README.md

This uses physical-intelligence's FAST action tokenizer but is only trained on LIBERO dataset. Action horizon = 10 (1 second).