Skip to content

Local Minimas #2

@kapsl

Description

@kapsl

Hi,
thanks for your code! I tested it and it worked pretty well! Basically we have a 7 DOF Robot trying to solve a peg in hole task. It looks like this

https://sites.google.com/view/rl-wo-dynamics-randomization

You see that if the target is inside the peg, being outside of the peg is a local minimum, you would need to go up first and then insert. A mode free approach like SAC manages to do this, but your method fails, which makes sense, as the reward is taken directly from the reward function - which cannot lead it out of the local minimum. Do you know a solution for this?

Another question: What is the score vs. reward in your code? Isnt't the score the same as reward?

lg Manuel

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions