Paper Submission & Registration
8th Dutch Bio-Medical Engineering Conference
16:30   Man-machine interfacing
Chair: Ying Wang
15 mins
Design of a soft exosuit for the lower back
Saivimal Sridar, Allan Veale, Massimo Sartori, Herman van der Kooij
Abstract: In recent decades, the incidence of low-back pain and musculoskeletal injuries in factory workers has increased, leading to increased sick leave and reduced quality of life. These workers often perform tasks such as forward leaning, twisting and repetitive lifting of heavy loads that are associated with the risk of back injuries [1]. Therefore, to reduce fatigue, low-back pain, and injuries, there is a need for physical interventions such as braces and wearable devices that are designed to support the lower back and spine. In the context of workplace back support and fatigue relief, the need for ergonomics and effective assistance is critical for widespread adoption. Orthopedic braces are ergonomic, but do not reduce muscle fatigue or spinal loading. Passive exoskeletons and exosuits can also be ergonomic but lack task versatility and high support. Active back exoskeletons and exosuits provide appropriate assistance levels but currently lack ergonomics. Methods of application of assistive forces on the body for these devices include a contractile force parallel to the back extensor muscle, which could increase spinal compression loads, or normal forces on the torso that generate a sagittal moment about the lower back and sometimes hip. Therefore, there is a need for more ergonomic active devices that do not increase spinal compression. This work presents a comparative study of contractile and moment generating exosuits. We implement contractile and bending actuation schemes using cable-driven and inflatable actuators in a wearable interface and perform static and dynamic characterisation. The actuators are sized and operated to ideally have similar performance outputs with minimal ergonomic footprint. The anchoring and actuation efficiency is estimated using a testbench. The anchor and actuation migration and effect on muscle activity is measured in tests on healthy subjects. We choose the most efficient and effective anchor and actuator for future embedding of sensors and intelligence with eventual use as a haptic and assistive support of the lower back and spine. [1] Hoogendoorn, Wilhelmina E., et al. "Flexion and rotation of the trunk and lifting at work are risk factors for low back pain: results of a prospective cohort study." Spine 25.23 (2000): 3087-3092.
15 mins
Examination of different features and machine learning models in mental workload detection
Liucija Svinkunaite, Jörn M. Horschig, Marianne J. Floor-Westerdijk
Abstract: Humans operate effectively when the demands of the imposed task correspond to their mental processing capabilities. Suboptimal mental workload levels might lead to poor engagement or compromise the performance and safety of the operator by increasing the error rate and fatigue. Among the methods used to measure mental workload, wearable functional Near-Infrared Spectroscopy (fNIRS) is a candidate technology to provide the means to continuously and objectively quantify mental workload, that could be readily transferred to a workplace. While the potential of fNIRS to monitor hemodynamic changes induced by mental task has been demonstrated in multiple studies, the optimal feature extraction and selection methods as well as machine learning models for reliable classification between different mental workload levels remain under investigation. In this study, we exploited a variety of features and models to determine which combinations lead to the highest mental workload classification performance. Wearable fNIRS devices were used to acquire hemodynamic response from prefrontal and parietal brain regions of 10 healthy participants performing mental workload task (n-back) in three difficulty levels (1-back, 2-back and 3-back). Features in time and frequency domain were extracted and used to train and test random forest, k-nearest neighbour and support vector machines classifiers. We reported the accuracies of 2-class (task vs rest, 1-back vs 2-back, 2-back vs 3-back, 1-back vs 3-back) and 3-class (1-back vs 2-back vs 3-back) classification using different learning models. In addition, feature selection criteria were introduced to determine a subset of most informative features. By employing coverage of relevant brain areas and investigating a range of features and models, this study contributes to developing effective strategies for mental workload assessment.
15 mins
Neuromechanical model-based closed-loop control of locomotion via muscle reflexes and synergies
Huawei Wang, Massimo Sartori
Abstract: Current real-time neuromechanical model-based (NM) controllers for wearable robotics, such as exoskeletons and prostheses, often use electromyography sensors (EMGs) to detect users’ motion intentions and then provide assistance that proportional to users’ joint torques [1]. Even though, recent studies demonstrate that this control framework has the ability to adapt to variety locomotion conditions [1-2], neither placing surface EMGs nor implanting intramuscular EMGs is appreciated by both patients and healthy users. Moreover, the more sensors the more susceptible the system is to noise and movement artifacts. Computational models based on the concepts of muscle reflexes and synergies have been studied via both experiments and avatar-based simulations [3-4], showing that they can explain complex locomotion, for instance walking on uneven terrains, speed up/slow down, with up/down slopes, and turnings. Our work focuses on modelling the reflex and synergy controls in human locomotion and using them to close the loop within NM models so that EMGs will no longer be needed. In the first stage of our study, we answer the question about whether a common structure of reflex and synergy control models can be directly identified from walking data. To answer it, kinetic and muscle activation data of 5 healthy adults at 4 walking speeds (0.9, 1.8, 2.7, 3.6km/h) were collected. Models of muscle reflexes and synergies were identified from the dataset, starting from a fully connected state-feedback structure. This was done through trajectory optimization with the direct collocation method [5]. To our knowledge, the structures of reflex control models are normally defined manually, and the reflex gains are mostly optimized based on the simulations of avatars. This study enables for first time to identify both the structure of reflex control and its reflex gains directly from human walking data. In addition, combining reflex and synergy together to explain locomotion data have not been examined yet. This first stage study will also provide the computational cost information of such identifications, which can be used as reference for the next study stage where an online calibration tool of muscle reflex and synergy controls will be developed.
15 mins
Comparison of kinematics of imposed head movements and head movements in free space
Anoek Geers, Erik Prinsen, Ralf de Jong, Bart Koopman, Jaap Buurke, Hans Rietman
Abstract: Main research question: The correct positioning of the head and neck is essential during electrical wheelchair use. Therefore, proper head support is important for persons with impaired head stabilization and/or positioning. Existing systems often provide only static, fixed head support, while during daily life support of different head support positions and position changes are needed. Therefore, an adaptive, dynamic head support is being developed by the authors. As design input, the kinematics of the head in free space and during imposed and restricted head movements were studied to define the degrees of freedom and movements that the system should be able to support. Research methods: An observational study (one measurement session) was set up with non-impaired individuals. Besides head movement in free space, imposed passive and restricted active head movements were measured, using a measurement device that allowed the head to rotate around fixed rotation points. Participants performed flexion/extension, lateral rotation, lateral flexion, and a combination of lateral rotation and lateral flexion. Results: A total of nineteen participants were included in the study. For this abstract, the kinematic data of thirteen participants were studied. In flexion-extension, the rotation was isolated with minimal rotation over the other two axes (< 5 degrees). Lateral rotation showed variable flexion/extension and contralateral lateral flexion at the movement extremes. Lateral flexion showed variable rotations over the other axes. For the imposed head movements, especially in flexion-extension but also in lateral flexion there was a mismatch at the movement extremes in which the head lost contact with the measurement device. In lateral rotation, the head and measurement device followed approximately the same path. However, in the combined trial, together with lateral rotation, the head showed contralateral lateral flexion (10-15 degrees). Conclusions: The research results provide input for the movement paths to be implemented in the new head support. In order to approximate the head kinematics in free space and to account for user variability, flexion extension, lateral rotation and lateral flexion movements should not be controlled over fixed rotation points. Rotations should be combined and the translations of the biomechanical rotation axes should also be incorporated.

end %-->