Skip to content
← Back to Videos

mimic-video: Video-Action Models for Generalizable Robot Control

Loading video
Loading video

mimic-video keeps a pretrained video backbone frozen, extracts latent visual plans, and decodes low-level robot commands. The authors report 10x better sample efficiency and 2x faster convergence than VLA baselines across simulated and real bimanual manipulation.

Category: research
Author: @robotsdigest
Date: 2026-11-18T00:00:00
Duration: 127.893s

As an Amazon Associate, we earn from qualifying purchases.