Skip to content

fix: Flashsac for g1_walk_flat - #320

Merged
TATP-233 merged 1 commit into
mainfrom
flahsac_g1
Apr 20, 2026
Merged

TATP-233 merged 1 commit into
mainfrom
flahsac_g1

Conversation

@lemon-star608

@lemon-star608 lemon-star608 commented Apr 20, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • 修复 FlashSAC learner 中三个导致策略完全无法学习的算法级 bug
  • 调整 flashsac训练g1_walk_flat 的超参数,匹配原始 FlashSAC 仓库的 UTD 比例、replay buffer 大小
  • 调整奖励权重优化步态

Linked Work

Validation

  • make check
  • uv run pytest -m "not slow"

Commands actually run:

# 完整训练
uv run python scripts/train_offpolicy.py algo=flashsac task=flashsac/g1_walk_flat/mujoco

Impact

  • Backend impact: mujoco
  • Platform impact: Linux
  • Training effect expected: yes——目前可以稳定收敛

Artifacts:

tensorboard训练曲线:
image
image

Checklist

  • Added or updated tests where needed
  • Updated docs if behavior or workflow changed
  • Linked the driving issue
  • Noted any follow-up work explicitly

@TATP-233
TATP-233 merged commit 9f9dab7 into main Apr 20, 2026
10 checks passed
@TATP-233
TATP-233 deleted the flahsac_g1 branch April 20, 2026 11:44
TATP-233 added a commit that referenced this pull request May 4, 2026
fix: Flashsac for g1_walk_flat
TATP-233 added a commit that referenced this pull request May 30, 2026
fix: Flashsac for g1_walk_flat
TATP-233 added a commit that referenced this pull request Sep 4, 2026
fix: Flashsac for g1_walk_flat
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants