Spatial language grounding — 20K episodes of synthetic 14-DoF joint trajectories for Neon VLA training.
Spatial reasoning with relative directions (left, right, above, near)
Each episode contains:
language_instruction: Natural language task description
actions: JSON array of joint position trajectories (T × 14)
length: Episode length (timesteps)
import pyarrow.parquet as pq
table = pq.read_table("data.parquet")
df =… See the full description on the dataset page:
https://huggingface.co/datasets/cagataydev/neon-spatial-language-20k.
