Project Type:
MFA Student Work
Focus:
Machine Learning
Software Used:
Houdini
Date:
Fall 2025
Description:
A realtime muscle deformer I made for a research project in my masters program. Uses a locally trained machine learning model.
256 Short Muscle Simulations
To train the model I created a small data set of 256 arm poses using Houdini’s new Otis muscle solver. To the right is one of those simulations.
I automated this process using a Houdini’s PDG tools.
Preprocessed Data
In order to train the model efficiently I became familiar with principled componant analysis to simplify the muscle diplacement data from the simulations.
This meant following the PCA workflow outlined by side effects for simplifying the displacement data and the pose serialization node for preparing the skeleton pose data.
(PCA analaysis results depicted on the left)