Adaptive Event--Triggered Consensus Control of Heterogeneous Multi--Robot Systems Based on Deep Reinforcement Learning

    The coordinated control of heterogeneous multi\u2011robot systems is an important research direction in the current field of intelligent systems and has broad application prospects in industrial manufacturing, disaster rescue, and intelligent logistics. However, the existing consensus control methods generally face challenges such as limited communication resources, difficulties in modeling heterogeneous dynamics, and environmental uncertainties, which make it difficult to achieve efficient and adaptive cooperative control. This study proposes an adaptive event\u2011triggered control framework that integrates deep reinforcement learning (DRL\u2011AETC) for the consensus control problem of heterogeneous multi\u2011robot systems. First, a heterogeneity\u2011aware dynamics encoder based on a graph attention network is designed to map the states of different types of robots into a unified latent space, effectively capturing the heterogeneous characteristics and topological relationships among robots; second, an adaptive event\u2011triggering mechanism is proposed, which dynamically optimizes the triggering thresholds via deep reinforcement learning, significantly reducing the communication frequency while ensuring consensus convergence; finally, a two\u2011layer policy optimization architecture based on the actor\u2013critic framework is constructed to realize joint optimization and fast adaptation of the event\u2011triggering policy and the control policy. Theoretical analysis under the Lyapunov stability framework proves the convergence of the proposed method under bounded disturbances and its Zeno\u2011free property. Experiments on a simulation platform containing three types of heterogeneous robots show that, compared with traditional periodic triggering methods, the proposed method reduces communication overhead by $67.3\\%$, and compared with existing event\u2011triggered methods, it increases the convergence speed by $42.8\\%$, while also exhibiting excellent robustness under scenarios with noise disturbances and topology switching. This research provides a new theoretical framework and technical approach for resource\u2011constrained coordinated control of heterogeneous multi\u2011robot systems.