Abstract

Autonomous Mobile Systems (AMS) offer sig-nificant advantages for industry and private sectors by adapting to diverse and dynamic environments. To train these systems, large amounts of data are required, typically obtained from simulated environments. However, the creation of these environments is often labor-intensive. Here, we propose a generative pipeline that provides a streamlined approach to virtual training and testing while allowing users to apply automated methods including generative AI.  
Our pipeline consists of four, partly iterative main steps. The pipeline spans from the creation of individual assets to the utilization of the simulated environments. The pipeline is then implemented for an exemplary scenario, utilizing multiple methods including generative AI.  Furthermore, we propose a novel application of our pipeline to provide robots with the capabilities to “imagine” virtual experiences based on anticipated tasks.  
The presented pipeline not only simplifies the process of generating simulated environments, but also resembles a scalable framework for developing increasingly complex AMS.