RoboStressBench: Robotics Vision-Language Benchmark Under Visual Stress

RoboStressBench is a robotics-oriented vision-language benchmark for evaluating model robustness under visual stress conditions. It contains two task formats: visual question answering and localization tasks. The dataset was created by RoboStressBench and was last updated on June 2, 2026.

Use Cases

• Evaluate model robustness for scene interpretation under stress based on the visual question answering tasks.
• Benchmark planning capabilities for embodied actions based on the planning task format.
• Test spatial reasoning for relative-position questions based on the spatial task format.
• Assess state understanding for task completion based on the state understanding task.

Strengths

• Designed specifically for evaluating robustness under visual stress conditions.
• Includes multiple task formats (visual question answering and localization) for comprehensive evaluation.
• Dataset was last updated on June 2, 2026, suggesting recent maintenance.

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 license information are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Freshness

Last updated 2026-06-02 02:45:30