AI-Based Assembly Sequence Planning in a Robotic On-Orbit Assembly Application.

Due to size and weight limitations imposed by launch systems, the on-orbit assembly has increasingly become the focus of robotics research in recent years [1] [2]. However, remote-controlled assembly is problematic due to restricted viewing angles, delays, and limited sensory feedback. Artificial intelligence (AI) approaches can provide solutions in these cases and enable further progress towards fully autonomous robotic on-orbit assembly systems. This paper provides a novel AI-based approach to generate an assembly plan for new satellite modules autonomously. A Seq2Seq Encoder-Decoder approach is used to generate the assembly plans, which are then checked by a sandbox simulation approach to prevent any damage due to incorrect assembly plans. The presented approach generates correct and safe-to-use configurations while outperforming traditional approaches in speed and runtime requirements. The paper presents and evaluates the architecture based on an assembly example of a building block-based satellite module.