In the github repo of TF Agents there are three driver classes for stepping through tensorflow environments using tensorflow policies. The first two are dynamic_step_drive
dynamic_step_drive