System identification to model limb biomechanics: Toward a re-evaluation of joint stiffness and viscosity
Files
Ninite_22222000_2025.pdf
Embargoed access from 2027-09-01 - Adobe PDF
- 4.74 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Accurately estimating the biomechanical properties of the human arm, in particular joint stiffness and viscosity, is essential for understanding how the central nervous system controls upper limb movements. Arm dynamics results from both neural inputs and intrinsic biomechanical properties. Separating these two contributions is therefore necessary to build reliable models of motor control. However, most existing studies estimating stiffness and viscosity do not explicitly distinguish these contributions. Recent evidence suggests that standard estimates may be overestimated due to contamination by neural feedback. This motivates the need to re-evaluate current estimates and develop approaches that better isolate biomechanical parameters from neural contributions. In this thesis, we propose a system identification approach that relies on data obtained from reaching movements to re-assess previous estimates of stiffness and viscosity. In particular, we apply the Prediction Error Method to both one-joint and two-joint biomechanical models of the upper limb. We provide a closed-form expression of the estimates for the one-joint case and express them as the solution to a linear system of equations in the two-joint case. In addition, the estimates are shown to be globally optimal. Our methods are tested on both one-joint and two-joint synthetic reaching data, generated using a state-of-the-art control framework, a stochastic optimal controller. We apply various system identification validation techniques and determine task conditions that improve estimation quality, which could inform the design of experimental protocols. Future work will focus on developing such experiments and applying our methods to the corresponding data.