Solve a Rubik’s Cube with Computer Vision | GPT-6 + SAM 3 Auto Labeling

See how GPT-6 and SAM 3 were used to train a computer vision model, which powers an application to guide people through the process of solving a Rubik's cube. In this video, Lucas Fochesatto of Roboflow shares his inspiration for creating an app to him solve a Rubik's Cube (hint: he didn't know how to do it on his own) and shows a few details behind the app. You'll see how training images were auto-labeled with GPT-6 Astra + SAM 3 in Roboflow and used to train an RF-DETR Medium segmentation model. Finally, you'll see a demo of the app running on an iPhone. = Chapters = 0:00 Intro: An iPhone App That Coaches a Rubik's Cube Solve 2:00 Auto Label with GPT-6 Astra + SAM 3 4:40 Polygons vs. Masks: Speed, Cost, and Precision 6:38 Review the Auto-Labels, Then Train a Small Model 7:57 Training RF-DETR Medium for Segmentation 8:54 Running the Model On-Device with Core ML 9:19 Live Demo: Solving the Rubik's Cube 11:08 Step-by-Step Guidance for Assembly Tasks = Resources = GPT-6 Astra for Segmentation: https://roboflow.com/blog/gpt-6-astra-for-segmentation The Rubik's Cube dataset on Roboflow Universe: https://universe.roboflow.com/lucass-workspace-fkstf/rubiks2 RF-DETR on GitHub: https://github.com/roboflow/rf-detr Join a future live session: https://roboflow.com/webinar