Planning and control for simulated robotic Sandia hand for the DARPA Robotic Challenge

Abstract

The DARPA Robotic Challenge (DRC) required the development of user interface, perception, and planning and control modules for a robotic humanoid. This paper focuses on the planning and control component for the manipulation qualification task of the virtual section of the DRC. Nonlinear algorithms were employed for the planning systems, such as the grasp optimization system and the robot state trajectory computation system. However, for closed-loop control, a linear proportional-derivative (PD) joint position controller was used. The nonlinear algorithms used for the planning systems may be improved, but their current functionality allows the successful completion of the manipulation qualification task. Also, even though PD controllers seem appropriate for the closed-loop control, PID controllers might yield a higher level of accuracy if tuned properly. In conclusion, a linear controller appears sufficient for certain control of the highly nonlinear ATLAS humanoid robot and Sandia hand as long as accurate optimization and planning systems complement such control.