{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/bc79d24d015646cd8f821503f1c13b3c\" frameborder=\"0\" width=\"1152\" height=\"864\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":864,"width":1152,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":864,"thumbnail_width":1152,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/bc79d24d015646cd8f821503f1c13b3c-9d0fa21aee1f939a.gif","duration":300.229,"title":"Trainable Robot Training with Realistic Simulations","description":"This Loom explains the Trainable simulation pipeline for testing and training robot behavior in realistic physics. It describes sourcing assets from BrightData, including object images, descriptions, and physical details like a 0.1 kg fruit, then converting them into interactable meshes with segmentation and movable parts. The robot runs an agent loop using a VLA model and a Luna agent, collecting telemetry and run data through signals while a governance layer via port.io stores context every few milliseconds or seconds to support future decisions. It also covers using the Frenka 10.7 DOF arm (and the ability to add GLB assets) with a control center for failure analysis and exporting runs for future training."}