| title | R^2Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop |
|---|---|
| authors | Mingda Zhang, Yun Huang, Qiang Huang, Yanjin Li, Rui Qian, Qichao Zhao, Tiesunlong Shen, Na Zhao, Jian Wang, Xiaoying Tang |
| date | 2026-10-22 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.22740 |
| affiliation | Chinese University of Hong Kong, Shenzhen |
| upvotes | 89 (Hugging Face Daily Papers, 2026-11-03) |
| links | abs pdf hf |
| abstract | In recent years, agentic orchestration has been widely applied to solve complex human tasks. However, existing orchestration learning still faces key challenges, including hand-designed curricula, advantage collapse when a task's rollouts all reach the same outcome, and unreliable evaluation signals that judge each skill by one noisy score. To address these challenges, we propose R^2Orch, a new paradigm of Recursive Self-Play Orchestration: an inner loop of self-play that improves the orchestrator on self-chosen tasks, and an outer loop that improves how the inner loop improves. To support this paradigm, we introduce Recursive Learnability, a training objective that rewards a Proposer for choosing tasks whose rollouts keep reaching different outcomes as skills are withdrawn. Built on this objective, we further propose Skill Withdrawal, in which the outer loop withdraws each skill once a controlled evaluation within self-play shows that the orchestrator performs as well without it. Moreover, the two loops form one cycle: failures are distilled into skills, skills into parameters, and withdrawals lead the Proposer to new failures, so the orchestrator moves toward zero-skill deployment. Experimental results on twelve datasets show that R^2Orch consistently outperforms baselines across question answering, math, code, and interactive tasks. Our code is available at https://github.com/beita6969/R2Orch. |
cite
@misc{zhang2026r,
title={R^2Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop},
author={Mingda Zhang and Yun Huang and Qiang Huang and Yanjin Li and Rui Qian and Qichao Zhao and Tiesunlong Shen and Na Zhao and Jian Wang and Xiaoying Tang},
year={2026},
eprint={2610.22740},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.22740}
}