Market training lets agent swarms self-orchestrate

Market-based post-training lets decentralized agent swarms learn actions and bidding without a central coordinator.

Two reports describe the paper Learning Self-Orchestration in Decentralized Agent Swarms, first listed on Hugging Face daily papers on 2026-10-14 and then on arXiv in cs.AI, cs.LG, and cs.CL on 2026-10-15. The work introduces market-based post-training (MPT), which jointly learns action policies and bidding behavior so a decentralized agent swarm can decide who participates without a central coordinator. Bids determine who acts and how collective outcomes assign local training credit. On embodied crafting, robot navigation, and deep research, MPT outperformed monolithic post-training, and learned orchestration beat other coordination strategies that used the same action policies. Trained swarms adapted faster to related tasks, improved as populations grew, and lost less performance when agents dropped out. The sources do not disagree; the later arXiv report adds that code and trained models are released on GitHub.

• Paper: Learning Self-Orchestration in Decentralized Agent Swarms, noted on Hugging Face daily papers on 2026-10-14 and on arXiv (cs.AI/cs.LG/cs.CL) on 2026-10-15.
• Market-based post-training (MPT) jointly learns action policies and bidding-based self-orchestration so decentralized swarms decide who participates without a central coordinator.
• Bids allocate who acts and supply local training credit from collective outcomes; no central coordinator assigns roles or credit.
• On embodied crafting, robot navigation, and deep research, MPT beat monolithic post-training, and learned orchestration outperformed other coordination strategies that used the same action policies.
• Trained swarms adapted faster to related tasks, scaled better with larger populations, and lost less performance under agent dropout.
• The arXiv report adds that code and trained models are released on GitHub.