Research map / Large language models / Agents, code and math
Designing Observation and Action Models for Efficient Reinforcement Learning with LLMs
Deunsol Yoon, Sunghoon Hong, Whiyoung Jung, Junseok Park, Geon-Hyeong Kim, Woohyung Lim, Soonyoung Lee, Seung Hwan Kim, Kanghoon Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add both observation and action models (full LOAM) instead of observation-only or action-only (ablation) | LOAM · HumanoidBench basic locomotion | return / convergence speed | Improved |
| use Bayesian ridge regression acquisition instead of Gaussian process regression (ablation) | LOAM-Race · HumanoidBench basic locomotion | return | No clear effect |
| use racing instead of successive halving (SHA4, SHA8) for candidate allocation (ablation) | LOAM-Race · HumanoidBench basic locomotion | convergence speed | Improved |
| use UCB-style acquisition instead of always selecting current best candidate (CBS) (ablation) | LOAM-Race · HumanoidBench basic locomotion | return | Improved |
| increase candidates from K=3 to K=4 (ablation) | LOAM-Race · h1hand-cube, h1hand-insert_small | convergence speed | Hurt |
| increase number of candidates K (up to 3) (ablation) | LOAM-Race · h1hand-cube, h1hand-insert_small | performance variance | Improved |
| add racing (LOAM-Race) when using smaller LLM backbones (GPT-5-nano, Qwen3-Coder-30B) (ablation) | LOAM · HumanoidBench basic locomotion | return | Improved |
| replace GPT-5 with smaller LLMs in LOAM without racing (ablation) | LOAM · HumanoidBench basic locomotion | return | Hurt |
| apply LOAM design directly to raw observations rather than appending LLM features to handcrafted observations (LESR) | FastTD3 · HumanoidBench run, hurdle, slide, push | return | Improved |
| add LOAM-Race selection over single LOAM design | LOAM · insert_small and cube | asymptotic return | Mixed |
Closest papers by meaning
- Can large language models explore in-context? (NeurIPS 2024 · similarity 0.91)
- From Words to Actions: Unveiling the Theoretical Underpinnings of LLM-Driven Autonomous Systems (ICML 2024 · similarity 0.90)
- Expanding LLM Agent Boundaries with Strategy-Guided Exploration (NeurIPS 2026 · similarity 0.90)
- LLM-Explorer: A Plug-in Reinforcement Learning Policy Exploration Enhancement Driven by Large Language Models (NeurIPS 2025 · similarity 0.89)
- Reinforcement World Model Learning for LLM-based Agents (NeurIPS 2026 · similarity 0.89)
- LLM-Empowered State Representation for Reinforcement Learning (ICML 2024 · similarity 0.89)
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.
