Graph networks as learnable physics engines for inference and control

Open in new window