A time-consuming configuration is necessary to find fast and feasible robot movements and prevent collisions in a bin picking unit. With the simulation software RoboDK a real bin picking unit is emulated and two approaches are tested. The first approach simulates various paths of movement for each object and chooses the best one. In the second approach an artificial neural network is used. A multilayer perceptron without shortcut connections represents the implemented structure of the network. Both approaches are implemented, tested and analyzed. The brute force approach provides for each object a robot path if one exists. The more movements are simulated the more likely the best possible path is found. A trained neural network will always provide a solution too. But in contrast to the simulation this result does not have to be a feasible robot path. Therefore, the analysis of the output from the neural network is an important part.

Studienbetreuer: Thierry Prud’homme