LaGO: Latent Action Guidance for Online Reinforcement Learning

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Kuan-Yen Liu, Ren-Jyun Huang, Ti-Rong Wu
• Venue: arXiv: cs.AI
• Published: 2026-06-23
• Collected: 2026-06-24
• URL: http://arxiv.org/abs/2606.24669v1
• PDF: https://arxiv.org/pdf/2606.24669v1
• DOI: UNKNOWN

Plain-Language Brief

The paper 'LaGO: Latent Action Guidance for Online Reinforcement Learning' proposes a new framework called LaGO. This framework uses a pre-trained language model (LLM) as a guide to help with online policy optimization in reinforcement learning, rather than relying on the LLM as an explicit planner or controller. The authors claim that LaGO improves both reward and success rate in experiments conducted on two benchmarks: CLEVR-Robot and Meta-World.

What the Document Says

• The paper discusses Latent Action Guidance for Online Reinforcement Learning (LaGO).
• LaGO uses a pretrained LLM as a latent action prior to softly guide online policy optimization.
• Experiments were conducted on CLEVR-Robot and Meta-World benchmarks.
• LaGO improves both reward and success rate in these experiments.

Key Concepts Mentioned

• Latent Action Guidance for Online Reinforcement Learning (LaGO)
• Large language models (LLMs)
• Online policy optimization
• CLEVR-Robot
• Meta-World

Claims Made by the Document

• LaGO improves both reward and success rate.

Evidence Used

• Increase in average success rate from 15.1% to 27.2% on CLEVR-Robot
• Increase in average success rate from 2.7% to 15.2% on Meta-World.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Large language models (LLMs) have shown strong potential for planning and sequential decision-making, but prior work often relies on using them as direct controllers, which requires precise action generation and can be unreliable in practice. This paper proposes Latent Action Guidance for Online Reinforcement Learning (LaGO), a framework that uses a pretrained LLM as a latent action prior to softly guide online policy optimization, rather than treating the LLM as an explicit planner or controller. Experiments on both a discrete-control benchmark, CLEVR-Robot, and a continuous-control benchmark, Meta-World, demonstrate that LaGO consistently improves both reward and success rate over Vanilla PPO. In particular, LaGO increases the average success rate from 15.1% to 27.2% on CLEVR-Robot and from 2.7% to 15.2% on Meta-World. Our analysis further shows that stronger pretrained LLMs provide more effective guidance, suggesting that LLM knowledge can improve planning and online decision-making.