EmbodimentSemantic: Spatial Scene Graphs from Robot Manipulation Trajectories

EmbodimentSemantic is a benchmark dataset for evaluating vision-language models on their ability to recover exact spatial scene graphs from robot manipulation trajectories. The dataset was created by researchers from Politecnico di Torino, Huawei Noah's Ark Lab, and University College London. It was last updated on 2026-06-03.

Use Cases

• Benchmarking VLM spatial reasoning based on robot manipulation trajectories
• Evaluating scene graph generation accuracy based on embodied interaction data
• Training models for robotic task planning based on spatial scene understanding

Strengths

• Designed as a benchmark for a specific task: evaluating VLMs on spatial scene graph recovery.
• Created by a collaboration of academic and industry research institutions.

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 is unknown, which may limit suitability assessment.

Provenance

Source

SemVLA (via Hugging Face)

Collection Method

Likely contains data from robot manipulation trajectories.

Freshness

Last updated 2026-06-03 15:00:44; freshness should be verified.