Embodied-R1.5: A Subset of Supervised Fine-Tuning Data for Embodied AI Agents

Embodied-R1.5-SFT-Dataset is a subset of the Stage 1 Supervised Fine-Tuning data used to train the Embodied-R1.5 model. The dataset is hosted by author IffYuan on HuggingFace, with a partial release noted as of June 9, 2026. The full dataset is described as a work in progress, with JSON files still being uploaded.

Use Cases

• Supervised fine-tuning of language models for embodied AI tasks based on the described SFT data.
• Benchmarking agent performance in simulated environments using the associated EmbodiedEvalKit.
• Research into multimodal instruction-following for robotics or virtual agents based on the dataset's domain.

Strengths

• Data is directly associated with a specific model release, Embodied-R1.5.
• The dataset is part of a larger project with a dedicated evaluation kit (EmbodiedEvalKit).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• The release is explicitly partial, with data files still being uploaded as of the last update.

Provenance

Source

IffYuan on HuggingFace.

Collection Method

Likely generated from simulations or human-agent interactions for model training.

Freshness

Last updated 2026-06-09 17:13:55; freshness should be verified.