When training with OmniIsaacGymEnvs, a PyTorch
.pth file is saved. I’m exploring how to implement this in ROS for a real robot. Since OmniIsaacGymEnvs requires Omniverse libraries and the initiation of Isaac Sim, I believe a custom environment tailored for the actual robot is necessary.
I think I can get actions from the model as shown below, but I’m uncertain about how to port the structure of the model used during training to this new environment.
model = ~~~~
model = torch.load('weight.pth')
action = model(obs)