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

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"