Visuotactile pretraining with human data leads to robust manipulation policies trained in simulation.
Sudharshan Suresh (Wed,) studied this question.