GEM-250K: Generative Supervision for Embodied Vision-Language Models

GEM-250K is a dataset for training Embodied Vision-Language Models (VLMs) using generative supervision. The dataset was created by researchers from Tsinghua University and Tencent Hunyuan and was last updated on Hugging Face in May 2026. Its description suggests it contains 250,000 data points aimed at improving embodied intelligence tasks.

Use Cases

• Training embodied agents for navigation and manipulation tasks based on vision-language instructions.
• Benchmarking generative supervision methods for improving VLM performance in interactive settings.
• Fine-tuning multimodal models on tasks requiring joint understanding of visual scenes and language commands.

Strengths

• Dataset size is indicated as 250,000 entries by its title.
• Created by a collaboration between Tsinghua University and Tencent Hunyuan, suggesting institutional backing.
• Specifically designed for the research area of generative supervision for embodied AI.

Limitations

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

Provenance

Source

Tsinghua University and Tencent Hunyuan

Collection Method

Likely generated or curated for research on generative supervision for embodied AI.

Freshness

Last updated 2026-05-28 04:07:08; freshness should be verified.