new demo

scripts/gyms/sprint_demo.py

@@ -44,8 +44,8 @@ class sprint(gym.Env):
 
         # Step behavior defaults
         self.STEP_DUR = 10
-        self.STEP_Z_SPAN = 0.2
-        self.STEP_Z_MAX = 0.7
+        self.STEP_Z_SPAN = 0.4
+        self.STEP_Z_MAX = 0.8
         nao_specs = self.ik.NAO_SPECS
         self.leg_length = nao_specs[1] + nao_specs[3]  # upper leg height + lower leg height
         feet_y_dev = nao_specs[0] * 2  # wider step
@@ -298,7 +298,7 @@ class sprint(gym.Env):
 
         w = self.player.world
         current_time = time.time()
-        if current_time - self.last_target_update_time > 1:
+        if current_time - self.last_target_update_time > 2:
             self.change_target()
             self.last_target_update_time = current_time
 
@@ -310,7 +310,7 @@ class sprint(gym.Env):
         self.walk_rel_orientation = M.vector_angle(self.walk_rel_target) * 0.3
 
         # exponential moving average
-        self.act = 0.7 * self.act + 0.3 * action_mult * 0.7
+        self.act = 0.8 * self.act + 0.2 * 0.7
 
         # execute Step behavior to extract the target positions of each leg (we will override these targets)
         lfy, lfz, rfy, rfz = self.step_generator.get_target_positions(self.step_counter == 0, self.STEP_DUR,
@@ -343,16 +343,14 @@ class sprint(gym.Env):
         robot_speed = r.loc_torso_velocity[0]
         direction_error = abs(self.walk_rel_orientation)
         direction_error = min(direction_error, 10)
-        reward = robot_speed * (1 - direction_error / 10)
+        reward = robot_speed * (1.5 - 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 > 50000:
             self.terminal = True
         elif r.loc_torso_position[0] > 14.5:
             self.terminal = True
             reward += 500
-        elif r.loc_torso_position[0] > 0:
-            reward += 3 * r.loc_torso_position[0]
         else:
             self.terminal = False
         return obs, reward, self.terminal, {}