修改了算法代码,并建立了一个简单的训练脚本.修改bert处理二维输入,移除PPO的permute参数

Env/simple_idm_policy.py

@@ -6,13 +6,8 @@ class ConstantVelocityPolicy:
 
     def act(self):
         self.step_num += 1
-        if self.step_num % 30 < 15:
-            throttle = 1.0
-        else:
-            throttle = 1.0
-
-        steering = 0.1
-
-        # return [steering, throttle]
-
-        return [0.0,0.05]
+        # 简单的前进策略：直行 + 较大油门
+        steering = 0.0  # 直行
+        throttle = 0.5  # 中等油门，让车辆有明显运动
+        
+        return [steering, throttle]