Training musculoskeletal models to perform different tasks and study the rehabilitation (re-training) process after a neuromotor injury. The models are trained using the Deep Deterministic Policy Gradient (DDPG) algorithm in osim-rl, a Python package that integrates the RL library Keras with the biomechancs simulation software OpenSim.
Developing and implementing numerical methods and algorithms for co-simulation of mechanical systems with contact using reduced-order models for robotics applications. In co-simulation, several subsystems are simulated in parallel and exchanging information at discrete time points. Model-based co-simulation gives a robust prediction of the interface variables between subsystem communication points.
Developing coupling strategies for multirate co-simulation of multi-physics systems with particular interest in hydraulic-actuated machines. The fast dynamics of hydraulic systems requires simulation time-step sizes smaller than that of other mechanical systems. Reduced-order models can give a physics-based prediction of the mechanical system when co-simulating hydraulic subsystems.
Multibody system dynamics interface modelling for stable multirate co-simulation of multiphysics systems [PDF]
Co-simulation of multibody systems with contact using reduced interface models [PDF]
Real-time simulation of wheeled vehicles on soft soil.
Contact is a common phenomenon in the physical world, but simulating systems with contact efficiently can be challenging, especially if we consider friction. Transitions from static to kinetic friction (stick-slip) as well as jamming (as in the Painlevé paradox) are some of the complex phenomena that are hard to capture numerically in simulation.
A multi-scale study of the hip joint mechanics using rigid-body inverse dynamics and finite element analysis
This site is under construction...