Deformable Object Manipulation

Robotic manipulation of deformable objects presents several challenges that are not present in rigid objects. The configuration space of deformable objects is high dimensional and their dynamics are highly affected by the material properties, making even more complex observing the state or deformation of the object. Whereas for rigid objects their position and orientation is usually enough to describe their location in the world, deformable objects do not have a simple representation available, which requires for more complex representations to manipulate the material.

In this project we research on how to manipulate more efficiently deformable objects by using dynamic manipulation as well as the modeling deformable objects via graph structures. Our applications range from manipulation of granular materials such as ground coffee to cloth manipulation.

People involved

  • David Blanco-Mulero (david.blancomulero(at)aalto.fi), PhD candidate.
  • Gökhan Alcan (gokhan.alcan(at)aalto.fi), Postdoctoral researcher.
  • Ville Kyrki (ville.kyrki(at)aalto.fi), Professor, group leader.

Project updates

Master Thesis on “Model-based Dynamic Manipulation of Deformable Objects”

The goal of this thesis is to develop model-based approaches for the dynamic manipulation of a deformable object. Particularly, the focus will be on the task of throwing a lasso around a target (a bollard) with a robotic arm.

Master Thesis on “Adversarial Robotic Cloth Manipulation”

This thesis will focus on how to use adversarial learning for improving existing methods in robotic cloth manipulation.

Our work “Learning Visual Feedback Control for Dynamic Cloth Folding” was accepted to IROS 2022!

We are happy to announce that our work “Learning Visual Feedback Control for Dynamic Cloth Folding” was accepted to IROS 2022 and nomitated to both the IROS Best Student Paper award and the IROS Best RoboCup Paper Award.

Manipulation of Granular Materials by Learning Particle Interactions

In this work we propose to use a Graph Neural Network (GNN) surrogate model to learn the particle interactions of granular materials. We perform planning of manipulation trajectories with the learnt surrogate model to arrange the material into a desired configuration.

Evolving-Graph Gaussian Processes poster at the Time Series Workshop at ICML 2021

This work extends the current SotA of Graph Gaussian Processes (GGPs) to dynamic graphs and asses the performance of the proposed evolving-Graph Gaussian Process (e-GGP) in two simulated tasks where deformable objects are represented as a graph that evolves over time.