Think Outside the Policy: In-Context Steered Policy Optimization

TL;DR

ICPO is a novel reinforcement learning framework that enhances Large Reasoning Models' reasoning abilities by leveraging in-context learning for expert guidance, improving exploration, stability, and performance without requiring advanced model trajectories.

Contribution

The paper introduces ICPO, a unified RLVR framework that uses in-context learning for expert guidance, expanding exploration and stabilizing training without relying on costly expert models.

Findings

ICPO improves reasoning performance on mathematical benchmarks.

ICPO enhances training stability and exploration in RLVR.

ICPO achieves consistent performance gains without advanced model trajectories.

Abstract

Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods, such as Group Relative Policy Optimization (GRPO), have achieved remarkable progress in improving the reasoning capabilities of Large Reasoning Models (LRMs). However, they exhibit limited exploration due to reliance on on-policy rollouts which are confined to the current policy's distribution, resulting in narrow trajectory diversity. Recent approaches attempt to expand policy coverage by incorporating trajectories generated from stronger expert models, yet this reliance increases computational cost and such advanced models are often inaccessible. To address these issues, we propose In-Context Steered Policy Optimization (ICPO), a unified framework that leverages the inherent in-context learning capability of LRMs to provide expert guidance using existing datasets. ICPO introduces mixed-policy GRPO with implicit expert forcing, which expands exploration beyond the current policy distribution without requiring advanced LRM trajectories. To further stabilize optimization, ICPO integrates expert region reject sampling to filter unreliable off-policy trajectories and annealed expert-bonus reward shaping to balance early expert guidance with later autonomous improvement. Results demonstrate that ICPO consistently enhances RLVR performance and training stability on mathematical reasoning benchmarks, revealing a scalable and effective RLVR paradigm for LRMs. Our code is available at https://github.com/Celine-hxy/ICPO.