Dynamic Event‐Triggered RL for Heterogeneous MASs Bipartite Consensus

This research explores bipartite consensus in heterogeneous multi‐agent systems (MASs) by integrating reinforcement learning (RL) with an event‐triggered mechanism (ETM). Initially, the heterogeneous MASs are partitioned into two distinct groups, each following a designated leader, and an estimated velocity method is employed for standardization. Subsequently, an RL algorithm utilizing actor‐critic neural networks is developed to learn optimal control policies from the input–output data of the MASs. Communication between agents is restricted to instances triggered by events, aligning the learning process with the ETM to minimize resource consumption. Moreover, this study introduces a novel discontinuity detection mechanism within the triggering function, which diverges from traditional ETMs by eliminating the need for continuous monitoring of neighboring agents' states, thereby conserving communication resources. This approach rigorously ensures tracking control within subgroups, achieving consensus on both position and velocity while maintaining bounded neural network weights. Finally, the effectiveness of the proposed algorithm is validated through two simulation examples.