Movement impairments resulting from neurologic injuries, such as stroke, can be treated with robotic exoskeletons that assist with movement retraining. Exoskeleton designs benefit from low impedance and accurate torque control. We designed a 2 degree-of-freedom tethered exoskeleton that can provide independent torque control on elbow flexion/extension and forearm supination/pronation. Two identical series elastic actuators (SEAs) are used to actuate the exoskeleton. The two SEAs are coupled through a novel cable-driven differential. The exoskeleton is compact and lightweight, with a mass of 0.9 kg. Applied RMS torque errors were less than 0.19 Nm. Benchtop tests demonstrated a torque rise time of approximately 0.1 s, a torque control bandwidth of 3.7 Hz and an impedance of less than 0.03 Nm/deg at 1 Hz. The controller can simulate a stable maximum wall stiffness of 0.45 Nm/deg. The overall performance is adequate for robotic therapy applications and the novelty of the design is discussed.

Techniques for automatically designing deep neural network architectures such as reinforcement learning based approaches have recently shown promising results. However, their success is based on vast computational resources (e.g. hundreds of GPUs), making them difficult to be widely used. A noticeable limitation is that they still design and train each network from scratch during the exploration of the architecture space, which is highly inefficient. In this paper, we propose a new framework toward efficient architecture search by exploring the architecture space based on the current network and reusing its weights. We employ a reinforcement learning agent as the meta-controller, whose action is to grow the network depth or layer width with function-preserving transformations. As such, the previously validated networks can be reused for further exploration, thus saves a large amount of computational cost. We apply our method to explore the architecture space of the plain convolutional neural networks (no skip-connections, branching etc.) on image benchmark datasets (CIFAR-10, SVHN) with restricted computational resources (5 GPUs). Our method can design highly competitive networks that outperform existing networks using the same design scheme. On CIFAR-10, our model without skip-connections achieves 4.23% test error rate, exceeding a vast majority of modern architectures and approaching DenseNet. Furthermore, by applying our method to explore the DenseNet architecture space, we are able to achieve more accurate networks with fewer parameters.

Ankle inversion-eversion compliance is an important feature of conventional prosthetic feet, and control of inversion, or roll, in active prostheses could improve balance for people with amputation. We designed a tethered ankle-foot prosthesis with two independently actuated toes that are coordinated to provide plantarflexion and inversion-eversion torques. A Bowden cable tether provides series elasticity. The prosthesis is simple and lightweight, with a mass of 0.72kg. Strain gauges on the toes measure torque with less than 1% root mean squared (RMS) error. Benchtop tests demonstrated a step response rise time of less than 33 ms, peak torques of 250 N·m in plantarflexion and ±30 N·m in inversion-eversion, and peak power above 3 kW. The phase-limited closed-loop torque bandwidth is 20 Hz with a chirp from 10 to 90 N·m in plantarflexion, and 24 Hz with a chirp from -20 to 20 N·m in inversion. The system has low sensitivity to toe position disturbances at frequencies of up to 18 Hz. Walking trials with an amputee subject demonstrated RMS torque tracking errors of less than 5.1 N·m in plantarflexion and less than 1.5 N·m in inversion-eversion. These properties make the platform suitable for testing inversion-related prosthesis features and controllers in experiments with humans.

We present a new type of hydrostatic transmission that uses a hybrid air-water configuration, analogous to N+1 cable-tendon transmissions, using N hydraulic lines and 1 pneumatic line for a system with N degrees of freedom (DOFs). The common air-filled line preloads all DOFs in the system, allowing bidirectional operation of every joint. This configuration achieves the high stiffness of a water-filled transmission with half the number of bulky hydraulic lines. We implemented this transmission using pairs of rolling-diaphragm cylinders to form rotary hydraulic actuators, with a new design achieving a 600-percent increase in specific work density per cycle. These actuators were used to build a humanoid robot with two 4-DOF arms, connected via the hydrostatic transmission to an identical master. Stereo cameras mounted on a 2-DOF servo-controlled neck stream live video to the operator's head-mounted display, which in turn sends the real-time attitude of the operator's head to the neck servos in the robot. The operator is visually immersed in the robot's physical workspace, and through the bilateral coupling of the low-impedance hydrostatic transmission, directly feels interaction forces between the robot and external environment. We qualitatively assessed the performance of this system for remote object manipulation and use as a platform to safely study physical human-robot interaction.

Ankle inversion-eversion compliance is an important feature of conventional prosthetic feet, and control of inversion, or roll, in robotic prostheses could improve balance for people with amputation. We designed a tethered ankle-foot prosthesis with two independently-actuated toes that are coordinated to provide plantarflexion and inversion-eversion torques. This configuration allows a simple lightweight structure with a total mass of 0.72 kg. Strain gages on the toes measure torque with less than 2.7% RMS error, while compliance in the Bowden cable tether provides series elasticity. Benchtop tests demonstrated a 90% rise time of less than 33 ms and peak torques of 180 N·m in plantarflexion and ±30 N·m in inversion-eversion. The phase-limited closedloop torque bandwidth is 20 Hz with a 90 N·m amplitude chirp in plantarflexion, and 24 Hz with a 20 N·m amplitude chirp in inversion-eversion. The system has low sensitivity to toe position disturbances at frequencies of up to 18 Hz. Walking trials with five values of constant inversion-eversion torque demonstrated RMS torque tracking errors of less than 3.7% in plantarflexion and less than 5.9% in inversion-eversion. These properties make the platform suitable for haptic rendering of virtual devices in experiments with humans, which may reveal strategies for improving balance or allow controlled comparisons of conventional prosthesis features. A similar morphology may be effective for autonomous devices.