Task: Stack Bowls

Six same-sized bowls---three red and three yellow---are placed on the table: one red and one yellow at fixed positions near the robot, and the other two red and two yellow in a row farther away. The robot stacks the two far bowls of the instructed color onto the near bowl of that color.

A trial is successful when both far bowls of the instructed color are stacked on the matching near bowl.

As the bowls differ only in color, the task isolates instruction-conditioned instance selection.

π₀
π₀ + Aux-Loss
π₀ + Gaze-Prompt