Learning Self-Orchestration in Decentralized Agent Swarms
Market-based post-training lets decentralized agent swarms jointly learn actions and self-orchestration without a central coordinator.
The paper introduces market-based post-training (MPT), which jointly learns action policies and bidding-based self-orchestration in decentralized agent swarms without a central coordinator. Bids decide who acts and how collective outcomes assign local training credit. On embodied crafting, robot navigation, and deep research, MPT beat monolithic post-training, and learned orchestration outperformed other coordination strategies using the same action policies. Trained swarms adapted faster to related tasks, improved with larger populations, and lost less performance under agent dropout. Code and models are released.