cyy/Motrixlab
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: terrain start level 0, clear ep_report after consume

Browse Source
This commit is contained in:
8x54zj-m
2026-06-30 15:43:02 +08:00
parent 6348cf72bc
commit 361502a8a5
3 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
motrix_rl/src/motrix_rl/rslrl/torch/wrap_vec_env.py
View File

@@ -127,9 +127,10 @@ class RslrlNpEnvWrap(VecEnv):
if "time_outs" in state.info:
extras["time_outs"] = torch.from_numpy(state.info["time_outs"]).to(self._device)
# 将 episode 各项奖励传入 TensorBoard
# 将 episode 各项奖励传入 TensorBoard（消费后清除，防止重复上报）
if "ep_report" in state.info:
extras["episode"] = state.info["ep_report"]
extras["episode"] = dict(state.info["ep_report"])
del state.info["ep_report"]
return obs, rewards, dones, extras

2
motrix_rl/src/motrix_rl/tasks/go1.py
View File

@@ -135,7 +135,7 @@ class rslrl:
runner.actor.class_name = (
"motrix_rl.rslrl.torch.models.cenet_actor:CENetActorModel")
runner.actor.hidden_dims = [512, 256, 128]
runner.actor.init_noise_std = 0.5 # 降噪声先学站稳，再探索行走
runner.actor.init_noise_std = 1.0 # 上游原值
# Critic：标准 MLPModel，输入 privileged_obs
runner.critic.class_name = "MLPModel"