A stacked graph neural network with self-exciting process for robotic cognitive strategy reasoning in proactive human-robot collaborative assembly

Proactive human-robot collaborative assembly, a cognitively driven human-robot collaboration, requires research into robot cognitive strategy reasoning to ensure that the robot actively collaborates with the operator in task completion. However, current methods primarily focus on the pairwise relationships of assembly components in discrete snapshots. They could fail to represent the interconnected status of dynamic assembly, leading to inaccurate task allocation, thereby affecting robotic cognitive strategy. To address this problem, we propose a stacked graph neural network (GNN) with self-exciting process to capture the correlation and triggering mechanisms between time-varying tasks. Firstly, a temporal hypergraph with assembly knowledge is constructed to represent the non-pairwise relationships among assembly components in time-varying tasks, aiming to reduce the redundant information brought by pairwise relationships. Then, considering the characteristic of mutual influence between assembly events, a Hawkes process is introduced into the stacked GNN architecture to learn the event correlation representation in the temporal hypergraph. This point process models the self-exciting process of assembly events for simultaneously capturing the individual and collective features of events, thereby revealing the triggering mechanisms of the dynamic events. Finally, the effectiveness of proposed method is demonstrated by comparative experiments and the results of robotic cognitive strategy reasoning on dynamic assembly.