Exoskeletons and exosuits have witnessed unprecedented growth in recent years, especially in the medical and industrial sectors. In order to be successfully integrated into the current society, these devices must comply with several commercialization rules and safety standards. Due to their intrinsic coupling with human limbs, one of the main challenges is to test and prove the quality of physical interaction with humans. However, the study of physical human–exoskeleton interactions (pHEI) has been poorly addressed in the literature. Understanding and identifying the technological ways to assess pHEI is necessary for the future acceptance and large-scale use of these devices. The harmonization of these evaluation processes represents a key factor in building a still missing accepted framework to inform human–device contact safety. In this review, we identify, analyze, and discuss the metrics, testing procedures, and measurement devices used to assess pHEI in the last ten years. Furthermore, we discuss the role of pHEI in safety contact evaluation. We found a very heterogeneous panorama in terms of sensors and testing methods, which are still far from considering realistic conditions and use-cases. We identified the main gaps and drawbacks of current approaches, pointing towards a number of promising research directions. This review aspires to help the wearable robotics community find agreements on interaction quality and safety assessment testing procedures.

This paper describes the design, development, and characterization of two wireless instrumented crutches for gait monitoring in order to provide clinicians quantitative parameters of upper limbs' contributions during walking. These parameters could be used to teach orthopedic patients to correctly use these supports and minimize problems connected to their usage. These instrumented crutches allow monitoring axial forces and shear forces, tilt angles, and time of impact on the ground in real time. Each crutch is composed of three strain-gauge bridges for measuring axial and shear forces, a conditioning circuit with transmission module, a triaxial accelerometer, a power management circuit, two batteries, and a biofeedback. The data are wirelessly transmitted via Bluetooth without needing any further readout unit, from the crutches to a personal computer, where the data are processed and displayed by a program created in LabVIEW. Each instrumented crutch was tested to assess the response of the accelerometer and the three strain-gauge bridges using a setup designed ad hoc. The mean experimental standard deviation was about 42 mV for axial forces corresponding to about 8 N and about 35 mV for shear forces corresponding to about 4 N. Hysteresis, linearity, and drift were calculated, and the obtained accuracy was about 8-9 N for axial forces and 4-5 N for shear forces. Furthermore, the crutches were tested during a walking activity of ten healthy subjects along a straight path for several trials. These crutches were used for a common analysis usually reported in the literature for weight bearing evaluation. The subjects were monitored performing a nonweight bearing (NWB) and a partial weight bearing (PWB) during a three-point gait. The results showed a mean of 102% ± 16% for NWB tests and a mean of 19% ±% for 10% PWB tests; these values are in agreement with similar studies in the literature. The simplicity that includes only constitutive strain gauges and a separable circuit board allows the achievement of the objectives of simplicity, ease of use, and noninvasiveness. Therefore, these crutches could be used as a support tool for controlling the use of crutches during walking not only in hospitals but also at home.

This paper is a first step towards a smart hand gesture recognition set up for Collaborative Robots using a Faster R-CNN Object Detector to find the accurate position of the hands in RGB images. In this work, a gesture is defined as a combination of two hands, where one is an anchor and the other codes the command for the robot. Other spatial requirements are used to improve the performances of the model and filter out the incorrect predictions made by the detector. As a first step, we used only four gestures.