Human and automated demonstration data acquisition for learning repetitive industrial handling operations

Although automated guided vehicles with forklift functionality are increasingly deployed, most systems rely on path planning approaches with limited adaptability. At the same time, learning-based methods and regulatory requirements emphasize the demand for high-quality real-world datasets. In response to these requirements, this paper makes two main contributions: a real-world industrial truck demonstrator enabling synchronized acquisition of sensor and external ground-truth data, and an automated system for the generation of annotated 6D pose datasets. These advancements aim to reduce data acquisition effort while enabling accurate and verifiable datasets for industrial applications.