diff --git a/motrix_rl/src/motrix_rl/rslrl/torch/train/dreamwaq_ppo.py b/motrix_rl/src/motrix_rl/rslrl/torch/train/dreamwaq_ppo.py
index 6f9050d..5a5afa9 100644
--- a/motrix_rl/src/motrix_rl/rslrl/torch/train/dreamwaq_ppo.py
+++ b/motrix_rl/src/motrix_rl/rslrl/torch/train/dreamwaq_ppo.py
@@ -158,9 +158,9 @@ class DreamWaQPPO(PPO):
         old_actions_log_prob_batch, advantages_batch
     ) -> torch.Tensor:
         """计算 PPO 代理损失（从父类 PPO.update() 中提取）。"""
-        ratio = torch.exp(actions_log_prob_batch - old_actions_log_prob_batch)
-        surrogate = -advantages_batch * ratio
-        surrogate_clipped = -advantages_batch * torch.clamp(
+        ratio = torch.exp(actions_log_prob_batch - torch.squeeze(old_actions_log_prob_batch))
+        surrogate = -torch.squeeze(advantages_batch) * ratio
+        surrogate_clipped = -torch.squeeze(advantages_batch) * torch.clamp(
             ratio, 1.0 - self.clip_param, 1.0 + self.clip_param)
         return torch.max(surrogate, surrogate_clipped).mean()