General-purpose embodied foundation model for cross-embodiment robot manipulation
Visics is RoboScience's general-purpose embodied foundation model built on an object-centric VLOA (Vision-Language-Object-Action) architecture. An upper embodied world model simulates physical trajectories and a lower generalist manipulation model turns them into control signals for different robots, aiming to generalise across embodiments, objects and tasks, with pretraining on internet video and simulation data.