Workflow from Scientific Research

The world model learns to encode the visual output from the imaging pipeline into a latent representation. It uses an encoder–decoder architecture to reconstruct the images, effectively capturing the latent state ( Z t ) from the input state ( S t ). The world model simulates future states ( S t +1 , S t +2 , …) and actions ( A t , A t +1 , …) in the latent space, facilitating the prediction of microrobot movements.

DOI