RoMaDO-RA

Statement of Objectives

Deformable objects are encountered in a wide variety of industrial, service and health applications. However, robotizing tasks that involve deformable objects requires specific contributions, as the behavior of such objects affects all aspects of a robotic system: perception, actuation, planning and control. In particular, it seems to the organizers that planning, control and perception require significant contributions, since the usual tools that apply for rigid body manipulation in those fields cannot be used anymore. Thus, systems for deformable object manipulation often:
• cannot handle a large number of use cases, nor variabilities that can appear in such use cases;
• are unable to perform refined tasks where interactions, contacts and complex deformation patterns occur.

Hence, we propose to gather works on planning, control and perception for deformable object manipulation, with the intent of bringing agility and performance to robotic systems performing real-world applications. The focus is especially put on:
1. Low-level or trajectory planning: usual deterministic and probabilistic planning methods cannot be used directly for deformable object manipulation given the difficulty to take the state of the object into account. We search here methods to mitigate the infinite dimensionality curse, take into account the possible interactions and ensure the physical integrity of the manipulated object.
2. High-level or task planning: tools are lacking to describe and replicate manipulation tasks that rely on object deformation, as well as to generalize task analysis to multiple use cases. Promising solutions to achieve higher autonomy revolve around manipulation process analysis, learning from demonstration, manipulation pipeline generalization, planning problem formalization or agile software and hardware architectures.
3. Advanced deformation control: the high nonlinearity introduced by deformation makes the formulation of a control problem more complex. In the literature, deformation control revolves strongly around the perception system and the behaviour of the object. Main challenges consist in closing the control loop to achieve higher performance and take into account variabilities of deformation patterns due to object behaviour, external interaction or contacts.
4. Dedicated perception systems: due to limits in simulation capacity and modeling precision, deformation sensing becomes almost mandatory for correct estimation of the state of the manipulated object. To go forward on this topic, we need to address practical implementation of multi-modal sensing and find methods for deformation state estimation from low amounts of data.

The aims of the workshop are twofold. Firstly, we want to propose to the community a detailed review of state-of-the-art contributions in terms of planning, control and perception for deformable object manipulation. To do so, we will mix keynote presentations from recognized researchers and interactive presentations that highlight recent and innovative results, both with a focus on real applications. Secondly, we wish to enable communication and dissemination within the community. To accomplish this, we propose a program where presentation and discussion times are balanced, and where neighboring fields such as soft robotics and medical robotics can intervene. With regards to dissemination, our goal is to ensure online availability of all material (presentations and discussions, but also multimedia files attached to contributions) and to contribute to an upcoming special issue in RA-L.