Merge branch 'main' into deepsource-autofix-5c035f8a

model-gym-ros-env/car_node/car_node/main.py

@@ -29,7 +29,7 @@
 class PPOModelEvaluator(Node):
 
     def __init__(self):
-        super().__init__('wall_follow_node')
+        super().__init__('car_node')
 
         #Topics & Subs, Pubs
         lidarscan_topic = '/scan'

model-gym-ros-env/car_node/car_node/wrapper.py

@@ -1,6 +1,5 @@
 import gym
 import numpy as np
-#from gym import spaces
 from gym import spaces
 from pathlib import Path
 from argparse import Namespace
@@ -157,11 +156,6 @@ def step(self, action):
             np.power((observation["poses_y"][0] - self.last_position['y']), 2)
                                             )
 
-        # if distance_from_last_position > 0.0005:
-        #     reward += 0.8
-        # else:
-        #     reward = 0
-        # # Update the last position
         reward += distance_from_last_position
 
         self.last_position['x'] = observation['poses_x'][0]
@@ -228,8 +222,6 @@ def reset(self):
         self.race_line_y = race_line_y
         self.race_line_theta = race_line_theta
 
-        # else:
-        #     x, y, t = self.start_position()
 
         self.episode_returns = []
 
@@ -243,9 +235,6 @@ def update_map(self, map_name, map_extension, update_render=False):
         self.env.map_name = map_name
         self.env.map_ext = map_extension
         self.env.update_map(f"{map_name}.yaml", map_extension)
-        # if update_render and self.env.renderer:
-        #     self.env.renderer.close()
-        #     self.env.renderer = None
 
     def seed(self, seed):
         self.current_seed = seed

report/bibliography.bib

@@ -58,7 +58,7 @@ @misc{OpenAIGym
   eprint = {arXiv:1606.01540}
 }
 
-@article{ROS2,
+@article{Ros2,
   author  = {Steven Macenski and Tully Foote and Brian Gerkey and Chris Lalancette and William Woodall},
   title   = {Robot Operating System 2: Design, architecture, and uses in the wild},
   journal = {Science Robotics},
@@ -69,3 +69,12 @@ @article{ROS2
   doi     = {10.1126/scirobotics.abm6074},
   url     = {https://www.science.org/doi/abs/10.1126/scirobotics.abm6074}
 }
+
+@inproceedings{F1tenthGym,
+  title        = {F1TENTH: An Open-source Evaluation Environment for Continuous Control and Reinforcement Learning},
+  author       = {O’Kelly, Matthew and Zheng, Hongrui and Karthik, Dhruv and Mangharam, Rahul},
+  booktitle    = {NeurIPS 2019 Competition and Demonstration Track},
+  pages        = {77--89},
+  year         = {2020},
+  organization = {PMLR}
+}

src/main.py

@@ -3,7 +3,7 @@
 import random
 import sys
 
-from train import train, run
+from train import train, run, evaluate
 
 
 base_path = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
@@ -13,7 +13,8 @@ def get_available_maps():
     dir_content = os.listdir(map_path)
     maps = []
     for item in dir_content:
-        if os.path.isdir(os.path.join(map_path, item)):
+        if os.path.isdir(os.path.join(map_path, item)) and \
+            os.path.exists(os.path.join(map_path, item, item + "_raceline.csv")):
             maps.append(item)
     return maps
 
@@ -39,12 +40,18 @@ def check_map(map_name):
     train_parser.add_argument('--min-learning-rate', type=float, help='Learning rate for the model. Default: 0.0005', default=0.0005)
     train_parser.add_argument('--max-learning-rate', type=float, help='Learning rate for the model. Default: 0.0001', default=0.0001)
     train_parser.add_argument('--num-of-steps', type=int, help='Number of steps to train the model. Default: 10', default=10)
+    train_parser.add_argument('--num-of-sub-steps', type=int, help='Number of steps for each traing cicle. Default: 10', default=10)
     train_parser.add_argument('--optimize-speed', action='store_true', help='Save the model to the specified path. Default: None')
 
     run_parser = subparser.add_parser('run', help='Run the model in the simulator')
     run_parser.add_argument('-m', '--map', type=str, help='Map to train the model on. Default: random maps', default='random')
     run_parser.add_argument('--timesteps', type=float, help='Number of timesteps to train the model. Default: 0.01', default=0.01)
 
+    evaluate_parser = subparser.add_parser('evaluate', help='Evaluate the model')
+    evaluate_parser.add_argument('-m', '--map', type=str, help='Map to train the model on. Default: random maps', default='random')
+    evaluate_parser.add_argument('--timesteps', type=float, help='Number of timesteps to train the model. Default: 0.01', default=0.01)
+    evaluate_parser.add_argument('--n-evaluate', type=int, help='Number of timesteps to train the model. Default: 20', default=20)
+
     args = parser.parse_args()
 
     available_maps = get_available_maps()
@@ -65,6 +72,7 @@ def check_map(map_name):
                 args.min_learning_rate,
                 args.max_learning_rate,
                 args.num_of_steps,
+                args.num_of_sub_steps,
                 args.optimize_speed)
         sys.exit(0)
     elif args.command == 'run':
@@ -74,6 +82,16 @@ def check_map(map_name):
         check_map(args.map)
         run(args.map, args.timesteps)
         sys.exit(0)
+    elif args.command == 'evaluate':
+        print("Evaluating the model")
+        if 'random' not in args.map:
+            check_map(args.map)
+        evaluate(
+            args.map == 'random',
+            random.choice(available_maps) if args.map == 'random' else args.map,
+            args.timesteps,
+            args.n_evaluate)
+        sys.exit(0)
     else:
         print("Invalid command")
         sys.exit(1)

src/train.py

@@ -18,6 +18,7 @@ def train(
         min_learning_rate: float,
         max_learning_rate: float,
         num_of_steps: int,
+        substeps: int,
         optimize_speed: bool) -> None:
 
     tensorboard_path = './train_test/'
@@ -28,7 +29,7 @@ def train(
     path = get_map(map)
     map_path = get_formatted_map(path)
     eval_env  = gym.make('f110_gym:f110-v0', map=map_path, map_ext=map_ext, num_agents=1, timestep=timestep, integrator=Integrator.RK4)
-    eval_env = F110_Wrapped(eval_env, random_map=True)
+    eval_env = F110_Wrapped(eval_env, random_map=random_map)
     eval_env.set_map_path(path)
     eval_env.seed(1773449316)
     eval_env.set_optimize_speed(optimize_speed)
@@ -67,7 +68,7 @@ def train(
                 model.save("./src/base_model")   
 
         eval_callback = EvalCallback(eval_env, best_model_save_path='./train_test/',
-                                    log_path='./train_test/', eval_freq= int(timesteps/20),
+                                    log_path='./train_test/', eval_freq= int(timesteps/substeps),
                                     deterministic=True, render=False)
 
         model.learn(total_timesteps=timesteps, progress_bar=True, callback=eval_callback)
@@ -94,6 +95,24 @@ def train(
 
         del model
 
+def evaluate(random_map: bool, map: str, timestep: int, n_evaluate: int) -> None:
+    device = 'cpu'
+
+    if random_map:
+        map = "BrandsHatch"
+    path = get_map(map)
+    map_path = get_formatted_map(path)
+
+    eval_env  = gym.make('f110_gym:f110-v0', map=map_path, map_ext=map_ext, num_agents=1, timestep=timestep, integrator=Integrator.RK4)
+    eval_env = F110_Wrapped(eval_env, random_map=random_map)
+    eval_env.set_map_path(path)
+
+    model = PPO.load("./train_test/best_global_model", eval_env, device=device)
+
+    mean_reward, std_reward = evaluate_policy(model, model.get_env(), n_eval_episodes=n_evaluate)
+
+    print(mean_reward)
+    print(std_reward)
 
 def run(map: str, timestep: float) -> None: