---
format: "aidr-story-markdown/v1"
id: "1e2c25139a2db2c0ec545320ade8be50fb74821442ad2d60188dfc4097a6bb3a"
canonical_url: "https://aidr.today/1e2c2513?lang=en"
title: "R^2Orch: Recursive Self-Improvement for Agent Orchestration via Self-Play within a Skill-Internalization Loop"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-02T00:00:00.000Z"
category: "Research"
topics: ["agent"]
source_urls: ["https://huggingface.co/papers/2610.22740"]
summary: "R^2Orch is a new paradigm of Recursive Self-Play Orchestration. It uses an inner loop of self-play to improve the orchestrator on self-chosen tasks. An outer loop improves how the inner loop improves."
---

# R^2Orch: Recursive Self\-Improvement for Agent Orchestration via Self\-Play within a Skill\-Internalization Loop

> [Open the canonical story](<https://aidr.today/1e2c2513?lang=en>)

**Published:** 2026-11-02T00:00:00.000Z
**Category:** Research
**Topics:** agent

## Summary

R^2Orch is a new paradigm of Recursive Self\-Play Orchestration\. It uses an inner loop of self\-play to improve the orchestrator on self\-chosen tasks\. An outer loop improves how the inner loop improves\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.22740>)