Research map / Reinforcement learning / Visual and representation RL
Explore with Long-term Memory: A Benchmark and Multimodal LLM-based Reinforcement Learning Framework for Embodied Exploration
Sen Wang, Bangwei Liu, Zhenkun Gao, Lizhuang Ma, Xuhong Wang, Yuan Xie, Xin Tan
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 |
|---|---|---|---|
| train with RL on navigation only (no memory tool) (ablation) | RA-Mem · LMEE-Bench (58-task subset) | Nav SR: 20.96 -> 20.59 | No clear effect |
| add memory retrieval tool in RL training (ablation) | RA-Mem · LMEE-Bench (58-task subset) | QA Score: 35.52 -> 43.62 | Improved |
| add memory retrieval tool in RL training (ablation) | RA-Mem · LMEE-Bench (58-task subset) | Nav SR: 20.96 -> 23.53 | Improved |
| add action-frontier consistency penalty (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | Nav SR: 22.43 -> 23.53 | Improved |
| add tool-usage penalty (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | Nav SR: 21.32 -> 23.53 | Improved |
| add tool-usage penalty (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | QA Score: 41.03 -> 43.62 | Improved |
| train on multiple-choice questions instead of simple progress question (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | QA Score: 41.33 -> 43.62 | Improved |
| train on all question types versus multiple-choice only (ablation) | MemoryExplorer · GOAT-Bench (Val Unseen subset) | Success rate: 46.40 -> 48.20 | Improved |
| train on all question types versus multiple-choice only (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | Nav SR: 23.53 -> 22.06 | Hurt |
| shift reward weights toward action and frontier (0.4,0.4,0.1,0.1) versus default (ablation) | MemoryExplorer · LMEE-Bench (58-task subset) | Nav SPL: 14.99 -> 12.43 | Hurt |
| use RL with active memory retrieval versus RA-Mem prompting | MemoryExplorer · GOAT-Bench (Val Unseen subset) | Success rate: 42.81 -> 46.40 | Improved |
| use RA-Mem query-based retrieval versus 3D-Mem memory pre-filtering | RA-Mem · GOAT-Bench (Val Unseen subset) | Success rate: 37.05 -> 42.81 | Improved |
Closest papers by meaning
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Exploratory Retrieval-Augmented Planning For Continual Embodied Instruction Following (NeurIPS 2024 · similarity 0.90)
- Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks (ICML 2026 · similarity 0.89)
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · similarity 0.89)
- Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making (NeurIPS 2024 · similarity 0.88)
- WorldMemArena: Evaluating Multimodal Agent Memory Through Action–World Interaction (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.
