Refactor gym to gymnasium

Fixes Update `gym` to `gymnasium`. #8

.vscode/extensions.json

@@ -4,6 +4,7 @@
         "github.copilot",
         "genieai.chatgpt-vscode",
         "ms-python.python",
-        "ms-python.pylint"
+        "ms-python.pylint",
+        "github.vscode-pull-request-github"
     ]
 }

README.md

@@ -4,7 +4,8 @@
 This project provides a training environment for the simulation of a robot moving
 in a pedestrian-filled space.
 
-The project interfaces with OpenAI Gym to facilitate trainings with various
+The project interfaces with Faram Foundations "Gymnasium" (former OpenAI Gym)
+to facilitate trainings with various
 SOTA reinforcement learning algorithms like e.g. StableBaselines3.
 For simulating the pedestrians, the SocialForce model is used via a dependency
 on a fork of PySocialForce.

demo_defensive.py

@@ -1,14 +1,13 @@
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 from stable_baselines3 import PPO
 
 from robot_sf.robot_env import RobotEnv, OBS_DRIVE_STATE, OBS_RAYS
 from robot_sf.sim_config import EnvSettings
 from robot_sf.sim.sim_config import SimulationSettings
 from robot_sf.robot.differential_drive import DifferentialDriveSettings
 
-
-def training():
+
     env_config = EnvSettings(
         sim_config=SimulationSettings(stack_steps=1, difficulty=0, ped_density_by_difficulty=[0.06]),
         robot_config=DifferentialDriveSettings(radius=1.0))

requirements.txt

@@ -13,7 +13,7 @@ fonttools==4.49.0
 fsspec==2024.2.0
 greenlet==3.0.3
 grpcio==1.60.1
-gym==0.26.2
+# gym==0.26.2
 gym-notices==0.0.8
 gymnasium==0.29.1
 iniconfig==2.0.0

robot_sf/feature_extractor.py

@@ -4,7 +4,7 @@
 from typing import List
 
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 import torch as th
 from torch import nn
 from stable_baselines3.common.torch_layers import BaseFeaturesExtractor

robot_sf/robot/bicycle_drive.py

@@ -3,7 +3,7 @@
 from typing import Tuple
 
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 
 
 Vec2D = Tuple[float, float]

robot_sf/robot/differential_drive.py

@@ -4,7 +4,7 @@
 from typing import Tuple
 
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 
 
 Vec2D = Tuple[float, float]
@@ -207,7 +207,7 @@ def observation_space(self) -> spaces.Box:
         Defines the observation space for the robot based on its configuration.
 
         Returns:
-            Box: An instance of gym.spaces.Box representing the continuous
+            Box: An instance of gymnasium.spaces.Box representing the continuous
                  observation space where each observation is a vector containing
                  linear and angular speeds.
         """
@@ -223,7 +223,7 @@ def action_space(self) -> spaces.Box:
         Defines the action space for the robot based on its configuration.
 
         Returns:
-            Box: An instance of gym.spaces.Box representing the continuous
+            Box: An instance of gymnasium.spaces.Box representing the continuous
                  action space where each action is a vector containing
                  linear and angular speeds.
         """

robot_sf/robot_env.py

@@ -15,7 +15,7 @@
 timeout condition, simulation time elapsed, and timestep count.
 
 3. `RobotEnv`: A class that represents the robot's environment. It inherits from `VectorEnv`
-from the `gym` library, which is a base class for environments that operate over
+from the `gymnasium` library, which is a base class for environments that operate over
 vectorized actions and observations. It includes methods for stepping through the environment,
 resetting it, rendering it, and closing it.
 It also defines the action and observation spaces for the robot.
@@ -31,9 +31,9 @@
 from multiprocessing.pool import ThreadPool
 
 import numpy as np
-from gym.vector import VectorEnv
-from gym import Env, spaces
-from gym.utils import seeding
+from gymnasium.vector import VectorEnv
+from gymnasium import Env, spaces
+from gymnasium.utils import seeding
 from robot_sf.nav.map_config import MapDefinition
 from robot_sf.nav.navigation import RouteNavigator
 
@@ -373,7 +373,7 @@ def init_spaces(env_config: EnvSettings, map_def: MapDefinition):
 
 class RobotEnv(Env):
     """
-    Representing an OpenAI Gym environment for training a self-driving robot
+    Representing a Gymnasium environment for training a self-driving robot
     with reinforcement learning.
     """
 
@@ -562,7 +562,7 @@ def exit(self):
 
 
 class MultiRobotEnv(VectorEnv):
-    """Representing an OpenAI Gym environment for training
+    """Representing a Gymnasium environment for training
     multiple self-driving robots with reinforcement learning"""
 
     def __init__(

robot_sf/sensor/goal_sensor.py

@@ -2,7 +2,7 @@
 from typing import Tuple, Union
 
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 
 Vec2D = Tuple[float, float]
 PolarVec2D = Tuple[float, float]

robot_sf/sensor/range_sensor.py

@@ -4,7 +4,7 @@
 
 import numpy as np
 import numba
-from gym import spaces
+from gymnasium import spaces
 
 from robot_sf.nav.occupancy import ContinuousOccupancy
 

robot_sf/sensor/sensor_fusion.py

@@ -6,7 +6,7 @@
 from dataclasses import dataclass, field
 
 import numpy as np
-from gym import spaces
+from gymnasium import spaces
 
 PolarVec2D = Tuple[float, float]
 

scripts/evaluate.py

@@ -3,8 +3,8 @@
 import json
 
 import numpy as np
-import gym
-from gym import spaces
+import gymnasium
+from gymnasium import spaces
 from tqdm import tqdm
 from stable_baselines3 import PPO, A2C
 
@@ -56,7 +56,7 @@ class EvalSettings:
 
 
 @dataclass
-class AdaptedEnv(gym.Env):
+class AdaptedEnv(gymnasium.Env):
     orig_env: RobotEnv
     config: GymAdapterSettings
 
@@ -78,7 +78,7 @@ def reset(self):
         return self.config.obs_adapter(obs)
 
 
-def evaluate(env: gym.Env, model: DriveModel, num_episodes: int) -> EnvMetrics:
+def evaluate(env: gymnasium.Env, model: DriveModel, num_episodes: int) -> EnvMetrics:
     eval_metrics = EnvMetrics(cache_size=num_episodes)
 
     for _ in tqdm(range(num_episodes)):
@@ -97,7 +97,7 @@ def evaluate(env: gym.Env, model: DriveModel, num_episodes: int) -> EnvMetrics:
     return eval_metrics
 
 
-def prepare_env(settings: EvalSettings, difficulty: int) -> gym.Env:
+def prepare_env(settings: EvalSettings, difficulty: int) -> gymnasium.Env:
     env_settings = EnvSettings()
     env_settings.sim_config.prf_config = settings.prf_config
     env_settings.sim_config.ped_density_by_difficulty = settings.ped_densities
@@ -108,7 +108,7 @@ def prepare_env(settings: EvalSettings, difficulty: int) -> gym.Env:
     return AdaptedEnv(orig_env, settings.gym_config)
 
 
-def prepare_model(model_path: str, env: gym.Env) -> DriveModel:
+def prepare_model(model_path: str, env: gymnasium.Env) -> DriveModel:
     return A2C.load(model_path, env=env)
 
 

scripts/training_ppo.py

@@ -41,7 +41,7 @@ def make_env():
         )
 
     model.learn(
-        total_timesteps=1_000_000,
+        total_timesteps=10_000_000,
         progress_bar=True,
         callback=combined_callback
         )

setup.py

@@ -5,15 +5,15 @@
 PACKAGE_NAME = "robot_sf"
 PACKAGE_VERSION = "2.0.0"
 PACKAGE_AUTHORS = "Marco Tröster"
-PACKAGE_DESCRIPTION = """This package allows implementing a "gym-style" environment
+PACKAGE_DESCRIPTION = """This package allows implementing a "gymnasium-style" environment
 for navigating a crowd with autonomous micromobility vehicles
 """
 HOME_REPO = "https://github.com/Bonifatius94/robot_env"
 EXCLUDE_FILES = []
 PACKAGE_DATA = {"robot_sf": ["maps/*.json"]}
-INSTALL_REQUIREMENTS = ["numpy", "gym", "pylint", "pytest", "scalene", "numba", "pygame",
+INSTALL_REQUIREMENTS = ["numpy", "gymnasium", "pylint", "pytest", "scalene", "numba", "pygame",
                         "stable-baselines3", "tqdm", "rich", "tensorboard"]
-
+# TODO Update this package information
 
 def get_ext_paths(root_dir, exclude_files):
     paths = []

tests/env_test.py

@@ -1,4 +1,4 @@
-from gym import spaces
+from gymnasium import spaces
 from robot_sf.robot_env import RobotEnv, OBS_DRIVE_STATE, OBS_RAYS