Modified sumo_rl/environment/env.py file

experiments/ppo_4x4grid.py

@@ -20,6 +20,7 @@
 
 
 if __name__ == "__main__":
+    print(os.getcwd())
     ray.init()
 
     env_name = "4x4grid"

experiments/sb3_grid4x4.py

@@ -5,7 +5,6 @@
 import numpy as np
 import supersuit as ss
 import traci
-from pyvirtualdisplay.smartdisplay import SmartDisplay
 from stable_baselines3 import PPO
 from stable_baselines3.common.callbacks import EvalCallback
 from stable_baselines3.common.evaluation import evaluate_policy
@@ -16,34 +15,26 @@
 
 
 if __name__ == "__main__":
-    RESOLUTION = (3200, 1800)
 
-    env = sumo_rl.grid4x4(use_gui=True, out_csv_name="outputs/grid4x4/ppo_test", virtual_display=RESOLUTION)
+    env = sumo_rl.grid4x4(use_gui=False, out_csv_name="outputs/grid4x4/ppo_train")
 
     max_time = env.unwrapped.env.sim_max_time
     delta_time = env.unwrapped.env.delta_time
 
     print("Environment created")
 
     env = ss.pettingzoo_env_to_vec_env_v1(env)
-    env = ss.concat_vec_envs_v1(env, 2, num_cpus=1, base_class="stable_baselines3")
+    env = ss.concat_vec_envs_v1(env, 2, num_cpus=16, base_class="stable_baselines3")
     env = VecMonitor(env)
 
     model = PPO(
         "MlpPolicy",
         env,
         verbose=3,
         gamma=0.95,
-        n_steps=256,
-        ent_coef=0.0905168,
         learning_rate=0.00062211,
-        vf_coef=0.042202,
-        max_grad_norm=0.9,
-        gae_lambda=0.99,
-        n_epochs=5,
-        clip_range=0.3,
         batch_size=256,
-        tensorboard_log="./logs/grid4x4/ppo_test",
+        tensorboard_log="./logs/grid4x4/ppo_train",
     )
 
     print("Starting training")
@@ -55,28 +46,19 @@
     print(mean_reward)
     print(std_reward)
 
+    model.save('ppo_output')
+
     # Maximum number of steps before reset, +1 because I'm scared of OBOE
     print("Starting rendering")
     num_steps = (max_time // delta_time) + 1
 
     obs = env.reset()
 
-    if os.path.exists("temp"):
-        shutil.rmtree("temp")
-
-    os.mkdir("temp")
-    # img = disp.grab()
-    # img.save(f"temp/img0.jpg")
-
     img = env.render()
     for t in trange(num_steps):
         actions, _ = model.predict(obs, state=None, deterministic=False)
         obs, reward, done, info = env.step(actions)
-        img = env.render()
-        img.save(f"temp/img{t}.jpg")
-
-    subprocess.run(["ffmpeg", "-y", "-framerate", "5", "-i", "temp/img%d.jpg", "output.mp4"])
+        env.render()
 
     print("All done, cleaning up")
-    shutil.rmtree("temp")
     env.close()

sumo_rl/environment/env.py

@@ -104,6 +104,7 @@ def __init__(
         sumo_warnings: bool = True,
         additional_sumo_cmd: Optional[str] = None,
         render_mode: Optional[str] = None,
+        ignore_route_errors: Optional[bool] = True
     ) -> None:
         """Initialize the environment."""
         assert render_mode is None or render_mode in self.metadata["render_modes"], "Invalid render mode."
@@ -138,6 +139,7 @@ def __init__(
         self.additional_sumo_cmd = additional_sumo_cmd
         self.add_system_info = add_system_info
         self.add_per_agent_info = add_per_agent_info
+        self.ignore_route_errors = ignore_route_errors
         self.label = str(SumoEnvironment.CONNECTION_LABEL)
         SumoEnvironment.CONNECTION_LABEL += 1
         self.sumo = None
@@ -227,6 +229,8 @@ def _start_simulation(self):
                 self.disp = SmartDisplay(size=self.virtual_display)
                 self.disp.start()
                 print("Virtual display started.")
+        if self.ignore_route_errors:
+            sumo_cmd.append("--ignore-route-errors")
 
         if LIBSUMO:
             traci.start(sumo_cmd)