IPSA - Inventor Physical Modelling API for Dynamics Simulation in Manipulation

When studying the dynamic, physical interaction of robots with the environment a generic physics simulator becomes most important when robot and environment exceed a certain complexity. Investigation of current problems in robotics, e.g. in multifingered manipulation or two-legged walking requires simulations which reflect physical details of the real robot system in a virtual scene. The development of suitable control systems for complex manipulators or walking robots in simulation may save a lot of time, otherwise spent for operating the system, and prevents unintented damage on a costly robot system. Further, the application and benefit of using specific sensors may be investigated by physical modelling of a robot system, e.g. when investigating the ideal placement and required resolution of tactile sensors. Finally, model-based simulation allows to distribute models of expensive robot system among researchers, which means saving costs directly. Dynamic simulations have attained focus much earlier in other domains as in development of automotive and space vehicles and for creating realistic computer animations. Consequently, a lot of commercial and non-commercial dynamic simulation frameworks have evolved. A reference multipurpose modelling and simulation system, not only in robotics, is Matlab/Simulink [1]. By using additional extensions to Matlab/Simulink such as SimMechanics, the VR Toolbox or The Robotics Toolbox [2] a lot of mechanisms are provided for simulating robots, their control systems and environmental dynamics. Yet, the assembly of a detailed virtual simulation scene from these building blocks is cumbersome and often takes a considerable portion of time compared to the whole investigation period. Also for the purpose of developing robot controllers in simulation Microsoft has made its Microsoft Robotics Studio (MSRS) available to the public [3]. This IDE provides a comprehensive software framework aiming for rapid development of robot control programs with unified interfaces and deploys the PhysX physics engine by NVIDIA for simulation of the virtual scene. Although MSRS appears quite powerful, the construction of a complex scene is not intuitive and requires different models for the physical and the visual domain. Further, the usability of the framework is limited to Microsoft operating systems. A quite useful robot simulator for developing manipulation and grasp planning algorithms was introduced with GraspIt! in [4]. This software provides a robot library with several robot hand models and a robot arm model. The simulator may be invoked via an interactive UI or via a TCP/IP interface for external applications. For static grasp analysis a collision detection function and consideration of friction forces were integrated and used to compute grasp quality measures. Later in [5] a dynamics simulator was added for simulation and investigation of dynamic grasping tasks. Specification of the physics model data such as joint types and location or inertia matrices must be provided by the user in addition to the visualization model. But by the authors practical experience these specifications were hard to tune to achieve a stable simulation behavior.