Learning to Act and Cooperate: Trajectory-Driven Multi-Agent Blackbox Consensus Optimization

arXiv·low signal

A new approach to distributed blackbox consensus optimization replaces handcrafted update rules with learned cooperation patterns for multi-agent systems. Agents improve a global objective using only local queries and limited neighbor communication, learning to balance local adaptation, global coordination, and communication efficiency in heterogeneous nonconvex environments. Relevant for multi-agent system designers moving from hand-tuned to learned coordination protocols.