Table of contents
Sample Tasks
Below, we show a sampling of the 300 tasks in our benchmark. The container positions are shuffled upon each environment reset, even within the same task. In the natural language instruction, the target container is identified by either its color or position, and the target object is identified by either its name, color, or shape.
Selected Train Tasks
Task #1: "Put fountain vase in green bin." | Task #87: "Put red colored object in red bin." | Task #141: "Put chalice shaped object in front bin." | Task #270: "Put bottle in right bin." |
Selected Test Tasks
Task #32: "Put black and white colored object in green bin." | Task #97: "Put trapezoidal prism shaped object in red bin." | Task #178: "Put box sofa in back bin." | Task #248: "Put cylinder shaped object in left bin." |
Train/Test Splits
All Train and Test Tasks
Scenario A (novel objects, colors, and shapes) trains on all gray tasks and tests on yellow, blue, and green tasks.
Scenario B (novel colors and shapes) trains on all gray and yellow tasks and tests on blue and green tasks.