MaskedVisualActions Turns Robot Actions into Pixel Trajectories

A joint research team introduced MaskedVisualActions, a pixel-space action interface that renders robot actions as pixel-level trajectories for video models to directly perceive. This approach boosts robot task success rates by up to 26 percentage points on the RoboCasa benchmark.

2026-07-22 ~ 2026-07-23 · 2 related posts

• A 15-hour video fine-tune turns masked trajectories into a robot control interface — Hadi Alzayer · 2026-07-22
• MaskedVisualActions turns robot motion into pixels and lifts RoboCasa success by up to 26 points — 机器之心 · 2026-07-23