Design, Dynamics, and Control of Active-passive Upper-limb Exoskeleton Robots
MetadataShow full item record
Modern power assistive technologies, neuromuscular rehabilitation engineering, and haptic research have all entered into a new age thanks to wearable exoskeleton robotics. Exoskeleton robots, which range from fully passive to fully active-assisted movements, have become an important instrument for assisting industrial employees and stroke rehabilitation therapy. Further technological advancements in this realm should continue to increase the upper limb functionality needed for activities of daily living (ADLs). Although exoskeleton technology is rapidly developing, interdisciplinary research is still required to address technological issues such as designing efficient power sources, evaluating human-robot interaction, estimating user intent, and developing control systems. This thesis presents: 1) a scalable musculoskeletal (MapleSim Biomechanics) model for the dynamic simulations of movement, 2) an advanced high-fidelity human-exoskeleton multibody biomechatronic model, 3) a novel design for an efficient active-passive power source for wearable portable robots, 4) a human intent machine learning estimator using surface electromyography (sEMG) biological signals as input, 5) an efficient assist-as-needed (AAN) hierarchical control scheme of upper-limb power assist exoskeletons, and 6) a short-term experimental human-in-the-loop (HITL) evaluation of sEMG-based machine learning-driven hierarchical control of active-passive exoskeleton. A musculoskeletal (MSK) model is a valuable tool for assessing complex biomechanical problems, estimating joint torques during motion, optimizing motion in sports, and designing exoskeletons and prostheses. This study proposes an open-source upper body MSK model that supports biomechanical analysis of human motion. The MSK model of the upper body consists of 8 body segments (torso, head, left/right upper arm, left/right forearm, and left/right hand). The model has 20 degrees of freedom (DoFs) and 40 muscle torque generators (MTGs), which are constructed using experimental data. The model is adjustable for different anthropometric measurements and subject body characteristics: sex, age, body mass, height, dominant side, and physical activity. Joint limits are modeled using experimental dynamometer data within the proposed multi-DoF MTG model. The model equations are verified by simulating the joint range of motion (ROM) and torque; simulation results have a good agreement with previously published research. As an alternative to static optimization for resolving the redundancy issue in inverse muscle models, we proposed a new machine-learning method. A recurrent neural network (RNN) was optimally configured, trained, and tested to estimate the pattern of muscle activation signals. Five biomechanical variables (joint angle, velocity, acceleration, torque, and activation signal) were used as inputs to the RNN, while the outputs were sEMG signals. The proposed approach permits a fast and direct estimation ability instead of iterative solutions for the inverse muscle model. It raises the possibility of integrating such a model in a device for functional rehabilitation and sports evaluation devices with real-time estimation and tracking. This method provides clinicians with a means of estimating sEMG activity without an invasive electrode setup. Designing the exoskeleton actuation is challenging and time-consuming due to closed-kinematic loops in the 3D human-exoskeleton multibody model, complicated interactions, and interdependent selection of power transmission features. This research proposes a process for dynamic syntheses of passive and active assistive shoulder exoskeletons. First, a multibody model was developed using six components: an upper-body musculoskeletal model, an optimal controller, the exoskeleton’s structure/mechanism, a passive mechanism, a powered actuator, and an assistance model. The system design was optimized by choosing features of the passive mechanism and exoskeleton motor such that the human joint active torque, power, muscle metabolic energy expenditure, and actuator electricity consumption were minimized. The resultant optimized active-passive exoskeletons allow for the creation of lighter and smaller wearable robots that reduce the user’s muscular activation torque for the tasks being studied. To estimate the human intent, motion, and force/torque in the high-level controller, a machine learning model for the regression of interrupted sEMG signals to future control-oriented signals (e.g., robot’s joint angle and assistive torque) of an active biomechatronic device is proposed. The regression models, collectively known as MuscleNET, take one of four forms: artificial neural network (ANN), RNN, convolutional neural network (CNN), and recurrent convolutional neural network (RCNN). Inspired by conventional biomechanical muscle models, delayed kinematic signals were used along with sEMG signals as the machine learning model’s input. The results indicated that the RNN (with filtered sEMG signals) and RCNN (with raw sEMG signals) models, both with delayed kinematic data, can extract underlying motor control information (such as joint activation torque or joint angle) from sEMG signals for pick-and-place tasks. The CNNs and RCNNs were able to filter raw sEMG signals. The RNN with a 0.1-second prediction horizon could predict the control-oriented joint angle and assistive torque with 92% and 86.5% regression accuracy, respectively, on the test dataset. This proposed approach permits a fast, predictive, and direct estimation of control-oriented signals, instead of an iterative process that optimizes assistive torque in the inverse dynamic simulation of a multibody human-robot system. Training with these interrupted input signals significantly improved the regression accuracy in the case of sEMG signal disconnection. This robust predictive control-oriented model (RobustMuscleNET) can support volitional high-level myoelectric-based control of biomechatronic devices such as exoskeletons, prostheses, and assistive/resistive robots. The RNN model with filtered sEMG and delayed kinematic signals is particularly appropriate for applications in musculoskeletal simulation and biomechatronic device control. A closed-loop human-robot system requires the development of an effective robotic controller that considers models of both the human and the robot, as well as the human adaptation to the robot. Doing so requires optimal coordination with the central nervous system (CNS) that manages the MSK. To develop an AAN upper-limb controller and assess human-robot interactions, we compared the AAN-proportional rule, AAN-computed-torque method (CTM), AAN-fuzzy logic rule, and AAN-model predictive control (MPC) frameworks. The human-robot adaptation was simulated using a nonlinear model predictive control (NMPC) as the human CNS for the three conditions of initial (the initial session of wearing the robot, without any previous experience), short-term (the entire first session, e.g., 45 minutes), and long-term experiences. The results showed that the desired strength of the robot should be increased gradually to suppress unexpected human-robot interactions (e.g., robot vibration, human spasticity). Although both controllers derived the necessary torques for following the target shoulder joint trajectories and desired assistance strength, from experimental testing with a human participant, model-predictive control was able to outperform the CTM control in the presence of a substantial control loop delay, but at a greater computation cost. This research studied exoskeleton control challenges, developed a test framework for evaluating a new active-passive shoulder exoskeleton, and proposed an sEMG-based human-robot cooperative control method to carry out the wearer’s movement intentions. The active-passive assistance was compared with fully-passive and fully-active exoskeletons using the following criteria: 1) post-test personal survey, 2) load tolerance duration, and 3) computed human active torque, power, computed metabolic energy expenditure using sEMG signals and inverse dynamic simulation. The experiments outcomes showed that active-passive exoskeletons required less muscular activation torque (50%) from the user, and reduced fatigue duration indicators by a factor of 3, compared to fully-passive ones.
Cite this version of the work
Ali Nasr (2022). Design, Dynamics, and Control of Active-passive Upper-limb Exoskeleton Robots. UWSpace. http://hdl.handle.net/10012/18985