Skip to content

teacher policy evaluation performance #44

Description

@ZZQ0316

Hello, is there any friend successfully trained teacher policy on whole Amass dataset and got an acceptable result? I just trained it for nearly 100k steps (num_envs=4096) on ACCAD, a subset of Amass. But when i evaluated the trained teacher policy using the following command, the success rate could only reach to 60%.
${ISAACLAB_PATH:?}/isaaclab.sh -p scripts/rsl_rl/play.py
--num_envs 1024
--reference_motion_path third_party/human2humanoid/data/h1/amass_all.pkl
--teacher_policy.resume_path ckpts/teacher/1
--teacher_policy.checkpoint model_100000.pt
I tried serval checkpoints saved between 50k to 100k, but the success rates were all below to 60%.

Image

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