Research map / Reinforcement learning / Visual and representation RL
Program Synthesis Guided Reinforcement Learning for Partially Observed Environments
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use MPPS program guidance instead of end-to-end policy | RL agent · 2D-craft | average reward: 0.22 (0.01) -> 0.70 (0.03) | Helped |
| Use MPPS program guidance instead of end-to-end policy | RL agent · Box-world | average reward: 0.85 (0.02) -> 0.90 (0.00) | Helped |
| Use MPPS program guidance instead of end-to-end policy | RL agent · Ant-craft | average reward: 0.12 (0.03) -> 0.40 (0.01) | Helped |
| Use MPPS instead of end-to-end policy | RL agent · 2D-craft | average finish step: 82.3 (1.3) -> 56.4 (2.0) | Helped |
| Replace synthesized programs with handcrafted oracle programs | MPPS · 2D-craft | average reward: 0.70 (0.03) -> 0.76 (0.02) | No clear effect |
| Replace synthesized programs with handcrafted oracle programs | MPPS · Box-world | average finish step: 38.6 (0.4) -> 30.8 (0.5) | Mixed |
| Remove learned hallucinator; naive random exploration before synthesis (ablation) | MPPS · 2D-craft | average reward: 0.70 (0.03) -> 0.024 | Hurt |
| Replace learned hallucinator with optimistic synthesizer or random hallucinator (ablation) | MPPS · 2D-craft | average reward | Hurt |
| Increase number of hallucinator samples m (ablation) | MPPS · Box-world | mean reward | No clear effect |
| Increase number of hallucinator samples m (ablation) | MPPS · Box-world | variance of reward | Helped |
| Make environment non-deterministic (20% action failure) | MPPS · 2D-craft | average reward: 0.70 (0.03) -> 0.47 (0.03) | Hurt |
| Make environment non-deterministic (20% action failure) | MPPS · 2D-craft | average finish step: 56.4 (2.0) -> 73.1 (1.2) | Hurt |
Closest papers by meaning
- Synthesizing Programmatic Reinforcement Learning Policies with Large Language Model Guided Search (ICLR 2025 · similarity 0.87)
- $\mathcal{B}$-Coder: Value-Based Deep Reinforcement Learning for Program Synthesis (ICLR 2024 · similarity 0.85)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · similarity 0.85)
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.85)
- Generating Code World Models with Large Language Models Guided by Monte Carlo Tree Search (NeurIPS 2024 · similarity 0.85)
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.