File size: 2,503 Bytes
fe72315
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc6cc73
 
 
 
 
 
 
 
 
 
 
fe72315
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
---
license: cc-by-4.0
language:
  - en
pretty_name: HomeSafeBench
tags:
  - embodied-ai
  - vision-language
  - home-safety
  - virtualhome
configs:
  - config_name: benchmark
    default: true
    data_files:
      - split: train
        path: data/train/*.json
      - split: test
        path: data/test/*.json
  - config_name: sft_action
    data_files:
      - split: train
        path: cueback/sft_action.jsonl
  - config_name: sft_both
    data_files:
      - split: train
        path: cueback/sft_both.jsonl
  - config_name: cueback
    data_files:
      - split: train
        path: cueback/cueback.jsonl
---

# HomeSafeBench

This repository contains the official data release for the paper
[HomeSafeBench: Benchmarking Embodied Vision-Language Models in Free-Exploration Home Safety Inspection](https://arxiv.org/abs/2509.23690).

HomeSafeBench is a benchmark for free-exploration home safety inspection with
embodied vision-language models, built on
[VirtualHome](https://github.com/xavierpuigf/virtualhome).

## Dataset Contents

- `data/train/`: 3,400 training tasks.
- `data/test/`: 1,000 human-validated evaluation tasks.

The CueBack data contains 3,158 aligned trajectories in three variants. They
share the same first-person observations, tool calls, and tool feedback, but
differ in the supervision text preceding each action:

- `cueback/sft_action.jsonl`: executable actions without reasoning.
- `cueback/sft_both.jsonl`: the same actions with rationales generated directly
  from the trajectory prefix, current observation, and action.
- `cueback/cueback.jsonl`: the same actions with clue-based reasoning
  constructed by CueBack.
- `cueback/images/`: the referenced first-person observations.

The benchmark format, setup instructions, and evaluation code are documented
in the [HomeSafeBench code repository](https://github.com/BITHLP/HomeSafeBench).

## Citation

```bibtex
@misc{yao2026homesafebenchbenchmarkembodiedvisionlanguage,
      title={HomeSafeBench: Benchmarking Embodied Vision-Language Models in Free-Exploration Home Safety Inspection},
      author={Jiashu Yao and Haoyu Wen and Siyuan Gao and Yuhang Guo and Zeming Liu and Heyan Huang},
      year={2026},
      eprint={2509.23690},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2509.23690},
}
```

## License

The HomeSafeBench dataset is released under the
[Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/).