Perceptionbased Intelligent Materialhandling in Industrial Logistics Environments

Abstract

In response to increased costs in logistics, the degree of automation is to be significantly increased over the next few years. Therefore, in this publication the logistics use case, and thus the challenges of the field of application are first discussed. Subsequently, various possibilities for determining gripping points for robots will be addressed. These can be subdivided into holistic and modular approaches. Due to the improved comprehensibility as well as more influencing variables and better-developed fundamentals, a modular approach was chosen for the algorithm presented here. The 3-part algorithm first locates the position of all objects of the relevant class in space. The system then cuts out the objects that are not process relevant and the objects that are not execution-relevant. Finally, by applying a filter cascade to the depth image of the remaining object, the challenging surface of the container is searched for suitable gripping surfaces for the vacuum suction pad used. This algorithm is then implemented in a depalletizing robot and tested in the real environment under various scenarios. Under difficult industrial conditions (all containers dirty, with labels or even damaged) 72 % of the material flow could be successfully depalletized. The error rate is made up of approximately 10 % process error detections and 20 % incorrectly determined gripping points under more stringent industrial conditions. These instructions give you basic guidelines for preparing camera-ready papers.