Research map / Reinforcement learning / Visual and representation RL
Alchemy: A benchmark and analysis toolkit for meta-reinforcement learning agents
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 |
|---|---|---|---|
| switch VMPO from 3D pixels to symbolic observations (ablation) | VMPO · Alchemy | episode score: 156.2 $\pm$ 1.6 -> 155.4 $\pm$ 1.6 | No clear effect |
| add ground-truth chemistry as extra input (ablation) | VMPO · symbolic Alchemy | episode score | Helped |
| add ideal-observer belief state as extra input (ablation) | VMPO · symbolic Alchemy | episode score | Helped |
| add ground-truth or belief-state input (ablation) | VMPO · 3D Alchemy | episode score | Helped |
| give symbolic observations to the 3D agent without extra chemistry input (ablation) | VMPO · 3D Alchemy | episode score | No clear effect |
| add symbolic observations plus ground-truth or belief-state input in 3D (ablation) | VMPO · 3D Alchemy | episode score | Helped |
| add auxiliary feature-prediction tasks (stone and potion counts) (ablation) | VMPO · Alchemy | episode score | Helped |
| add auxiliary chemistry-prediction task (ablation) | VMPO · Alchemy | episode score | Helped |
| add feature-prediction auxiliary tasks in symbolic Alchemy without other augmentation (ablation) | VMPO · symbolic Alchemy | episode score | Helped |
| add symbolic observations when auxiliary tasks already present (ablation) | VMPO · 3D Alchemy | episode score | No clear effect |
| compare VMPO with IMPALA | IMPALA · Alchemy 3D | episode score: 156.2 $\pm$ 1.6 -> 140.2 $\pm$ 1.5 | Hurt |
Closest papers by meaning
- Meta-World+: An Improved, Standardized, RL Benchmark (NeurIPS 2025 · similarity 0.89)
- MAMBA: an Effective World Model Approach for Meta-Reinforcement Learning (ICLR 2024 · similarity 0.88)
- DynaMITE-RL: A Dynamic Model for Improved Temporal Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (ICLR 2025 · similarity 0.87)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.87)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.87)
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.