Stochastic Object Pose Calculation for a Space Robot Manipulator
Exploring other planetary sufaces with autonomous robots is a very complex task and requires a lot of research. To give an impulse, the DLR organized the first SpaceBot Cup in 2013. The team Berlin Rockets from the Freie Universität constructed a six-legged robot to take part in the competition. This thesis presents an approach for the stochastic estimation of an object's pose. The estimate will be used by the robot to grasp the object. The estimation is done with a particle filter that uses depth and color data from the robot's Kinect for WindowsTM. The object models consist of 3D point that are evenly distribued over the object's surface. Additionally, a few specific pixels from the vicinity of the object are also taken into account to reduce the rate of false positives. To test the particle filter, several test cases were conducted. The accuracy of the results is unfortunately very low, which shows that the measures taken did not have the desired effect.