Towards Latent Space Based Manipulation of Deformable Objects Using Autoencoder Models

EasyChair Preprint 11765

Abstract

This research paper explores an innovative approach towards deformable object manipulation by leveraging latent space representations facilitated by autoencoder models. Dealing with the inherent challenges posed by deformable objects in robotic manipulation, our proposed framework integrates autoencoder models to learn and encode latent features that capture the complex deformations and interactions of flexible materials. The autoencoder's ability to compress high-dimensional deformable object data into a meaningful latent space facilitates efficient manipulation planning and control. The study investigates the training of autoencoder models on diverse deformable object datasets, allowing the network to learn robust representations of the underlying physics governing deformations. The outcomes of this research pave the way for more intelligent and adaptive robotic systems capable of handling a wide range of deformable materials in various applications.

Keyphrases: Exploration, latent, space

@booklet{EasyChair:11765,
  author    = {Jane Smith and Patric Bruze},
  title     = {Towards Latent Space Based Manipulation of Deformable Objects Using Autoencoder Models},
  howpublished = {EasyChair Preprint 11765},
  year      = {EasyChair, 2024}}