Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Scene Memory Transformer for Embodied Agents in Long-Horizon Tasks
Check 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 |
|---|---|---|---|
| Store separate observations and attend over memory instead of max-pooling it (ablation) | SMT · Roaming, Coverage, and Search | Reward: 394.7 vs 366.8; 474.6 vs 443.9; 428.4 vs 332.5 | Helped |
| Use memory factorization in place of full self-attention (ablation) | SMT · Roaming, Coverage, and Search | Reward: 394.7 -> 376.4; 474.6 -> 450.1; 428.4 -> 432.6 | Mixed |
| Increase scene-memory capacity (ablation) | SMT · Roaming, Coverage, and Search | Reward | Helped |
| Remove the previous-action modality (ablation) | SMT · Roaming, Coverage, and Search | Reward | Hurt |
| Remove the pose modality (ablation) | SMT · Roaming, Coverage, and Search | Reward | Hurt |
| Remove the depth image (ablation) | SMT · Roaming, Coverage, and Search | Reward | Hurt |
| Remove RGB observations (ablation) | SMT · Coverage and Search | Reward | No clear effect |
| Remove RGB observations (ablation) | SMT · Roaming | Reward: 10 | Hurt |
| Remove segmentation observations (ablation) | SMT · Coverage | Reward | No clear effect |
| Remove segmentation observations (ablation) | SMT · Roaming | Reward: 10 | Hurt |
| Remove RGB observations (ablation) | SMT · Search | Reward: 20 | Hurt |
| Remove segmentation observations (ablation) | SMT · Search | Reward: 15 | Hurt |
Closest papers by meaning
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Spatially-Aware Transformers for Embodied Agents (ICLR 2024 · similarity 0.89)
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · similarity 0.88)
- MemoNav: Working Memory Model for Visual Navigation (CVPR 2024 · similarity 0.87)
- SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025 · similarity 0.87)
- Kinaema: a recurrent sequence model for memory and pose in motion (NeurIPS 2025 · 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.