OSCAR Robot: Multi-Embodiment Teleoperation Episodes for World Models

OSCAR_robot is the robot half of the training corpus for OSCAR, an Omni-Embodiment Action-Conditioned World Model for Robotics. It contains curated, filtered, and deduplicated multi-embodiment robot teleoperation episodes re-rendered into a unified conditioning format featuring a kinematic-skeleton overlay. The dataset was authored by zywu2115 and last updated on Hugging Face in June 2026.

Use Cases

• Training action-conditioned world models based on multi-embodiment robot teleoperation data.
• Evaluating cross-embodiment robot policy performance based on a unified video representation.
• Studying robot kinematics and gripper control based on rendered skeleton overlay episodes.

Strengths

• Data is curated, filtered, and deduplicated, which suggests quality control.
• Episodes are re-rendered into a unified conditioning format, which suggests consistency.
• The dataset is specifically designed for training the OSCAR world model.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and data size are unknown, which limits suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face user zywu2115.

Collection Method

Multi-embodiment robot teleoperation episodes were curated, filtered, deduplicated, and re-rendered.

Freshness

Last updated 2026-06-04 06:59:00; freshness should be verified.