The repertoire of human manipulation is filled with creative use of contacts to move the object about the hand and the environment. However, in most robotic applications the robot fixes all contact points on the object and do grasping. Reliable robot manipulation beyond grasping is still rarely seen.
The difficulties come from the inevitable modeling uncertainties and disturbances in any robotic system. It's even worse for manipulation systems, because they tend to have multiple possible contact modes. Failure or even physical damage could happen to the execution of a motion plan if the system falls into an unexpected contact mode.
Seeing those difficulties, in this talk I will share our work on robust manipulation by utilizing robot compliance. In the first half I will introduce Hybrid Servoing, a method that computes hybrid force-velocity control to execute a given contact-rich motion plan. In the second half I will introduce Shared Grasping, a geometrical mechanical analysis tool for reliable motion planning. For both methods, we demonstrate repeatable manipulation behaviors in experiments for a variety of problems where grasping is not possible.
Yifan Hou is a PhD Candidate at Robotics Institute, Carnegie Mellon University, working with Professor Matt Mason in the Manipulation Lab. Before that he obtained MS in Robotics degree from Carnegie Mellon University in 2017, and Bachelor's degree from Tsinghua University, Beijing, China in 2015. He has also spent time in Toyota Research Institute and University of Washington. His research focuses on model-based motion planning and control for manipulation problems, especially simple hand regrasping and in-hand manipulation. http://www.cs.cmu.edu/~yifanh/