fix: dual-shuffle bug in terrain_origins — robots now spawn on correct terrain columns

motrix_envs/src/motrix_envs/locomotion/go1/dreamwaq.py

@@ -410,11 +410,16 @@ class DreamWaQTask(Go1WalkTask):
 
         # 游戏启发式地形课程（per-env，与上游一致）
         if not hasattr(self, '_terrain_origins'):
-            all_levels = np.repeat(np.arange(self._num_rows), self._num_cols)
-            all_indices = np.tile(np.arange(self._num_cols), self._num_rows)
-            all_origins = self._make_origins(all_levels, all_indices)
-            self._terrain_origins = all_origins.reshape(self._num_rows, self._num_cols, 2)
-            self._max_init_level = 0  # 从平地起步，靠课程升级
+            all_origins = np.zeros((self._num_rows, self._num_cols, 2), dtype=np.float32)
+            half_x = self._border + self._num_cols * self._cell_size / 2.0
+            half_y = self._border + self._num_rows * self._cell_size / 2.0
+            for row in range(self._num_rows):
+                for col in range(self._num_cols):
+                    cx = -half_x + self._border + col * self._cell_size + self._cell_size / 2
+                    cy = half_y - self._border - row * self._cell_size - self._cell_size / 2
+                    all_origins[row, col] = [cx, cy]
+            self._terrain_origins = all_origins
+            self._max_init_level = 5  # 上游原值，随机 0-5 起步
 
         if num_reset > 0 and self._init_done and state is not None and hasattr(state, 'info'):
             old_info = state.info

motrix_rl/src/motrix_rl/tasks/go1.py

@@ -112,7 +112,7 @@ class rslrl:
 
             # Runner 设置（严格对齐上游 LeggedRobotCfgPPO + Go1RoughCfgPPO）
             runner.seed = 5                     # 上游 seed=5
-            runner.max_iterations = 3000
+            runner.max_iterations = 5000  # 2048 envs 需要更多迭代补偿采样量
             runner.num_steps_per_env = 24
             runner.experiment_name = "go1_dreamwaq_walk"
             runner.save_interval = 50
@@ -126,7 +126,7 @@ class rslrl:
             runner.algorithm.entropy_coef = 0.01      # 上游 Go1RoughCfgPPO
             runner.algorithm.desired_kl = 0.01        # 上游 0.01 (默认 0.008)
             runner.algorithm.clip_param = 0.2
-            runner.algorithm.schedule = "fixed"   # 固定 schedule，防止 noise_std 发散
+            runner.algorithm.schedule = "adaptive"  # 上游原值，前期 bug 已修
             runner.algorithm.gamma = 0.99
             runner.algorithm.lam = 0.95
             runner.algorithm.max_grad_norm = 1.0
@@ -136,7 +136,7 @@ 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  # 降低探索噪声，1024 envs 不需要太高
+            runner.actor.init_noise_std = 1.0  # 上游原值，adaptive 会自行调节
 
             # Critic：标准 MLPModel，输入 privileged_obs
             runner.critic.class_name = "MLPModel"