Is icra_benchmark/rgl.py trained with ground truth position?

Hi,

I've ran training for all the benchmarks and it appears that rgl.py has the best results. Is this trained with ground truth positions (no state predictions)?

Regards,
Hafiq