Improvement of a Multi-Body Collision Computation Framework and Its Application to Robot (Self-)Collision Avoidance

Kurzfassung

One of the fundamental demands on robotic systems is a safe interaction with their environment. In order to fulfill that condition, both collisions with obstacles and own structure have to be avoided. This problem has been addressed before at the German Aerospace Center (DLR) through the use of different algorithms. In this work, a novel solution that differentiates itself from previous implementations due to its geometry-independent, flexible thread structure and computationally robust nature is presented. In a first step, in order to achieve self-collision avoidance, collision detection must be handled. In this line, the Robotics and Mechatronics Center of the DLR developed its own version of the Voxmap-Pointshell (VPS) Algorithm. This penalty based collision computation algorithm uses two types of haptic data structures for each pair of potentially colliding objects in order to detect contact points and compute forces of interfering virtual objects; voxelmaps and pointshells. Prior to the work presented, a framework for multi-body collision detection already existed. However, it was not designed nor optimized to handle mechanisms. This thesis resents a framework that handles collision detection, force computation and physics processing of multi-body virtual realities in real-time integrating the DLR VPS Algorithm implementation. Due to the high number of available robots and mechanisms, a method that is both robust and generic enough to withstand the forthcoming developments would be desirable. In this work, an input configuration file detailing the mechanism’s structure is used, based on the Denavit-Hartenberg convention, so that any type of robotic system or virtual object can use this method without any loss of validity. Experiments to prove the validity of this work have been performed both on DLR’s HUG simulator and on DLR’s HUG haptic device, composed of two DLR-KUKA light weight robots (LWRs).