Research map / Vision-language and multimodal / Robot action and navigation
HomeGuard: VLM-based Embodied Safeguard for Identifying Contextual Risk in Household Task
Xiaoya Lu, Yijin Zhou, Zeren Chen, Ruocheng Wang, Bingrui Sima, Enshen Zhou, Lu Sheng, Dongrui Liu, Jing Shao
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 constraint IoU reward (ablation) | HomeGuard-4B · HomeGuard-Bench | RMR: 50.20 -> 63.72 | Improved |
| add target IoU reward (ablation) | HomeGuard-4B · HomeGuard-Bench | oversafety rate: 37.03 -> 21.96 | Improved |
| add target IoU reward (ablation) | HomeGuard-4B · HomeGuard-Bench | RMR: 56.27 -> 63.72 | Improved |
| add visual grounding (IoU) supervision overall (ablation) | HomeGuard-4B · HomeGuard-Bench | oversafety rate: 42.35 -> 21.96 | Improved |
| add visual grounding (IoU) supervision overall (ablation) | HomeGuard-4B · HomeGuard-Bench | RIR: 90.98 -> 90.00 | No clear effect |
| add RFT with CG-CoT on top of prompting only the 4-step CoT (ablation) | Qwen3-VL-4B-Thinking · HomeGuard-Bench | RMR: 43.50 -> 63.72 | Improved |
| train with CG-CoT instead of prompting safety principles directly (ablation) | Qwen3-VL-4B-Thinking · HomeGuard-Bench | oversafety rate: 35.56 -> 21.96 | Improved |
| prompt with ground-truth boxes as visual anchors (ablation) | Qwen3-VL-4B-Thinking · HomeGuard-Bench | RIR: 66.67 -> 84.71 | Improved |
| two-stage SFT+RFT training with CG-CoT | Qwen3-VL-4B-Thinking · HomeGuard-Bench | RIR: 66.67 -> 90.00 | Improved |
| two-stage SFT+RFT training with CG-CoT | Qwen3-VL-8B-Thinking · HomeGuard-Bench | oversafety rate: 32.62 -> 13.14 | Improved |
| integrate HomeGuard safety tips into planner | Qwen3-VL-8B-Thinking planner · IS-Bench | safe success rate: 29.54 -> 45.65 | Improved |
| integrate HomeGuard safety tips into planner | Qwen3-VL-8B-Thinking planner · IS-Bench | task success rate: 61.36 -> 73.91 | Improved |
Closest papers by meaning
- AGENTSAFE: Benchmarking the Safety of Embodied Agents on Hazardous Instructions (CVPR 2026 · similarity 0.90)
- IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Tasks (AAAI 2026 · similarity 0.90)
- Evolving Contextual Safety in Multi-Modal Large Language Models via Inference-Time Self-Reflective Memory (CVPR 2026 · similarity 0.87)
- Multimodal Situational Safety (ICLR 2025 · similarity 0.87)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (ICLR 2025 · similarity 0.87)
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 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.
