CRAFT: Video Diffusion for Bimanual Robot Data Generation

1. 1Utilize video diffusion transformers to synthesize temporally coherent manipulation videos
2. 2Apply Canny-guided techniques to refine generated data
3. 3Integrate CRAFT with existing robot learning frameworks to improve policy robustness
4. 4Evaluate generated data for diversity and quality across viewpoints, object configurations, and embodiments

Robotics engineers and AI researchers on a team can benefit from CRAFT as it generates scalable and diverse bimanual demonstration data, improving policy robustness and reducing the need for real-world data collection

💡 CRAFT can reduce the cost and increase visual diversity of real-world data for bimanual robot learning

💡 CRAFT: a video diffusion framework for generating bimanual robot demo data