Human-in-the-loop optimization of rocker shoes via different cost functions during walking.

Personalised footwear could be used to enhance the function of the foot-ankle complex to a person's maximum. Human-in-the-loop optimization could be used as an effective and efficient way to find a personalised optimal rocker profile (i.e., apex position and angle). The outcome of this process likely depends on the selected optimization objective and its responsiveness to the rocker parameters being tuned. This study aims to explore whether and how human-in-the-loop optimization via different cost functions (i.e., metabolic cost, collision work as measure for external mechanical work, and step distance variability as measure for gait stability) affects the optimal apex position and angle of a rocker profile differently for individuals during walking. Ten healthy individuals walked on a treadmill with experimental rocker shoes in which apex position and angle were optimized using human-in-the-loop optimization using different cost functions. We compared the obtained optimal apex parameters for the different cost functions and how these affected the selected gait related objectives. Optimal apex parameters differed substantially between participants and optimal apex positions differed between cost functions. The responsiveness to changes in apex parameters differed between cost functions. Collision work was the only cost function that resulted in a significant improvement of its performance criteria. Improvements in metabolic cost or step distance variability were not found after optimization. This study showed that cost function selection is important when human-in-the-loop optimization is used to design personalised footwear to allow conversion to an optimum that suits the individual.

The gait1415+2 OpenSim musculoskeletal model of transfemoral amputees with a generic bone-anchored prosthesis.

This short communication presents the gait1415+2 musculoskeletal model, that has been developed in OpenSim to describe the lower-extremity of a human subject with transfemoral amputation wearing a generic lower-limb bone-anchored prosthesis. The model has fourteen degrees of freedom, governed by fifteen musculotendon units (placed at the contralateral and residual limbs) and two generic actuators (one placed at the knee joint and one at the ankle joint of the prosthetic leg). Even though the model is a simplified abstraction, it is capable of generating a human-like walking gait and, specifically, it is capable of reproducing both the kinematics and the dynamics of a person with transfemoral amputation wearing a bone-anchored prosthesis during normal level-ground walking. The model is released as support material to this short communication with the final goal of providing the scientific community with a tool for performing forward and inverse dynamics simulations, and for developing computationally-demanding control schemes based on artificial intelligence methods for lower-limb prostheses.

Learning to Walk With Deep Reinforcement Learning: Forward Dynamic Simulation of a Physics-Based Musculoskeletal Model of an Osseointegrated Transfemoral Amputee.

This paper leverages the OpenSim physics-based simulation environment for the forward dynamic simulation of an osseointegrated transfemoral amputee musculoskeletal model, wearing a generic prosthesis. A deep reinforcement learning architecture, which combines the proximal policy optimization algorithm with imitation learning, is designed to enable the model to walk by using three different observation states. The first is a complete state that includes the agent's kinematics, ground reaction forces, and muscle data; the second is a reduced state that only includes the kinematics and ground reaction forces; the third is an augmented state that combines the kinematics and ground reaction forces with a prediction of the muscle data generated by a fully-connected feed-forward neural network. The empirical results demonstrate that the model trained with the augmented observation state can achieve walking patterns with rewards and gait symmetry ratings comparable to those of the model trained with the complete observation state, while there are no symmetric walking patterns when using the reduced observation state. This paper shows the importance of including muscle data in a deep reinforcement learning architecture for the forward dynamic simulation of musculoskeletal models of transfemoral amputees.

Musculoskeletal model of osseointegrated transfemoral amputees in OpenSim.

This study presents a generic OpenSim musculoskeletal model of people with an osseointegrated unilateral transfemoral amputation wearing a generic prosthesis. The model, which consists of seventy-six musculotendon units and two ideal actuators at the knee and ankle joints of the prosthesis, is tested by designing an optimal control strategy that guarantees the tracking of experimental amputee data during level-ground walking while finding the actuators' torques and minimizing the muscle forces. The model can be made subject-specific and, as such, is able to reproduce the kinematics and dynamics of both healthy and amputee subjects. The model provides a tool to analyze the biomechanics of level-ground walking and to understand the contribution of the muscles and of the prosthesis' actuators. The proposed OpenSim musculoskeletal model is released as support material to this study.

Self-Sensing Soft Skin Based on Piezoelectric Nanofibers.

The development of electronic skins and wearable devices is rapidly growing due to their broad application fields, such as for biomedical, health monitoring, or robotic purposes. In particular, tactile sensors based on piezoelectric polymers, which feature self-powering capability, have been widely used thanks to their flexibility and light weight. Among these, poly(vinylidenefluoride-trifluoroethylene) (PVDF-TrFE) presents enhanced piezoelectric properties, especially if manufactured in a nanofiber shape. In this work, the enhanced piezoelectric performances of PVDF-TrFE nanofibers were exploited to manufacture a flexible sensor which can be used for wearable applications or e-skin. The piezoelectric signal was collected by carbon black (CB)-based electrodes, which were added to the active layer in a sandwich-like structure. The sensor was electromechanically characterized in a frequency range between 0.25 Hz and 20 Hz-which is consistent with human activities (i.e., gait cycle or accidental bumps)-showing a sensitivity of up to 4 mV/N. The parameters of the signal acquisition circuit were tuned to enable the sensor to work at the required frequency. The proposed electrical model of the nanofibrous piezoelectric sensor was validated by the experimental results. The sensitivity of the sensor remained above 77.5% of its original value after 106 cycles of fatigue testing with a 1 kN compressive force.

IMU-Based Classification of Locomotion Modes, Transitions, and Gait Phases with Convolutional Recurrent Neural Networks.

This paper focuses on the classification of seven locomotion modes (sitting, standing, level ground walking, ramp ascent and descent, stair ascent and descent), the transitions among these modes, and the gait phases within each mode, by only using data in the frequency domain from one or two inertial measurement units. Different deep neural network configurations are investigated and compared by combining convolutional and recurrent layers. The results show that a system composed of a convolutional neural network followed by a long short-term memory network is able to classify with a mean F1-score of 0.89 and 0.91 for ten healthy subjects, and of 0.92 and 0.95 for one osseointegrated transfemoral amputee subject (excluding the gait phases because they are not labeled in the data-set), using one and two inertial measurement units, respectively, with a 5-fold cross-validation. The promising results obtained in this study pave the way for using deep learning for the control of transfemoral prostheses with a minimum number of inertial measurement units.

Learning to Ascend Stairs and Ramps: Deep Reinforcement Learning for a Physics-Based Human Musculoskeletal Model.

This paper proposes to use deep reinforcement learning to teach a physics-based human musculoskeletal model to ascend stairs and ramps. The deep reinforcement learning architecture employs the proximal policy optimization algorithm combined with imitation learning and is trained with experimental data of a public dataset. The human model is developed in the open-source simulation software OpenSim, together with two objects (i.e., the stairs and ramp) and the elastic foundation contact dynamics. The model can learn to ascend stairs and ramps with muscle forces comparable to healthy subjects and with a forward dynamics comparable to the experimental training data, achieving an average correlation of 0.82 during stair ascent and of 0.58 during ramp ascent across both the knee and ankle joints.

IMU-Based Deep Neural Networks: Prediction of Locomotor and Transition Intentions of an Osseointegrated Transfemoral Amputee.

This paper focuses on the design and comparison of different deep neural networks for the real-time prediction of locomotor and transition intentions of one osseointegrated transfemoral amputee using only data from inertial measurement units. The deep neural networks are based on convolutional neural networks, recurrent neural networks, and convolutional recurrent neural networks. The architectures' input are features in both the time domain and the time-frequency domain, which are derived from either one inertial measurement unit (placed above the prosthetic knee) or two inertial measurement units (placed above and below the prosthetic knee). The prediction of eight different locomotion modes (i.e., sitting, standing, level ground walking, stair ascent and descent, ramp ascent and descent, walking on uneven terrain) and the twenty-four transitions among them is investigated. The study shows that a recurrent neural network, realized with four layers of gated recurrent unit networks, achieves (with a 5-fold cross-validation) a mean F1 score of 84.78% and 86.50% using one inertial measurement unit, and 93.06% and 89.99% using two inertial measurement units, with or without sitting, respectively.

Deep Reinforcement Learning for Physics-Based Musculoskeletal Simulations of Healthy Subjects and Transfemoral Prostheses' Users During Normal Walking.

This paper proposes to use deep reinforcement learning for the simulation of physics-based musculoskeletal models of both healthy subjects and transfemoral prostheses' users during normal level-ground walking. The deep reinforcement learning algorithm is based on the proximal policy optimization approach in combination with imitation learning to guarantee a natural walking gait while reducing the computational time of the training. Firstly, the optimization algorithm is implemented for the OpenSim model of a healthy subject and validated with experimental data from a public data-set. Afterwards, the optimization algorithm is implemented for the OpenSim model of a generic transfemoral prosthesis' user, which has been obtained by reducing the number of muscles around the knee and ankle joints and, specifically, by keeping only the uniarticular ones. The model of the transfemoral prosthesis' user shows a stable gait, with a forward dynamic comparable to the healthy subject's, yet using higher muscles' forces. Even though the computed muscles' forces could not be directly used as control inputs for muscle-like linear actuators due to their pattern, this study paves the way for using deep reinforcement learning for the design of the control architecture of transfemoral prostheses.

Towards Poly(vinylidene fluoride-trifluoroethylene-chlorotrifluoroethylene)-Based Soft Actuators: Films and Electrospun Aligned Nanofiber Mats.

In the pursuit of designing a linear soft actuator with a high force-to-weight ratio and a stiffening behavior, this paper analyzes the electrostrictive effect of the poly(vinylidene fluoride-trifluoroethylene-chlorotrifluoroethylene) polymer in the form of film and aligned electrospun nanofiber mat. An experimental setup is realized to evaluate the electrostrictive effect of the specimens disjointly from the Maxwell stress. In particular, an uniaxial load test is designed to evaluate the specimens' forces produced by their axial contraction (i.e., the electrostrictive effect) when an external electric field is applied, while an uniaxial tensile load test is designed to show the specimens' stiffening properties. This electro-mechanical analysis demonstrates that both the film and the nanofiber mat are electrostrictive, and that the nanofiber mat exhibits a force-to-weight ratio ∼65% higher than the film and, therefore, a larger electrostrictive effect. Moreover, both the film and the nanofiber mat show a stiffening behavior, which is more evident for the nanofiber mat than the film and is proportional to the weight of the material. This study concludes that, thanks to its electro-mechanical properties, the poly(vinylidene fluoride-trifluoroethylene-chlorotrifluoroethylene), especially in the form of aligned electrospun nanofiber mat, has high potential to be used as electro-active polymer for soft actuators in biomedical and biorobotics applications.

Conceptual Design of a Fully Passive Transfemoral Prosthesis to Facilitate Energy-Efficient Gait.

In this paper, we present the working principle and conceptual design toward the realization of a fully-passive transfemoral prosthesis that mimics the energetics of the natural human gait. The fundamental property of the conceptual design consists of realizing an energetic coupling between the knee and ankle joints of the mechanism. Simulation results show that the power flow of the working principle is comparable with that in human gait and a considerable amount of energy is delivered to the ankle joint for the push-off generation. An initial prototype in half scale is realized to validate the working principle. The construction of the prototype is explained together with the test setup that has been built for the evaluation. Finally, experimental results of the prosthesis prototype during walking on a treadmill show the validity of the working principle.

A novel energy-efficient rotational variable stiffness actuator.

This paper presents the working principle, the design and realization of a novel rotational variable stiffness actuator, whose stiffness can be varied independently of its output angular position. This actuator is energy-efficient, meaning that the stiffness of the actuator can be varied by keeping constant the internal stored energy of the actuator. The principle of the actuator is an extension of the principle of translational energy-efficient actuator vsaUT. A prototype based on the principle has been designed, in which ball-bearings and linear slide guides have been used in order to reduce losses due to friction.

Stiffness and position control of a prosthetic wrist by means of an EMG interface.

In this paper, we present a novel approach for decoding electromyographic signals from an amputee and for interfacing them with a prosthetic wrist. The model for the interface makes use of electromyographic signals from electrodes placed in agonistic and antagonistic sides of the forearm. The model decodes these signals in order to control both the position and the stiffness of the wrist.