Gym environments for predictive biomechanical simulations using reinforcement learning.
- Make sure Python 3.9 is installed on your system
- Install the latest version of SCONE
- Optional: for faster simulations and access to all gym environments, be sure to activate the Hyfydy simulation engine inside SCONE. Without Hyfydy, you are limited to using OpenSim, which takes much longer to optimize. More information and a free trial can be found on the Hyfydy website.
- Clone the sconegym repository
- Open a console, navigate to your local sconegym folder and type:
pip install -e .
(the-e
flag will ensure the package automatically gets updated when you update the sconegym repository) - To see if everything works, try out the
example_environment.py
ortest_environments.py
from the sconegym folder - To test sconegym in combination with depRL, try running
example_deprl.py
, after following the instructions on the depRL website
Results of an optimization can be rendered and analyzed in SCONE Studio, using the following steps:
- Open SCONE Studio
- In the Optimization Results pane on the left, navigate to any checkpoint file (extension
.pt
) and double-click the file - A number of rollouts will be performed, and results will be stored as
.sto
files inside the SCONE data folder, in a subfolder below the.pt file
. The name of the subfolder starts withrun_checkpoint
and ends with the checkpoint number. - Double-click on any of the
.sto
files to open and display them - In addition to the 3D renders, results can be analyzed via the Analysis Window, and gaits analysis can be performed via
Tools -> Gait Analysis
- For more information, see the SCONE Website
- Pierre Schumacher @P-Schumacher
- Thomas Geijtenbeek @tgeijten