Research map / Reinforcement learning / Visual and representation RL
Memory Gym: Towards Endless Tasks to Benchmark Memory Capabilities of Agents
Marco Pleines, Matthias Pallasch, Frank Zimmer, Mike Preuss
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 observation reconstruction loss (ablation) | TrXL · Finite tasks aggregate | mean task progression: 56% -> 99% | Improved |
| add observation reconstruction loss (ablation) | GRU · Searing Spotlights | sample efficiency | Improved |
| replace TrXL with GRU | PPO agent · Endless Mortar Mayhem | commands executed: 17 -> 115 | Improved |
| replace TrXL with GRU | PPO agent · Endless Searing Spotlights | coins collected: 6.7 -> 23.9 | Improved |
| replace TrXL with GRU | PPO agent · Endless Mystery Path | tiles visited: 20 -> 26 | Improved |
| scale TrXL layers, embedding and memory window (ablation) | TrXL · Endless Mortar Mayhem | commands executed | No clear effect |
| add ground truth estimation head (ablation) | TrXL · Endless Mortar Mayhem | mean commands: 16 -> 36 | Improved |
| raise final learning rate to 1e-4 (ablation) | TrXL · Endless Mortar Mayhem | mean commands: 16 -> 25 | Improved |
| add absolute positional encoding to query (ablation) | TrXL · Endless Mortar Mayhem | mean commands: 16 -> 18 | Improved |
| combine tuned LR, query positional encoding and ground truth head (ablation) | TrXL · Endless Mortar Mayhem | mean commands: 16 -> 47 | Improved |
| use learned or relative positional encoding (ablation) | TrXL · Endless Mortar Mayhem | mean commands | No clear effect |
| turn off advantage normalization (ablation) | recurrent PPO agent · Searing Spotlights (full observability) | success rate: 0.43 -> 1.00 | Improved |
Closest papers by meaning
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Mem-T: Densifying Rewards for Long-Horizon Memory Agents (ICML 2026 · similarity 0.87)
- Mastering Memory Tasks with World Models (ICLR 2024 · similarity 0.86)
- Stable Hadamard Memory: Revitalizing Memory-Augmented Agents for Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.86)
- Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks (ICML 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.
