Machines & Dance

Project Overview

With this project, I am trying to generate novel dance choreography using a recurrent neural network (RNN) and motion capture data.

The basic foundation of a phrase of choreography or a small sequence of human movement is the change of a position in 3D space. What did this remind me of? Vectors! (Really complicated ones...)

I wanted to see if machines can create choreography in ways that a human may not think of. To do this, I am using deep learning techniques by implementing RNNs and LSTMs. I define an RNN with output nodes of weights, biases, and tensor placeholders. With these, a n-element sequence of inputs is generated to be fed into the LSTM. While I am currently testing with motion capture data from the CMU Computer Graphics Project, I plan to gather enough motion capture data of myself dancing to train the network on my own movement.

Primary Research Topics/Keywords:

Deep Learning

Machine Learning

Recurrent Neural Network

Long Short-Term Memory Networks

Natalie Monger