TimewarpVAE: Simultaneous Time-Warping and Representation Learning of Trajectories
This work addresses the challenge of data scarcity in complex tasks like dexterous manipulation by providing a method to learn disentangled representations from trajectories, though it is incremental as it builds on existing VAE and DTW techniques.
The paper tackled the problem of learning efficient representations from human demonstration trajectories by factoring out timing variations from spatial characteristics, achieving lower spatial reconstruction test error than baselines and enabling generation of semantically meaningful novel trajectories for robotic tasks.
Human demonstrations of trajectories are an important source of training data for many machine learning problems. However, the difficulty of collecting human demonstration data for complex tasks makes learning efficient representations of those trajectories challenging. For many problems, such as for dexterous manipulation, the exact timings of the trajectories should be factored from their spatial path characteristics. In this work, we propose TimewarpVAE, a fully differentiable manifold-learning algorithm that incorporates Dynamic Time Warping (DTW) to simultaneously learn both timing variations and latent factors of spatial variation. We show how the TimewarpVAE algorithm learns appropriate time alignments and meaningful representations of spatial variations in handwriting and fork manipulation datasets. Our results have lower spatial reconstruction test error than baseline approaches and the learned low-dimensional representations can be used to efficiently generate semantically meaningful novel trajectories. We demonstrate the utility of our algorithm to generate novel high-speed trajectories for a robotic arm.