BEAT: Visual Backdoor Attack Training Data for Embodied AI Agents

BEAT is a training dataset accompanying the research paper 'BEAT: Visual Backdoor Attacks on VLM-based Embodied Agents via Contrastive Trigger Learning'. The dataset is authored by Qiusi Zhan and collaborators and was last updated on April 20, 2026. It is hosted on Hugging Face and is gated, requiring access requests for academic research on AI security.

Use Cases

• Training and evaluating backdoor attack methods based on the described contrastive trigger learning framework.
• Studying the security robustness of VLM-based embodied agents based on the dataset's adversarial examples.
• Benchmarking defense mechanisms against visual backdoor attacks based on the provided attack data.

Strengths

• Dataset is directly linked to a peer-reviewed arXiv research paper (2510.27623).
• Last update timestamp is precisely recorded (2026-04-20 06:12:32).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and data size are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face dataset repository by author qiusizhan, accompanying the cited arXiv paper.

Collection Method

Likely generated for research on adversarial attacks against embodied AI agents.

Time Range

null

Freshness

Last updated 2026-04-20 06:12:32; freshness should be verified.

Geography

null