Variable Impedance Robotic Skills: From the Foundations to the Current Challenges and Perspectives

  • Thiago Boaventura Cunha (Järjestäjä)
  • Fares Abu-Dakka (Järjestäjä)
  • Matteo Saveriano (Järjestäjä)
  • Meghan E. Huber (Järjestäjä)

Aktiviteetti: Työpajan, paneelin, session, tutoriaalin tai tapahtuman järjestäminen

Description

To have robots safely and robustly interacting with their surroundings and cooperating with people, they must present some degree of mechanical compliance. While such compliance can be achieved passively, using elastic and flexible elements in the robot's mechanical design, control algorithms are able to shape the robot's mechanical compliance (e.g., stiffness, damping, inertia) in a much more versatile manner. A well-known approach to control physical interaction is via impedance control, where it is possible to control the robot's mechanical impedance (i.e., the dynamic relation between force and velocity) at a given interaction point. In this context, there are several fundamental research problems, such as how to design control systems that have to work with potentially uncertain environment models and uncertain sensory feedback, deal with unpredictable and complex interactions, and adapt in real-time.

To deal with such uncertain physical interactions, it is possible to use variable impedance control (VIC) as a feasible solution to overcome position uncertainties and subsequently avoid large impact forces, since robots are controlled to modulate their motion or compliance according to force and visual perception. However, knowing what the desired robot impedance should be for a given scenario is still an open question. In addition, it is desirable to avoid hard-coding such skills in the software. In this direction, machine learning has been providing suitable tools to learn VIC skills, e.g. from human demonstrations (Learning from Demonstration) or through exploration. Another current trend to approach this uncertainty problem during physical interaction is using model predictive control.



This proposal aims to be the third event on this topic of Variable Impedance Learning and Control. The first one was organized at IROS18 and the second one at IROS21. Both were very successful and had about 70 attendees on average. This time, for RSS, we proposed a hybrid tutorial/workshop that aims at, firstly, introducing this topic of physical interaction control to junior researchers and students in a tutorial-like way. The very basic theoretical foundations on this topic will be presented, followed by hands-on activities, as described in the next section. This half-day tutorial will give this junior audience the basic background needed to better understand the second part of the workshop, where different experts will be brought together to discuss the current challenges and perspectives of this field in robotics. The main intended audience of this second-half part of the workshop is more senior researchers and students that already work on this topic, as the talks will be more detailed and focused. Special attention was given to providing a diverse spectrum of prestigious speakers, from different seniority, gender, ethnicity, as well as geographic location, as detailed in the next sections.
Aikajakso27 kesäk. 2022
Tapahtuman tyyppiWorkshop
Konferenssinumero3
SijaintiNew York, Yhdysvallat, New YorkNäytä kartalla
Tunnustuksen arvoInternational