CLARU
VincentNiapache-2.0

RoboTwin stack_blocks_three — Wan2.2 TI2V Rollouts

160 text+image-to-video rollouts for the stack_blocks_three task from RoboTwin, generated with the Wan2.2 TI2V (5B) diffusion model and scored with the blocks_stack_v2 reward model. Companion dataset to the EmbodiedVideoRL / DanceGRPO reward-model work.

Downloads4
Episodes160

Why This Matters for Physical AI

This dataset enables training and evaluation of reward models for video-based robotic control by providing ground-truth task success labels and detailed per-frame tracking annotations for a complex manipulation task.

Technical Profile

Modalities
rgblanguage
Robot Embodiments
dual_arm_manipulator
Action Space
joint_positions
Environment
simulation
Task Types
pick_and_placestackingmanipulation
Episodes
160
Data Format
mp4
Annotation Types
language_instructionsreward_labelsaction_labelssegmentationbounding_boxes
License
apache-2.0
Part of the RoboTwin family

Access

View on HuggingFaceRead Paper

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

Xperience-10M

A large-scale egocentric multimodal dataset of human experience containing 10 million interactions and 10,000 hours of synchronized first-person recordings with six video streams, audio, stereo depth, camera pose, hand mocap, full-body mocap, IMU, and hierarchical language annotations for embodied AI, robotics, and world modeling research.

rgbaudiodepthproprioception+3
2.3M downloadsMar 2026other

LingBot-Depth Dataset

Self-curated RGB-D dataset for training masked depth modeling approaches, containing real-world indoor scenes, VLA robot manipulation tasks, and simulated data across multiple camera types and robot platforms.

rgbdepth
247K downloadsApr 2026CC BY-NC-SA 4.0

Open-H-Embodiment

A multi-embodiment community-driven dataset of paired kinematics and video for training AI autonomy models in surgical robotics and ultrasound applications. The dataset includes tabletop exercises, clinical procedures, and simulations of healthcare robotics tasks.

rgbproprioception
140K downloadsMar 2026cc-by-4.0

OmniAction

A large-scale multimodal dataset for proactive robot manipulation comprising 141,162 episodes with cross-modal contextual instructions derived from spoken dialogue, environmental sounds, and visual cues rather than explicit commands.

rgbaudiolanguage
63K downloadsMar 2026cc-by-nc-4.0

ManiTwin-100K: Manipulation-Ready Digital Object Twins

A large-scale dataset of 100K manipulation-ready digital object twins with simulation-ready 3D meshes, physical properties, functional point annotations, grasp configurations, and language descriptions validated through physics-based simulation.

3d_meshlanguage
51K downloadsApr 2026cc-by-nc-4.0

OmniAction

A large-scale multimodal dataset for proactive robot manipulation comprising 141,162 episodes across 112 skills and 748 objects, enriched with audio, visual, and contextual instruction data for cross-modal intention recognition.

rgbaudiolanguage
46K downloadsMar 2026cc-by-nc-4.0