bridge_orig_lerobot
VerifiedWidowX robot episodes recorded at 5 FPS via LeRobot.
What is bridge_orig_lerobot?
bridge_orig_lerobot holds synchronized state, action, and video streams from WidowX robot trajectories generated through the LeRobot pipeline.
It supports training and evaluation of imitation-learning and control models for robotic manipulation research.
What you can build with bridge_orig_lerobot
Train imitation learning policies
Use the 53k episodes of WidowX arm trajectories to train visuomotor policies that map video observations to actions for manipulation tasks.
Benchmark robot learning algorithms
Evaluate new methods on the 19,974 tasks using the provided parquet files and 212k videos as a standardized LeRobot-format dataset.
Develop task-conditioned models
Leverage the large number of distinct tasks to train or fine-tune models that generalize across varied WidowX manipulation scenarios.
Load bridge_orig_lerobot
from datasets import load_dataset
ds = load_dataset("IPEC-COMMUNITY/bridge_orig_lerobot")- 1pip install datasets
- 2from datasets import load_dataset
- 3dataset = load_dataset('IPEC-COMMUNITY/bridge_orig_lerobot')
- 4Access episodes via dataset['train'] and load associated videos/parquet files
- 5Use LeRobot v2.0 utilities to process frames and actions for training
bridge_orig_lerobot: pros & cons
Pros
- +Very large scale with over 53k episodes and 1.8M frames
- +Includes both video and structured parquet data
- +Produced with standardized LeRobot v2.0 format
- +Covers nearly 20k distinct tasks on WidowX hardware
Cons
- –Only a single training split provided
- –Limited to WidowX robot arm platform
- –Requires LeRobot-specific tooling for full use
Frequently asked questions
A large-scale robot learning dataset with 53,192 episodes collected on a WidowX arm using LeRobot v2.0, containing videos and parquet files for 19,974 tasks.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…