Research map / Reinforcement learning / Visual and representation RL
Read and Reap the Rewards: Learning to Play Atari with the Help of Instruction Manuals
Yue Wu, Yewen Fan, Paul Pu Liang, Amos Azaria, Yuanzhi Li, Tom Mitchell
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 Read and Reward auxiliary reward | A2C · Pacman (delayed reward) | game score: 452 -> 580 | Improved |
| add Read and Reward auxiliary reward | A2C · Tennis (delayed reward) | game score: -23 -> -5 | Improved |
| add Read and Reward auxiliary reward | A2C · Breakout (delayed reward) | game score: 2 -> 14 | Improved |
| add Read and Reward auxiliary reward | PPO · Pacman (delayed reward) | game score: 204 -> 253 | Improved |
| add Read and Reward auxiliary reward | R2D1 · Pacman (delayed reward) | game score: 2116 -> 3115 | Improved |
| add Read and Reward auxiliary reward | Agent57 · Pacman (delayed reward) | game score: 708 -> 1000 | Improved |
| add Read and Reward auxiliary reward | Agent57 · Breakout (delayed reward) | game score: 2 -> 38 | Improved |
| add Read and Reward auxiliary reward | A2C · Breakout (delayed reward, 3 seeds) | game score: 2.1 -> -4.5 | Hurt |
| add Read and Reward auxiliary reward | A2C · Pacman (delayed reward, 3 seeds) | game score: 387.1 -> 455.2 | Improved |
| add Read and Reward auxiliary reward | R2D1 · Pacman (delayed reward, 3 seeds) | game score: 1999.2 -> 3001.3 | Improved |
| use Wikipedia gameplay text instead of official manual (ablation) | Read and Reward · Tennis, Pacman, Breakout | game score: 580 -> 580 | No clear effect |
| remove delayed reward schedule | Read and Reward with Agent57 · Atari | game score | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Agent57 — Agent57: Outperforming the Atari Human Benchmark (ICML 2020) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning to Follow Instructions in Text-Based Games (NeurIPS 2022 · similarity 0.89)
- Model Based Reinforcement Learning for Atari (ICLR 2020 · similarity 0.88)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.86)
- Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use (ICML 2026 · similarity 0.86)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.86)
- Leveraging Sub-Optimal Data for Human-in-the-Loop Reinforcement Learning (ICLR 2025 · 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.
