Robotics and Biology Laboratory

Experimental Validation of Contact Dynamics for Prehensile Pushing

Images

People

Roman Kolbert
Alberto Rodriguez
Oliver Brock

Motivation

Most models in contact dynamics show some unrealistic behavior due to assumptions that were made for the  sake of computational convenience. Unfortunately, there is a lack of experimental work to validate these assumptions and to evaluate how realistic these contact modeling approaches are, which is the purpose of this thesis.

Description of Work

The application of choice is part of an interesting manipulation problem, which is called Prehensile Pushing. Prehensile Pushing is a form of In-Hand manipulation, that exploits the environment to change the orientation and position of a grasped object inside the gripper. An experimental setup was developed to measure forces and torques at all contacts as well as the motion of a grasped object in three dimensions while autonomously performing three different Prehensile Pushing actions - linear pushing, pivoting and rolling.

Results

Kolbert validated a variety of assumptions used in state of the art contact dynamics by examining results of Prehensile Pushing experiments. These assumptions were applied to the experimental data and validated their predictions. Some of these assumptions proved to be valid only to some degree. The thesis shows that the Law of Dry Friction by Coloumb is one of the weakest assumptions, although it is used in many contact models. Kolbert found that the Prehensile Pushing Model predicts forces, torques, and motions to first order degree while being computationally more expensive and the model used in MuJoCo predicts the behavior only to some degree while being computationally cheap. Kolbert also found that the variability of local forces within runs with the same parameters limits the predictability of these forces and questions the uniqueness of a valid solution.