Statement of Objectives

Deformable objects are pervasive in applications across sectors including industry, agriculture, and healthcare. Having robots that could reliably manipulate such objects could yield significant benefits. However, the robotic handling of soft objects in dynamic and unstructured environments poses a challenging research area spanning perception, planning, control, machine learning, and human-robot interaction. Thus, we propose a workshop dedicated to deformable object manipulation given the vast potential for robotic systems and current unsolved research questions. In particular, the workshop focuses on:

  • Grippers/tools design and grasping for manipulation of deformable objects
    • Grasping of deformable objects can be found in different health and industrial applications. This is a very challenging task due to the non-rigidity properties of the objects. This imposes challenges from different aspects such as gripper and tool design, object grasping, and how to control the grasping points during the manipulations. Some answers exist in the literature composed of having prior knowledge of the deformed object or/and designing a specific gripper or tool. However, these solutions will have some limitations to the manipulation of deformable objects in general.
  • Data management of deformable objects
    • Data management is an important aspect of robotic control. It can be used in different approaches such as perception and learning. There are few benchmarks for evaluating and comparing different approaches for deformable object manipulation. In this topic, we highlight the importance of benchmarking where these data are the base for other manipulation topics such as learning techniques (sim2real) to explore the challenges of deformable objects.
  • Machine learning for deformable object manipulation (Perception and Control)
    • Deep learning from vision and touch data can be used for recognizing physical features and surface reconstruction of manipulated objects to determine how the robot should grasp the object or plan for the manipulation actions. In addition, deep reinforcement learning and learning from demonstration approaches have enabled the manipulation of a wide variety of deformable objects. Finally, modern machine learning now relies heavily on "foundation models" which have been trained on broad data for use in many downstream application tasks, some of which include deformable object manipulation. Machine learning is undoubtedly a very promising tool in the highly uncertain and complex scenarios of deformable object manipulation.
  • Deformable object manipulation educational tutorial
    • There is a rapid cycle of technology development and innovation in soft robotics and the manipulation of deformable objects. In parallel, there is a need for robots to interact physically with unstructured environments including deformable objects. In this workshop, an educational tutorial will be given on recent advancements in soft robots.

The workshop will present current advances and contributions in deformable object manipulation. A panel of recognized experts will present and discuss their recent and innovative research results. In addition, we propose an intervention between experts close to the subject to better understand the communication and dissemination within the community.
Akin to ROMADO’s previous workshops and to prompt discussion of current and future research, paper submissions will be requested, and a poster session will be organized for accepted papers. Furthermore, all materials (presentations and discussions, but also experimental data) will be available online.