cagataydev/

neon-long-horizon-15k

Long-horizon multi-step — 15K episodes of synthetic 14-DoF joint trajectories for Neon VLA training.

4-8 subtask chains with smooth phase transitions 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 = table.to_pandas()… See the full description on the dataset page: https://huggingface.co/datasets/cagataydev/neon-long-horizon-15k.