Lower-limb Wearable Exoskeleton
Different types of assistive controllers and intelligent algorithms are under development and implementation on wearable exoskeleton.
This controllers vary from a feedforward controller with subject-specific models of joint impedance, to variety of model-predictive and adaptive controllers and includes biologically inspired neuromechanical models of walking.
The controllers have intention decoder blocks to improve robot adaptation.
Subjects (SCI and healthy controls) will train with the exoskeleton and controllers in several repetitive sessions, which enables us to investigate the co-adaptation and improve in walking performance.
​
see these as examples:
Our robust gait phase estimator
Our controller based on ref. trajectory adaptation
Our virtual energy based controller
​
Prediction and optimizing blood pressure
using a wearable compression device
We are developing advanced machine learning techniques for reliable and accurate prediction of blood pressure both for long term cuffless blood pressure monitoring and optimizing the blood flow using a wearable compression device developed by Fluid Flow Physics Group at the University of Waterloo.
We used machine learning techniques to optimize the blood flow for each individual wearing the device.
Neuromechanical Modelling and Control
Robotic pushers and exoskeletons are used to perturb subjects both during walking and in isometric conditions. The perturbed kinetic data are benchmarked against unperturbed data. The way residual torque and kinematics relate can be modeled as variable mechanical joint impedance that has a significant role in control of our movements and interaction with the external world. Those residual data are fed into an optimization process to obtain a parametric model of joint impedance.
This process are underwent for subjects with neurological conditions such as incomplete SCI, and healthy control subjects.
Identified joint impedance models can be validated using other perturbation experiments and directly used in design of assistive controllers.
Neuromuscular Modelling and Inverse Optimal Control
We use Deep Reinforcement Learning and Inverse Optimal Control techniques with detailed musculoskeletal models to discover the underlying mechanisms and set goals for movement.
We also develops efficient inverse optimal control techniques
allowing effective estimation of optimal cost functions.
​
see these examples:
our Adaptive-reference IOC technique
our DRL-Curriculum Learning with Musculoskeletal models
https://twitter.com/Arash_Arami/status/1538914595036876814?s=20&t=gnD5yH4XIiAN8-U5ceM7Jw
Spasticity Modelling
The goal of this study is to build simple wearable tools together with modelling techniques to quantitatively evaluate and model spasticity in variety of subjects.
Spasticity is an involuntary velocity-dependent increase of tonic stretch reflexes.
In this study we use 3D printed handles instrumented with force sensors, together with wearable IMUs and EMG sensors to build a model based on tonic stretch reflex thresholds and how they change with change of velocity.
AI-driven manufacturing: from data to optimal forging and casting
In a close collaboration with Fatigue and Stress Analysis Lab (led by Prof. H. Jahed) and Prof. S. Lambert and Prof. M. Wells, we are shaping the future of the manufacturing by using the power of AI and machine learning in prediction of manufacturing performance outcome and optimizing it.
Smart Endoprostheses
The goal is to design implantable instrumentation to estimate clinically meaningful measures related to the function of a prosthetic joint. Variety of sensors for both kinematics estimation and joint contact forces have been used along with the data-driven models to provide an accurate estimate and diagnose any fault with the prostheses.
Parkinson's Disease Modelling
The goal is to build models for sensorimotor deficits for patient suffering from Parkinson's disease. These models can predict the motor disorders, such as balance deficiency, rigidity, bradikynesia and tremor can be used to evaluate and monitor the progression of the disease.
We are also developing predictive models of freezing of gait in PD subjects which allows us to anticipate the freezing events and enable some preventive actions.
​
​
​
​
​
