Master Thesis: arm movement prediction from visual data
We have all the methodologies in house to do movement prediction based on visual data. Using either our stochastic recurrent neural network (STORN) our our DMP-enhanced autoencoder neural networks (AE-DMP) we would like to use convolutional neural networks (cNN) to do the following:
- translate movement in pixel domain to movement in latent space, possibly using simulated data;
- translate that movement in latent space to movement in Cartesian / joint / ... space;
- generalise these methodologies from one arm to two arms;
- generalise these methodologies among arms.
We prefer candidates who are familiar with probabilistic machine learning. An affinity with deep learning is surely required. The work will involve applying our models in a python environment.