Open Access

Thomas Wendt, Sinan Süme, Kaithavalappil Ajay Amal

• In this paper an approach to enable mobile robots to localize and position correctly on special machines based on their unique fingerprint in manufacturing is introduced. This is achieved using a single RGB-D camera and the use of Convolutional Neural Networks, i.e. Mask-RCNN. The utilized network is trained on the individual shape of the machine to recognize, classify, and segment it, so theIn this paper an approach to enable mobile robots to localize and position correctly on special machines based on their unique fingerprint in manufacturing is introduced. This is achieved using a single RGB-D camera and the use of Convolutional Neural Networks, i.e. Mask-RCNN. The utilized network is trained on the individual shape of the machine to recognize, classify, and segment it, so the mobile robot can locate the correct machine without any further modifications to the machine. Additionally, the depth information of the identified object is processed to calculate the distance to the machine. For this, real data of special machines were taken, labeled and the Mask-RCNN model was trained. This approach shows promising results, with average values of the Jaccard Index at 95.9 % and the Dice Coefficient at 97.7 %, and an inference time of 0.083 s, enabling real-time detection.…