Skip to content

combining Equivariant Networks(symm) with RMA #52

Description

@wangqin666

Hello, and thank you so much for your contributions! Recently, I’ve been experimenting with combining Equivariant Networks with RMA. My approach is to feed the equivariant network with both the "raw observations" and the "terrain latent estimated by RMA."
The main challenge I'm facing is that equivariant networks require every observation dimension to have a well-defined symmetry representation, whereas the latent output from RMA doesn't inherently satisfy this. Therefore, I need to assign a specific representation to this latent (e.g., a trivial representation) to ensure that the RMA output complies with the symmetry contract of the equivariant network.
This is my current understanding, and I’d love to know if you think this approach is feasible. Have you tried anything similar before? My ultimate goal is to use RMA to solve the adaptation problem in blind locomotion, as my desired_feet_height is currently just a hardcoded constant. I’d really appreciate any suggestions or insights you might have. Thank you!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions