Trained policies can be loaded using the checkpoint cmdline argument: e.g. checkpoint=runs/Anymal/nn/Anymal.pth using the same python script used for training. If you’d like to run inferencing on with the policy, you an also add the test=True argument.