Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Towards mental time travel: a hierarchical memory for reinforcement learning agents
Andrew Kyle Lampinen, Stephanie C.Y. Chan, Andrea Banino, Felix Hill
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 |
|---|---|---|---|
| use HCAM memory instead of Gated TransformerXL | RL agent · Ballet 2 dances, delay 16 | % correct: 96.9 -> 99.8 | Improved |
| use HCAM memory instead of Gated TransformerXL | RL agent · Ballet 8 dances, delay 16 | % correct: 65.7 -> 98.1 | Improved |
| use HCAM memory instead of Gated TransformerXL | RL agent · Ballet 8 dances, delay 48 | % correct: 49.2 -> 97.2 | Improved |
| use HCAM memory instead of LSTM | RL agent · Ballet 8 dances, delay 16 | % correct: 25.2 -> 98.1 | Improved |
| use HCAM memory instead of TrXL | RL agent · Object permanence, no delay | % correct: 82.2 -> 96.7 | Improved |
| use HCAM memory instead of TrXL | RL agent · Object permanence, long delay | % correct: 46.1 -> 91.7 | Improved |
| train only on long object-permanence tasks (no short scaffold) with HCAM | HCAM · Object permanence, long delay | % correct: 91.7 -> 82.9 | Mixed |
| use HCAM instead of TrXL for extrapolation to 10 distractors | RL agent · Rapid word learning | % correct: 76.7 -> 93.0 | Improved |
| use HCAM instead of TrXL for recall across 4 episodes | RL agent · Rapid word learning | % correct: 49.3 -> 82.8 | Improved |
| use HCAM instead of TrXL for 2 episodes with 5 distractors each | RL agent · Rapid word learning | % correct: 48.7 -> 71.1 | Improved |
| use HCAM instead of TrXL | RL agent · One-Shot StreetLearn | average reward: 19.9 -> 26.8 | Improved |
| select a single memory chunk instead of top-k (ablation) | HCAM · Ballet | % correct | No clear effect |
Closest papers by meaning
- MemGen: Weaving Generative Latent Memory for Self-Evolving Agents (ICLR 2026 · similarity 0.89)
- Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks (ICML 2026 · similarity 0.89)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session Agents (ICLR 2026 · similarity 0.88)
- Meta-Cognitive Memory Policy Optimization for Long-Horizon LLM Agents (NeurIPS 2026 · similarity 0.88)
- AUTOMEM: Automated Learning of Memory as a Cognitive Skill (NeurIPS 2026 · similarity 0.88)
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.
