cagataydev/neon-g1-diverse-50k

--- license: mit task_categories: - robotics tags: - neon - vla - humanoid - g1 - action-trajectories size_categories: - 10K<n<100K --- # neon-g1-diverse-50k **Multi-scene diversity** — 50K episodes of synthetic 14-DoF joint trajectories for Neon VLA training. ## Description 10 scenes × 34 objects — diverse manipulation across environments Each episode contains: - `language_instruction`: Natural language task description - `actions`: JSON array of joint position trajectories (T × 14) - `length`: Episode length (timesteps) ## Usage ```python import pyarrow.parquet as pq table = pq.read_table("data.parquet") df = table.to_pandas() print(f"Episodes: {len(df)}, Columns: {list(df.columns)}") ``` ## Part of Neon VLA Training data for [Neon](https://github.com/cagataycali/neon) — open-source Vision-Language-Action model for humanoid whole-body control. Total Neon dataset collection: **160K episodes across 8 datasets (~1.27GB)**