Decentralized agent swarms can learn effective coordination through market-based bidding alongside action learning, enabling flexible, scalable multi-agent systems without central control.
This paper introduces market-based post-training (MPT), a framework where agent swarms learn both what actions to take and how to coordinate with each other through a bidding mechanism. Instead of a central controller, agents express willingness to participate via bids, allowing the group to self-organize while learning.