What the model imagined.
What the robot did.
Explore generated futures beside the recorded simulation.
Ten shared worlds per checkpoint. Every prediction window.
The same scenes, at each checkpoint.
World 01 SEED —
Loading recording cadence…
Reward over the rollout.
Click the chart to seek. Reward is measured every 50 actions. A value of 1 means the task is solved.
About these recordings Protocol, timing & interpretation
Closed-loop simulation
Each call sees the latest simulator cameras and measured state, predicts a 48-action block, and executes it. Generated video is saved for inspection and is never fed back as the next observation.
Independent prediction windows
Every forecast contains the current frame followed by eight generated frames. “Loop this forecast” isolates one window. “Play rollout” follows the full simulator timeline and selects its corresponding forecast.
Matching the training contract
Absolute canonical base-frame end-effector actions, ABC-source quantile normalization, canonical grippers, and inverse kinematics. The canvas puts the top camera above the left and right wrist cameras.
Read results at the right scope
Each checkpoint evaluates ten shared worlds. New recordings appear individually as they finish; the ready count shows the scope of the displayed results. All 22,000 actions are budgeted; the last stored simulator image is at action —.