Mastering physical interaction with frictional contact is essential to autonomous robotic manipulation and locomotion in a complex, cluttered and dynamically changing environment.
A common manipulation pipeline often consists of two stages. The first stage is to plan a collision free path to a pre-action pose followed by an open-loop action. The second stage is to adjust the in-hand pose with pick-and-place actions or fingertip motions. Contacts with the environment are minimized during the process. In sharp contrast, humans actively embrace environment contacts using compliant motions under rich feedback to guide manipulation, reduce uncertainty, and achieve dexterity.
Similarly, robust treatment of the changing contacts between legged robots and the ground remains a basic challenge for the field. For walking and running robots to successfully navigate complex and uncertain environments, they must intelligently manage and leverage these environmental contacts.
The stiffness and discontinuities inherent in models of impacts and frictional contact create significant problems for many traditional approaches to simulation, planning, and control and planning. These challenges have resulted in robotic approaches that seek to avoid, rather than leverage, contact and that are brittle to any unexpected contact. Many current techniques for both manipulation and locomotion therefore result in motions that are not nearly as dynamic or reliable as human activity.
The workshop seeks to address three major challenges: 1) high-fidelity yet tractable computational solution to rigid body frictional contacts modeling and soft material; 2) optimization-based motion planning without full mode scheduling; 3) control synthesis for contact-rich interaction with visual and tactile feedback.
Topics of Interest
- Beyond open-loop manipulation: closing the loop with visual and tactile sensing
- Non-prehensile manipulation and extrinsic dexterity
- Swing/stance switching control for locomotion
- Data-efficient contact model learning
- Modeling of soft surfaces
- Tactile and non-tactile sensing
- What model complexity is required for planning and control?
- Controlled slip as a resource
- Quasi-static vs. dynamic models of manipulation
- Mechanical vs. algorithmic intelligence
- Hybrid and mixed-integer optimization approaches
- Role of compliance and impedance control
- Todd Murphey, Northwestern
- Jeff Trinkle, RPI
- Aaron Johnson, CMU
- Alberto Rodriguez, MIT
- Kris Hauser, Duke
- Russ Tedrake, MIT
- Ludovic Righetti, Max Planck Institute
- Nathan Ratliff, Lula Robotics
- Jeannette Bohg, Max Planck Institute
Jiaji Zhou, The Robotics Institute, Carnegie Mellon University
jiajiz (at) cs.cmu.edu
Matthew Mason, The Robotics Institute, Carnegie Mellon University
matt.mason (at) cs.cmu.edu
Michael Posa, MEAM and GRASP Lab, University of Pennsylvania
posa (at) seas.upenn.edu