Some details about dp experiment

[Selen-Suyue]

When testing the DP effect, which vision model architecture did you choose, CNN or transformer?

[YanjieZe]

Hi, thank you for your interest. We use the original vision arch in Diffusion Policy, which is a CNN,

[Selen-Suyue]

Hello, could you please tell me if the CNN you used here is based on the Resnet series architecture or if you defined your own convolution blocks? I noticed in the DP3 experiments that a deeper visual model seems to affect the quality of the diffused actions.