Research map / Reinforcement learning / Visual and representation RL
Control of Memory, Active Perception, and Action in Minecraft
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 |
|---|---|---|---|
| add context-dependent memory retrieval with feedback (FRMQN vs DQN) | FRMQN · I-Maze size 40 | success rate (%): 46.6 -> 89.0 | Helped |
| FRMQN vs DRQN on large I-Mazes | FRMQN · I-Maze size 20 | success rate (%): 63.6 -> 98.8 | Helped |
| add feedback connection from retrieved memory (RMQN -> FRMQN) (ablation) | FRMQN · Pattern matching unseen | success probability (%): 62.3 -> 91.8 | Helped |
| add recurrence to MQN (MQN -> RMQN) (ablation) | RMQN · Seq+I unseen-l | success rate (%): 34.4 -> 49.5 | Helped |
| feedback connection in random mazes (RMQN vs FRMQN) (ablation) | FRMQN · Seq+I unseen random maze | average reward: -0.32 -> -0.28 | No clear effect |
| MQN on unseen patterns | MQN · Pattern matching | success probability (%): 99.0 -> 69.3 | Hurt |
| DRQN on pattern matching | DRQN · Pattern matching train | success probability (%): 49.7 | Hurt |
| increase corridor length beyond training for DQN | DQN · I-Maze | success rate (%): 99.0 -> 46.6 | Hurt |
| unseen larger maps vs training maps | FRMQN · Seq+I random maze | average reward: 0.48 -> -0.54 | Hurt |
| increase distance between indicator and goal | RMQN and FRMQN vs others · Single Goal with Indicator | precision | Helped |
| DQN on indicator tasks | DQN · Single+I unseen-l | failure rate (%): 31.6 | Hurt |
Closest papers by meaning
- RL-GPT: Integrating Reinforcement Learning and Code-as-policy (NeurIPS 2024 · similarity 0.87)
- MrSteve: Instruction-Following Agents in Minecraft with What-Where-When Memory (ICLR 2025 · similarity 0.87)
- MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perception (CVPR 2024 · similarity 0.87)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.87)
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.86)
- Auto MC-Reward: Automated Dense Reward Design with Large Language Models for Minecraft (CVPR 2024 · 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.