change_target

scripts/gyms/sprint.py

@@ -280,12 +280,16 @@ class sprint(gym.Env):
         self.player.terminate()
 
     def change_target(self):
-        # 生成一个 -10 到 +10 度之间的随机角度
+        original_angle = M.vector_angle(self.walk_rel_target)
-        orientations = random.choice([-1, 1])
+        random_angle_delta = np.random.uniform(-10, 10)
-        random_target = orientations * 10 # 单位是度
+        new_angle = (original_angle + np.radians(random_angle_delta)) * 3 * math.sin(time.time())
 
-        self.walk_rel_target = (15, random_target)
+        new_walk_rel_target = np.array([
+            np.cos(new_angle) * self.walk_distance,
+            np.sin(new_angle) * self.walk_distance
+        ])
 
+        self.walk_rel_target = new_walk_rel_target
 
     def step(self, action):
 
@@ -294,7 +298,7 @@ class sprint(gym.Env):
 
         w = self.player.world
         current_time = time.time()
-        if current_time - self.last_target_update_time > 0.8:
+        if current_time - self.last_target_update_time > 0.6:
             self.change_target()
             self.last_target_update_time = current_time
 
@@ -342,19 +346,16 @@ class sprint(gym.Env):
         reward = robot_speed * (1 - direction_error / 10)
         if self.player.behavior.is_ready("Get_Up"):
             self.terminal = True
-
+        elif w.time_local_ms - self.reset_time > 30000:
-        elif w.time_local_ms - self.reset_time > 10000 * 2:
             self.terminal = True
         elif r.loc_torso_position[0] > 14.5:
             self.terminal = True
             reward += 500
         elif r.loc_torso_position[0] > 0:
-            reward += r.loc_torso_position[0] * 10
+            reward += 3 * r.loc_torso_position[0]
         else:
             self.terminal = False
         return obs, reward, self.terminal, {}
-
-
 class Train(Train_Base):
     def __init__(self, script) -> None:
         super().__init__(script)