GSC: A graph-based skill composition framework for robot learning

Qiangxing Tian, Shanshan Zhang, Donglin Wang, Jinxin Liu, Shuyu Yang

Abstract: Highlights•Introduced a Graph-based Skill Composition (GSC) framework for integrating knowledge.•Modeled skill sets as graphs using embeddings with Graph Neural Networks (GNN).•Created a skill graph in GSC with nodes as skills and edges as their relations.•Developed a transformer-style graph architecture for superior skill representation.