Unpowered Knee Exoskeleton Reduces Quadriceps Activity during Cycling

Cycling is an eco-friendly method of transport and recreation. With the intent of reducing the energy cost of cycling without providing an additional energy source, we have proposed the use of a torsion spring for knee-extension support. We developed an exoskeleton prototype using a crossing four-bar mechanism as a knee joint with an embedded torsion spring. This study evaluates the passive knee exoskeleton using constant-power cycling tests performed by eight healthy male participants. We recorded the surface electromyography over the rectus femoris muscles of both legs, while the participants cycled at 200 and 225 W on a trainer with the developed wheel-accelerating system. We then analyzed these data in time-frequency via a continuous wavelet transform. At the same cycling speed and leg cadence, the median power spectral frequency of the electromyography increases with cycling load. At the same cycling load, the median power spectral frequency decreases when cycling with the exoskeleton. Quadriceps activity can be relieved despite the exoskeleton consuming no electrical energy and not delivering net-positive mechanical work. This fundamental can be applied to the further development of wearable devices for cycling assistance.

Image-based fall detection and classification of a user with a walking support system

The classification of visual human action is important in the development of systems that interact with humans. This study investigates an image-based classifica- tion of the human state while using a walking support system to improve the safety and dependability of these systems. We categorize the possible human behavior while utilizing a walker robot into eight states （i.e., sitting, standing, walking, and five falling types）, and propose two different methods, namely, normal distribution and hidden Markov models （HMMs）, to detect and recognize these states. The visual feature for the state classification is the centroid position of the upper body, which is extracted from the user＇s depth images. The first method shows that the centroid position follows a normal distribution while walking, which can be adopted to detect any non-walking state. The second method implements HMMs to detect and recognize these states. We then measure and compare the performance of both methods. The classification results are employed to control the motion of a passive-type walker （called ＂RT Walker＂） by activating its brakes in nonwalking states. Thus, the system can be used for sit/stand support and fall prevention. The experiments are performed with four subjects, including an experienced physiotherapist. Results show that the algorithm can be adapted to the new user＇s motion pattern within 40 s, with a fall detection rate of 96.25% and state classification rate of 81.0%. The proposed method can be implemented to other abnormality detection/classification applications that employ depth image-sensing devices.