Research map / Vision-language and multimodal / Robot action and navigation
HiMe: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control
Li Ji, Siyin Wang, Pengfang Qian, Xiaopeng Yu, Yihai Tian, Zhaoye Fei, Jingjing Gong, Xipeng Qiu
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 Sentry to a memory-less Transient Memory planner (ablation) | Transient Memory · Real-robot 3-task average | task progress (%): 14% vs. 26% | Improved |
| add Sentry to HiMe planner (ablation) | HiMe · Real-robot 3-task average | task progress (%): 68% to 90% | Improved |
| replace FIFO flat memory with hierarchical consolidated memory | HiMe · Real-robot 3-task average | task progress (%): 65% vs 90% | Improved |
| add Update/Delete management over append-only memory (ablation) | HiMe · Real-robot 3-task average | task progress (%): 86% vs. 90% | Improved |
| use active management instead of FIFO eviction (ablation) | HiMe · Real-robot 3-task average | task progress (%): 68% vs 90% | Improved |
| use cross-modal memory instead of text-only (ablation) | HiMe · Object Search | task progress (%): 74% -> 92% | Improved |
| use image-only memory instead of text-only (ablation) | HiMe · Object Search | task progress (%): 74% -> 86% | Improved |
| use text-only memory instead of image-only (ablation) | HiMe · Counting | task progress (%): 78% -> 91% | Improved |
| use HiMe instead of Flat Memory for planner calls | HiMe · Object Search | API calls per subtask: 5.4 to 1.8 | Improved |
| use infinite structured memory instead of FIFO window | HiMe · Object Search | memory hit: 68% vs. 94% | Improved |
| increase Sentry context window from 1 to 8 frames (ablation) | Sentry · Subtask completion detection | precision (%): from ~76% to ~82% | Improved |
| increase Sentry context window from 1 to 8 frames (ablation) | Sentry · Subtask completion detection | recall (%): from ~22% to ~35% | Improved |
Closest papers by meaning
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.89)
- ECHO: Continuous Hierarchical Memory for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025 · similarity 0.89)
- Embodied Agents Meet Personalization: Investigating Challenges and Solutions Through the Lens of Memory Utilization (ICLR 2026 · similarity 0.89)
- R2C: Mapping Room to Chessboard to Unlock LLM As Low-Level Action Planner (CVPR 2025 · similarity 0.88)
- RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation (NeurIPS 2025 · 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.
