metadata
license: mit
task_categories:
- reinforcement-learning
- robotics
language:
- en
pretty_name: Causality Two Room (Glitched Hue)
size_categories:
- 10K<n<100K
robomotic/causality-two-room
This dataset contains trajectories collected in the swm/GlitchedHueTwoRoom-v1
environment for causal world-model experiments.
What is inside
glitched_hue_tworoom.h5: HDF5 dataset with trajectories and rendered frames.- Includes observations, actions, rewards, episode indexing, variation values, and pixel renderings.
How it was generated
Collection script:
scripts/data/collect_glitched_hue.py
Configuration:
scripts/data/config/glitched_hue.yaml
Command used:
python scripts/data/collect_glitched_hue.py num_traj=10000 seed=3072
Generation details:
- Total episodes: 10,000
- Pass 1: 5,000 episodes with blue background + teleport enabled
- Pass 2: 5,000 episodes with green background + teleport disabled
- Render size: 224x224
- File size: 9.09 GB
- Uploaded at (UTC): 2026-04-07 09:09:48
Repository
Notes
The second background hue is green (not red) to preserve strong contrast with the red agent in pixel observations.