Clarification on the calculation of 166 variations in RLBench tasks

[zhanglixuan0720]

Hi,

I was going through the paper, and I noticed the following statement:

"Following [76], we utilize a curated subset of 10 challenging language-conditioned manipulation tasks from RLBench, which includes 166 variations in object properties and scene arrangement."

I would like to clarify how the 166 variations are computed. Specifically, are these variations based on a combination of different object properties (such as color, size, shape) and scene arrangements (e.g., positions, orientations)? Could you provide more details on the exact methodology for calculating these 166 variations?