Team member: Junhyek Han, Jaehyung Kim
Humans possess remarkable skills in non-prehensile manipulation techniques such as pulling, pushing, dragging, and tumbling, which enables them to handle objects that are hard or impossible to grasp. However, most of the research focused on pick-and-place manipulation and fail in situations where grasping is not an option due to physical or geometric limitations. Our objective is to equip robots with the ability to manipulate objects even in such circumstances, using a basic gripper and an RGB camera. Project pages: https://sites.google.com/view/nonprenehsile-decomposition?usp=sharing
https://sites.google.com/view/contact-non-prehensile
Our paper: Pre- and Post-Contact Policy Decomposition for Non-Prehensile Manipulation with Sim-to-Real Transfer
Contact_rich_manipulation_paper (1).pdf
Simulation to real-world transfer
Training in simulation (NVIDIA Isaac Gym [2])
The affordance and the optimality of the motion are determined by the environment and object configuration. Thus, to manipulate the object efficiently, the robot must know what kinds of motions are possible and required in the current circumstances. For example, to grab the object far away the robot must have to drag the object nearby and grasp the object afterward. To put an object in a small hole, the robot should pick and place the object near the hole, reorient the object to the right configuration, and push it in the right direction. These motions require both prehensile and non-prehensile motions, and the ability to intelligently combine them will enable a robot to execute various tasks.
Currently, one trained policy can handle objects with a similar shape only in a fixed environment. So handling various environments and general objects requires different policies for each task.