Research map / Reinforcement learning / RL from feedback
Multimodal LLM-assisted Evolutionary Search for Programmatic Control Policies
Qinglong Hu, Tong Xialiang, Mingxuan Yuan, Fei Liu, Zhichao Lu, Qingfu Zhang
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 visual behavioral-evidence analysis to EoH (MLES vs EoH) | MLES-EoH · Car Racing testing | score: 79.286 -> 96.358 | Improved |
| add visual behavioral-evidence analysis to EoH (MLES vs EoH) | MLES-EoH · Lunar Lander testing | score: 0.776 -> 0.819 | Mixed |
| compare MLES with PPO on test instances | MLES · Lunar Lander testing | score: 0.846 -> 0.819 | Hurt |
| compare MLES with PPO on Car Racing test | MLES · Car Racing testing | score: 94.546 -> 96.358 | Improved |
| remove multimodal modification operator M1_M (ablation) | MLES · Lunar Lander | score: 0.997 -> 1.090 | Improved |
| remove multimodal modification operator M2_M (ablation) | MLES · Car Racing | score: 84.77 -> 98.70 | Improved |
| add exploration operator E1 (ablation) | MLES · Lunar Lander | score: 1.093 -> 1.090 | No clear effect |
| add exploration operator E1 (ablation) | MLES · Car Racing | score: 87.53 -> 98.70 | Improved |
| provide an initial seed policy (GPT-4o-mini) (ablation) | MLES · Lunar Lander | score @ 2k: 0.667 -> 1.090 | Improved |
| use GPT-5-mini instead of GPT-4o-mini without seeds (ablation) | MLES · Lunar Lander cold start | score @ 2k: 0.667 -> 1.013 | Improved |
| use GPT-5-mini instead of GPT-4o-mini with seeds (ablation) | MLES · Lunar Lander | score @ 2k: 1.090 -> 1.065 | Hurt |
| add image evidence to M1 operator (ablation) | M1 · Lunar Lander operator test | best policy improvement: +1.18% -> +4.39% | Improved |
Closest papers by meaning
- Synthesizing Programmatic Reinforcement Learning Policies with Large Language Model Guided Search (ICLR 2025 · similarity 0.88)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.87)
- LLM-Explorer: A Plug-in Reinforcement Learning Policy Exploration Enhancement Driven by Large Language Models (NeurIPS 2025 · similarity 0.87)
- Self-evolving LLM agents with in-distribution Optimization (ICML 2026 · similarity 0.87)
- $\mathcal{B}$-Coder: Value-Based Deep Reinforcement Learning for Program Synthesis (ICLR 2024 · similarity 0.86)
- Reflective Prompted Policy Optimization: Trajectory-Grounded Revision and Salience Bias (NeurIPS 2026 · similarity 0.86)
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.
