mmHg during the entire exercise window, which improved to 9.0 mmHg when the model was periodically calibrated every 7 minutes. CONCLUSION: Population-based NARX models can estimate BP during moderate- and heavy-intensity exercise but need periodic calibration to account for the change in vascular resistance during exertion. SIGNIFICANCE: MAP can be continuously tracked during exercise using only wearable sensors, making monitoring exercise physiology more convenient and accessible.

The objective is to develop a cuffless method that accurately estimates blood pressure (BP) during activities of daily living. User-specific nonlinear autoregressive models with exogenous inputs (NARX) are implemented using artificial neural networks to estimate the BP waveforms from electrocardiography and photoplethysmography signals. To broaden the range of BP in the training data, subjects followed a short procedure consisting of sitting, standing, walking, Valsalva maneuvers, and static handgrip exercises. The procedure was performed before and after a six-hour testing phase wherein five participants went about their normal daily living activities. Data were further collected at a four-month time point for two participants and again at six months for one of the two. The performance of three different NARX models was compared with three pulse arrival time (PAT) models. The NARX models demonstrate superior accuracy and correlation with "ground truth" systolic and diastolic BP measures compared to the PAT models and a clear advantage in estimating the large range of BP. Preliminary results show that the NARX models can accurately estimate BP even months apart from the training. Preliminary testing suggests that it is robust against variabilities due to sensor placement. This establishes a method for cuffless BP estimation during activities of daily living that can be used for continuous monitoring and acute hypotension and hypertension detection.

In this work, we studied a combination of embedded magnetic measurement system in a knee prosthesis and wearable inertial sensors to estimate two knee joint rotations namely flexion-extension and internal-external rotations. The near optimal sensor configuration was designed for implantable measurement system, and linear estimators were used to estimate the mentioned angles. This system was separately evaluated in a mechanical knee simulator and the effect of the imposed Abduction-Adduction rotation was also studied on the angle estimations. To reduce the power consumption of the internal system, we reduced the sampling rate and duty cycled the implantable sensors. Then we compensated the lack of information via use of kinematic information from wearable sensors to provide accurate angle estimations. As long as this smart prosthesis is not implanted yet on a subject, the angles estimations from implantable sensors and wearable sensors are realistically simulated for four subjects. The simulated angle estimations were fed to the designed data fusion algorithms to boost the estimation performance. The results were considerably improved via use of Maximum Entropy Ordered Weighted Averaging (MEOWA) fusion for flexion angles, but not for internal-external angle estimations.

The recent advances in wearable inertial sensors opened a new horizon for pervasive measurement of human locomotion even in aquatic environment. In this paper we proposed an automatic approach of detecting the key temporal events of breaststroke swimming as a tentatively explored technique due to the complexity of the stroke. We used two inertial measurement units worn on the right arm and right leg of seven swimmers to capture the kinematics of the breaststroke. The detection of the temporal phases from the inertial signals was undertaken in the framework of a Hidden Markov Model (HMM). Supervised learning of the HMM parameters was achieved using the reference data from manual video analysis by an expert. The outputs of two well-known classifiers on the inertial signals were fused to unfold the input space of the HMM for an enhanced performance. An average correct phase detection of 93.5% for the arm stroke, 94.4% for the leg stroke and the minimum precision of 67 milliseconds in detection of the key events, suggests the accuracy of the method.

In this work, we present the general concept of an instrumented smart knee prosthesis for in-vivo measurement of forces and kinematics. This system can be used for early monitoring of the patient after implantation and prevent possible damage to the prosthesis. The diagnosis of defects can be done by detecting the load imbalance or abnormal forces and kinematics of the prosthetic knee in function. This work is a step towards the fabrication of an instrumented system for monitoring the function of the knee in daily conditions. Studying the constraints of commercially available prostheses, we designed a minimal sensory system and required electronics to be placed in the polyethylene part of prostheses. Three magnetic sensors and a permanent magnet were chosen and configured to measure the prosthetic knee kinematics. Strain gauges were designed to measure the forces applied to the polyethylene insert. Kinematic and force measurements were validated on a mechanical knee simulator by comparing them to different reference systems. Embedded electronics, including the A/D converters and amplifier were designed to acquire and condition the measurements to wirelessly transmit them to an external unit. By considering the necessary power budget for all components, the optimum coil for remote powering was investigated. The necessary rectifier and voltage doubler for remote powering were also designed. This is the first system capable of internally measuring force and kinematics simultaneously. We propose to package the system in the polyethylene part, bringing versatility to the instrumented system developed, as the polyethylene part can be easily modified for different types of prostheses based on the same principle, without changing the prosthesis design.

The method of multiple heterogeneous ant colonies with information exchange (MHACIE) is presented in this paper with emphasis on the speed of finding the optimal solution and the corresponding computational complexity. The proposed method which is inspired by biology and psychology has a structure composed of several ant colonies. These colonies participate in solving problems in a concurrently manner and also exchange information with each other in communicational steps. Each ant colony is considered as an intelligent agent with behavioral traits. These behavioral traits play a key role in the solving procedure, in interrelation circumstances and in installation of relations. Faster solutions have been achieved using different employments of agents in the algorithm structure. Experimental results show the superiority of Multiple heterogeneous ant colonies algorithm in comparison to the standard ant colony system (ACS) and particle swarm optimization (PSO) algorithms on different benchmarks. A dynamic, control engineering benchmark is also provided in order to gain a more complete evaluation of the proposed algorithm.