fracapuano2026MIT

LeWM TwoRooms

A Lance-converted dataset of 10,000 navigation episodes in a simulated TwoRoom environment, containing 920,809 timesteps with RGB observations, low-dimensional state, and 2D actions for world model training.

Downloads0
Episodes10,000

Why This Matters for Physical AI

This dataset provides large-scale trajectory data with visual observations and proprioceptive state for training world models, which are fundamental for embodied AI systems to predict environment dynamics and plan behaviors.

Technical Profile

Modalities
rgbproprioception
Robot Embodiments
simulated_agent
Action Space
end_effector_delta
Environment
simulation
Task Types
navigation
Episodes
10,000
Data Format
Lance
Annotation Types
reward_labelsaction_labels
License
MIT
Part of the LeWorldModel family

Access

Need custom rgb data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets