WCSE

Embedded AI-Box for Visual Detection and Inspection with Mobile Robots in ROS2

Abstract— Computer vision has made impressive progress in recent years due to the surge of deep learning (DL). Many DL approaches have been developed to detect, classify and segment many kinds of objects. Most of these models require a large amount of data and computing resources for training. This is specially inconvenient for mobile robots and embedded systems, because they have limited hardware and energy. We present a modular approach for a distributed system and an AI-Box for visual detection of objects and persons. The AI-Box has dedicated hardware for DL, a camera, and it runs ROS2 and Detectron2. The AIBox can easily be installed onto a mobile robot to extend its capabilities. The trained model is optimized and reduced using pruning and quantization. Finally, the model is executed on the AI-box installed on the robot. We show how the AI-Box can be integrated onto a quadruped robot (Spot). The AI-Box can run ”compressed” DL vision models and provide detection capabilities to the robot. We also show that AIBox can also be easily extended to run different models for detection, classification and segmentation. By using a system like this, mobile robots can be enhanced with powerful models trained offline for different capabilities.

Index Terms— embedded AI; mobile robots; vision; object and person detection.

Atanas Tanev
FZI Research Center for Information Technology, Karlsruhe, Germany
J. Camilo Vasquez Tieck
FZI Research Center for Information Technology, Karlsruhe, Germany
Georg Heppner
FZI Research Center for Information Technology, Karlsruhe, Germany
Philip Keller
FZI Research Center for Information Technology, Karlsruhe, Germany
Arne Roennau
FZI Research Center for Information Technology, Karlsruhe, Germany
RüdigerDillmann
FZI Research Center for Information Technology, Karlsruhe, Germany

[Download]

Cite: Atanas Tanev, J. Camilo Vasquez Tieck, Georg Heppner, Philip Keller, Arne Roennau, Rüdiger Dillmann, " Embedded AI-Box for Visual Detection and Inspection with Mobile Robots in ROS2, " WCSE 2022 Spring Event: 2022 9th International Conference on Industrial Engineering and Applications, pp. 455-462, Sanya, China, April 15-18, 2022.