AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

When Instinct Guides and Insight Grounds: Staged RL Training for LLM Agents

Zijing Zhang, Boning Zhang

AAAI 2026 · Multi-agent systems and games › LLM agents · approach: LLM multi-agent decision making · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a staged RL training approach for LLM agents in interactive environments, with the abstract only partially available; details beyond the abstract are unknown.

Problem

End-to-end RL training of LLM agents for dynamic, interactive decision-making is hard to optimize efficiently, and existing RL-based agent paradigms organize interactions as reasoning followed by action.

Approach

Staged RL training for LLM agents, framed around an instinct (fast action) versus insight (grounded reasoning) distinction; the abstract is truncated so the exact mechanism is not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add action-first exploration stage (stage 1) before reasoning-first training (ablation)ActRe (Qwen2.5-1.5B) · ALFWorldsuccess rate: 0.73 -> 0.83Improved
add reasoning-first alignment stage (stage 2) (ablation)ActRe (Qwen2.5-1.5B) · ALFWorldsuccess rate: 0.78 -> 0.83Improved
raise sampling temperature on key action tokens (ablation)ActRe (Qwen2.5-1.5B) · ALFWorldsuccess rate: 0.81 -> 0.83Improved
add action-first exploration stage (ablation)ActRe (Qwen2.5-1.5B) · WebShopsuccess rate: 0.57 -> 0.64Improved
add reasoning-first alignment stage (ablation)ActRe (Qwen2.5-1.5B) · WebShopsuccess rate: 0.59 -> 0.64Improved
ActRe two-stage training instead of vanilla GRPOGRPO (Qwen2.5-7B) · ALFWorldsuccess rate: 0.78 -> 0.86Improved
use higher sampling temperature during rolloutGRPO (Qwen2.5-1.5B) · ALFWorldsuccess rate: 0.73 -> 0.75No clear effect
use higher sampling temperature during rolloutGRPO (Qwen2.5-7B) · ALFWorldsuccess rate: 0.78 -> 0.80Mixed
add entropy regularization to the lossGRPO (Qwen2.5-7B) · ALFWorldsuccess rate: 0.78 -> 0.82Improved
remove explicit reasoning (no think) in RL training (ablation)GRPO (Qwen2.5-7B) · ALFWorldpass rate: 77.6 -> 86.8Improved
remove explicit reasoning (no think) in SFT (ablation)SFT (Qwen2.5-7B) · ALFWorldpass rate: 63.8 -> 46.4Hurt
remove explicit reasoning (no think) in PPO (ablation)PPO (Qwen2.5-1.5B) · ALFWorldpass rate: 64.6 -> 61.4Hurt

Limitations

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy