This repository contains the code to implement a predictive simulation of a planar rigid body avatar performing a backward tuck somersault. The predictive simulation is implemented using stochastic optimal control. The goal of this study was to compare the following implementations:
- OCP: deterministic optimal control problem with open-loop control
- SOCP: stochastic optimal control problem with open-loop control, direct feedback control based on proprioceptive and vestibular information affected by random sensory noise of predefined magnitude, and random motor noise of predefined magnitude
-
$\text{SOCP}_{\text{VN}}$ : stochastic optimal control problem with open-loop control, direct feedback control based on proprioceptive and vestibular information affected by random sensory noise modulated by the head angular velocity, and random motor noise modulated by the joint torque actuation -
$\text{SOCP}^{\text{AF}}$ : stochastic optimal control problem with open-loop control, direct feedback control based on proprioceptive and vestibular information affected by random sensory noise of predefined magnitude, anticipatory feedback control based on vestibular and visual information affected by random sensory noise of predefined magnitude, and random motor noise of predefined magnitude -
$\text{SOCP}_{\text{VN}}^{\text{AF}}$ : stochastic optimal control problem with open-loop control, direct feedback control based on proprioceptive and vestibular information affected by random sensory noise modulated by the head angular velocity, anticipatory feedback control based on vestibular and visual information affected by random sensory noise modulated by the head angular velocity and gaze orientation, and random motor noise modulated by the joint torque actuation
This work has been submitted. TODO: add ref to the paper when it is accepted.
| Type | Status |
|---|---|
| Zenodo |  |
In order to run the code, you need to install the environment.yml
- Install Bioptim from source pyomeca/bioptim.
You will then have to navigate to the commit
b0d8f43990c7a717600c6fb9387a6686e6244f0d(i.e.,Stochastic_tagfrom EveCharbie's fork), which is the version the results were generated with. - Install libhsl.so from hsl. You will first need to request a license (free for academics).
main_DMS.py: This script runs all predictive simulations (watch out for the large computational time).plot_optimal_solutions.py: This script generates the solution analysis and figures.
Do not hesitate to contact me if you have any questions or comments about this work eve.charbie@gmail.com.