cagataydev/neon-failure-recovery-5k
neon-failure-recovery-5k Failure recovery — 5K episodes of synthetic 14-DoF joint trajectories for Neon VLA training. Description Grasp failure → pause → retry patterns for robust policies Each episode contains: language_instruction: Natural language task description actions: JSON array of joint position trajectories (T × 14) length: Episode length (timesteps) Usage 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-failure-recovery-5k.
neon-failure-recovery-5k
Failure recovery — 5K episodes of synthetic 14-DoF joint trajectories for Neon VLA training.
Description
Grasp failure → pause → retry patterns for robust policies
Each episode contains:
language_instruction: Natural language task descriptionactions: JSON array of joint position trajectories (T × 14)length: Episode length (timesteps)
Usage
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 — open-source Vision-Language-Action model for humanoid whole-body control.
Total Neon dataset collection: 160K episodes across 8 datasets (~1.27GB)