Add support for Gymnasium v1.0

.github/workflows/tests.yml

@@ -30,7 +30,7 @@ jobs:
       matrix:
         os: [ubuntu-latest]
         # TODO: add Python 3.11 to test matrix
-        python-version: ["3.8", "3.9", "3.10"]
+        python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
       fail-fast: false
     timeout-minutes: 60
     steps:

pyproject.toml

@@ -20,15 +20,16 @@ classifiers = [
     "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
     "Intended Audience :: Science/Research",
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
     "Topic :: Software Development :: Libraries :: Python Modules",
 ]
 dependencies = [
-    "gymnasium == 0.28.1",
-    "gymnasium-robotics == 1.2.2",
-    "pygame == 2.1.0",
-    "mujoco == 2.3.3",
+    "gymnasium >= 0.28.1",
+    "gymnasium-robotics >= 1.2.2",
+    "pygame >= 2.1.0",
+    "mujoco >= 2.3.3",
     "xmltodict >= 0.13.0",
     "pyyaml >= 6.0",
     "imageio >= 2.27.0",
@@ -74,7 +75,7 @@ safe = true
 line-length = 100
 skip-string-normalization = true
 # Sync with requires-python
-target-version = ["py38", "py39", "py310", "py311"]
+target-version = ["py38", "py39", "py310", "py311", "py312"]
 
 [tool.isort]
 atomic = true

safety_gymnasium/assets/free_geoms/push_box.py

@@ -38,7 +38,7 @@ class PushBox(FreeGeom):  # pylint: disable=too-many-instance-attributes
     reward_box_dist: float = 1.0  # Dense reward for moving the agent towards the box
     reward_box_goal: float = 1.0  # Reward for moving the box towards the goal
 
-    color: np.array = COLOR['push_box']
+    color: np.array = field(default_factory=lambda: COLOR['push_box'])
     alpha: float = 0.25
     group: np.array = GROUP['push_box']
     is_lidar_observed: bool = True

safety_gymnasium/assets/free_geoms/vases.py

@@ -33,7 +33,7 @@ class Vases(FreeGeom):  # pylint: disable=too-many-instance-attributes
     placements: list = None  # Vases placements list (defaults to full extents)
     locations: list = field(default_factory=list)  # Fixed locations to override placements
     keepout: float = 0.15  # Radius of vases keepout for placement
-    alpha: float = COLOR['vase'][-1]
+    alpha: float = field(default_factory=lambda: COLOR['vase'][-1])
     sink: float = 4e-5  # Experimentally measured, based on size and density,
     # how far vases "sink" into the floor.
     # Mujoco has soft contacts, so vases slightly sink into the floor,
@@ -48,7 +48,7 @@ class Vases(FreeGeom):  # pylint: disable=too-many-instance-attributes
     velocity_cost: float = 1.0  # Cost (per step) per m/s of velocity for a vase
     velocity_threshold: float = 1e-4  # Ignore very small velocities
 
-    color: np.array = COLOR['vase']
+    color: np.array = field(default_factory=lambda: COLOR['vase'])
     group: np.array = GROUP['vase']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/assets/geoms/apples.py

@@ -44,7 +44,7 @@ class Apples(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['apple']
+    color: np.array = field(default_factory=lambda: COLOR['apple'])
     alpha: float = 0.25
     group: np.array = GROUP['apple']
     is_lidar_observed: bool = True

safety_gymnasium/assets/geoms/buttons.py

@@ -45,7 +45,7 @@ class Buttons(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['button']
+    color: np.array = field(default_factory=lambda: COLOR['button'])
     group: np.array = GROUP['button']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/assets/geoms/circle.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 """Circle."""
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 import numpy as np
 
@@ -33,7 +33,7 @@ class Circle(Geom):  # pylint: disable=too-many-instance-attributes
     locations: tuple = ((0, 0),)
     keepout: float = 0.0
 
-    color: np.array = COLOR['circle']
+    color: np.array = field(default_factory=lambda: COLOR['circle'])
     alpha: float = 0.1
     group: np.array = GROUP['circle']
     is_lidar_observed: bool = True

safety_gymnasium/assets/geoms/goal.py

@@ -39,7 +39,7 @@ class Goal(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.ndarray = COLOR['goal']
+    color: np.ndarray = field(default_factory=lambda: COLOR['goal'])
     alpha: float = 0.25
     group: np.ndarray = GROUP['goal']
     is_lidar_observed: bool = True

safety_gymnasium/assets/geoms/hazards.py

@@ -36,7 +36,7 @@ class Hazards(Geom):  # pylint: disable=too-many-instance-attributes
     alpha: float = 0.25
     cost: float = 1.0  # Cost (per step) for violating the constraint
 
-    color: np.array = COLOR['hazard']
+    color: np.array = field(default_factory=lambda: COLOR['hazard'])
     group: np.array = GROUP['hazard']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/assets/geoms/oranges.py

@@ -44,7 +44,7 @@ class Oranges(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['orange']
+    color: np.array = field(default_factory=lambda: COLOR['orange'])
     alpha: float = 0.25
     group: np.array = GROUP['orange']
     is_lidar_observed: bool = True

safety_gymnasium/assets/geoms/pillars.py

@@ -36,7 +36,7 @@ class Pillars(Geom):  # pylint: disable=too-many-instance-attributes
     keepout: float = 0.3  # Radius for placement of pillars
     cost: float = 1.0  # Cost (per step) for being in contact with a pillar
 
-    color: np.array = COLOR['pillar']
+    color: np.array = field(default_factory=lambda: COLOR['pillar'])
     alpha: float = 1.0
     group: np.array = GROUP['pillar']
     is_lidar_observed: bool = True

safety_gymnasium/assets/geoms/sigwalls.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 """Hazard."""
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 import numpy as np
 
@@ -37,7 +37,7 @@ class Sigwalls(Geom):  # pylint: disable=too-many-instance-attributes
     placements: list = None
     keepout: float = 0.0
 
-    color: np.array = COLOR['sigwall']
+    color: np.array = field(default_factory=lambda: COLOR['sigwall'])
     alpha: float = 0.1
     group: np.array = GROUP['sigwall']
     is_lidar_observed: bool = False

safety_gymnasium/assets/geoms/walls.py

@@ -36,7 +36,7 @@ class Walls(Geom):  # pylint: disable=too-many-instance-attributes
     locations: list = field(default_factory=list)  # This should be used and length == walls_num
     keepout: float = 0.0  # This should not be used
 
-    color: np.array = COLOR['wall']
+    color: np.array = field(default_factory=lambda: COLOR['wall'])
     group: np.array = GROUP['wall']
     is_lidar_observed: bool = True
     is_constrained: bool = False

safety_gymnasium/assets/mocaps/gremlins.py

@@ -39,7 +39,7 @@ class Gremlins(Mocap):  # pylint: disable=too-many-instance-attributes
     dist_cost: float = 1.0  # Cost for being within distance threshold
     density: float = 0.001
 
-    color: np.array = COLOR['gremlin']
+    color: np.array = field(default_factory=lambda: COLOR['gremlin'])
     alpha: float = 1
     group: np.array = GROUP['gremlin']
     is_lidar_observed: bool = True

safety_gymnasium/assets/robot.py

@@ -34,7 +34,7 @@ class Robot:  # pylint: disable=too-many-instance-attributes
     placements: list = None  # Robot placements list (defaults to full extents)
     locations: list = field(default_factory=list)  # Explicitly place robot XY coordinate
     keepout: float = 0.4  # Needs to be set to match the robot XML used
-    base: str = 'assets/xmls/car.xml'  # Which robot XML to use as the base
+    base: str = os.path.join('assets', 'xmls', 'car.xml')  # Which robot XML to use as the base
     rot: float = None  # Override robot starting angle
 
     def __post_init__(self, path) -> None:

safety_gymnasium/bases/base_agent.py

@@ -173,7 +173,7 @@ def __init__(  # pylint: disable=too-many-arguments
             keepout (float): Needs to be set to match the agent XML used.
             rot (float): Override agent starting angle.
         """
-        self.base: str = f'assets/xmls/{name.lower()}.xml'
+        self.base: str = os.path.join('assets', 'xmls', f'{name.lower()}.xml')
         self.random_generator: RandomGenerator = random_generator
         self.placements: list = placements
         self.locations: list = [] if locations is None else locations

safety_gymnasium/bases/underlying.py

@@ -46,7 +46,7 @@ class RenderConf:
     r"""Render options.
 
     Attributes:
-        libels (bool): Whether to render labels.
+        labels (bool): Whether to render labels.
         lidar_markers (bool): Whether to render lidar markers.
         lidar_radius (float): Radius of the lidar markers.
         lidar_size (float): Size of the lidar markers.
@@ -482,7 +482,7 @@ def _render_sphere(
     def render(
         self,
         width: int,
-        height: int,
+            height: int,
         mode: str,
         camera_id: int | None = None,
         camera_name: str | None = None,
@@ -577,7 +577,9 @@ def _get_viewer(
                     self.agent.keyboard_control_callback,
                 )
             elif mode in {'rgb_array', 'depth_array'}:
-                self.viewer = OffScreenViewer(self.model, self.data)
+                self.viewer = OffScreenViewer(self.model, self.data,
+                                              self.model.vis.global_.offwidth,
+                                              self.model.vis.global_.offheight)
             else:
                 raise AttributeError(f'Unexpected mode: {mode}')
 

safety_gymnasium/tasks/safe_multi_agent/assets/free_geoms/push_box.py

@@ -38,7 +38,7 @@ class PushBox(FreeGeom):  # pylint: disable=too-many-instance-attributes
     reward_box_dist: float = 1.0  # Dense reward for moving the agent towards the box
     reward_box_goal: float = 1.0  # Reward for moving the box towards the goal
 
-    color: np.array = COLOR['push_box']
+    color: np.array = field(default_factory=lambda: COLOR['push_box'])
     group: np.array = GROUP['push_box']
     is_lidar_observed: bool = True
     is_comp_observed: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/free_geoms/vases.py

@@ -48,7 +48,7 @@ class Vases(FreeGeom):  # pylint: disable=too-many-instance-attributes
     velocity_threshold: float = 1e-4  # Ignore very small velocities
     last_contact: dict = field(default_factory=lambda: {f'vase{i}': 0 for i in range(20)})
 
-    color: np.array = COLOR['vase']
+    color: np.array = field(default_factory=lambda: COLOR['vase'])
     group: np.array = GROUP['vase']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/apples.py

@@ -18,6 +18,7 @@
 
 import numpy as np
 
+
 from safety_gymnasium.tasks.safe_multi_agent.assets.color import COLOR
 from safety_gymnasium.tasks.safe_multi_agent.assets.group import GROUP
 from safety_gymnasium.tasks.safe_multi_agent.bases.base_object import Geom
@@ -44,7 +45,7 @@ class Apples(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['apple']
+    color: np.array = field(default_factory=lambda: COLOR['apple'])
     group: np.array = GROUP['apple']
     is_lidar_observed: bool = True
     is_constrained: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/buttons.py

@@ -45,7 +45,7 @@ class Buttons(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['button']
+    color: np.array = field(default_factory=lambda: COLOR['button'])
     group: np.array = GROUP['button']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/circle.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 """Circle."""
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 import numpy as np
 
@@ -33,7 +33,7 @@ class Circle(Geom):  # pylint: disable=too-many-instance-attributes
     locations: tuple = ((0, 0),)
     keepout: float = 0.0
 
-    color: np.array = COLOR['circle']
+    color: np.array = field(default_factory=lambda: COLOR['circle'])
     group: np.array = GROUP['circle']
     is_lidar_observed: bool = True
     is_constrained: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/goal.py

@@ -40,7 +40,7 @@ class Goal(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['goal']
+    color: np.array = field(default_factory=lambda: COLOR['goal'])
     group: np.array = GROUP['goal']
     is_lidar_observed: bool = True
     is_comp_observed: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/hazards.py

@@ -36,7 +36,7 @@ class Hazards(Geom):  # pylint: disable=too-many-instance-attributes
     alpha: float = COLOR['hazard'][-1]
     cost: float = 1.0  # Cost (per step) for violating the constraint
 
-    color: np.array = COLOR['hazard']
+    color: np.array = field(default_factory=lambda: COLOR['hazard'])
     group: np.array = GROUP['hazard']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/oranges.py

@@ -44,7 +44,7 @@ class Oranges(Geom):  # pylint: disable=too-many-instance-attributes
     # if reward_distance is 0, then the reward function is sparse
     reward_distance: float = 1.0  # Dense reward multiplied by the distance moved to the goal
 
-    color: np.array = COLOR['orange']
+    color: np.array = field(default_factory=lambda: COLOR['orange'])
     group: np.array = GROUP['orange']
     is_lidar_observed: bool = True
     is_constrained: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/pillars.py

@@ -36,7 +36,7 @@ class Pillars(Geom):  # pylint: disable=too-many-instance-attributes
     keepout: float = 0.3  # Radius for placement of pillars
     cost: float = 1.0  # Cost (per step) for being in contact with a pillar
 
-    color: np.array = COLOR['pillar']
+    color: np.array = field(default_factory=lambda: COLOR['pillar'])
     group: np.array = GROUP['pillar']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/sigwalls.py

@@ -14,7 +14,7 @@
 # ==============================================================================
 """Hazard."""
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 
 import numpy as np
 
@@ -34,7 +34,7 @@ class Sigwalls(Geom):  # pylint: disable=too-many-instance-attributes
     placements: list = None
     keepout: float = 0.0
 
-    color: np.array = COLOR['sigwall']
+    color: np.array = field(default_factory=lambda: COLOR['sigwall'])
     group: np.array = GROUP['sigwall']
     is_lidar_observed: bool = False
     is_constrained: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/geoms/walls.py

@@ -36,7 +36,7 @@ class Walls(Geom):  # pylint: disable=too-many-instance-attributes
     locations: list = field(default_factory=list)  # This should be used and length == walls_num
     keepout: float = 0.0  # This should not be used
 
-    color: np.array = COLOR['wall']
+    color: np.array = field(default_factory=lambda: COLOR['wall'])
     group: np.array = GROUP['wall']
     is_lidar_observed: bool = True
     is_constrained: bool = False

safety_gymnasium/tasks/safe_multi_agent/assets/mocaps/gremlins.py

@@ -39,7 +39,7 @@ class Gremlins(Mocap):  # pylint: disable=too-many-instance-attributes
     dist_cost: float = 1.0  # Cost for being within distance threshold
     density: float = 0.001
 
-    color: np.array = COLOR['gremlin']
+    color: np.array = field(default_factory=lambda: COLOR['gremlin'])
     group: np.array = GROUP['gremlin']
     is_lidar_observed: bool = True
     is_constrained: bool = True

safety_gymnasium/tasks/safe_multi_agent/assets/robot.py

@@ -34,7 +34,7 @@ class Robot:  # pylint: disable=too-many-instance-attributes
     placements: list = None  # Robot placements list (defaults to full extents)
     locations: list = field(default_factory=list)  # Explicitly place robot XY coordinate
     keepout: float = 0.4  # Needs to be set to match the robot XML used
-    base: str = 'assets/xmls/car.xml'  # Which robot XML to use as the base
+    base: str = os.path.join('assets', 'xmls', 'car.xml')  # Which robot XML to use as the base
     rot: float = None  # Override robot starting angle
 
     def __post_init__(self, path) -> None:

safety_gymnasium/tasks/safe_multi_agent/bases/base_agent.py

@@ -182,7 +182,7 @@ def __init__(  # pylint: disable=too-many-arguments
             keepout (float): Needs to be set to match the agent XML used.
             rot (float): Override agent starting angle.
         """
-        self.base: str = f'assets/xmls/multi_{name.lower()}.xml'
+        self.base: str = os.path.join('assets', 'xmls', f'multi_{name.lower()}.xml')
         self.random_generator: RandomGenerator = random_generator
         self.placements: list = placements
         self.locations: list = [] if locations is None else locations

safety_gymnasium/tasks/safe_multi_agent/tasks/velocity/safe_mujoco_multi.py

@@ -73,7 +73,7 @@ def __init__(  # pylint: disable=too-many-arguments
             render_mode,
             **kwargs,
         )
-        self.env.single_agent_env.model.light(0).castshadow = False
+        self.env.single_agent_env.unwrapped.model.light(0).castshadow = False
 
     def __getattr__(self, name: str) -> Any:
         """Returns an attribute with ``name``, unless ``name`` starts with an underscore."""
@@ -97,7 +97,7 @@ def step(self, action):
         for agents in self.env.possible_agents:
             costs[agents] = cost_n
 
-        viewer = self.env.single_agent_env.mujoco_renderer.viewer
+        viewer = self.env.single_agent_env.unwrapped.mujoco_renderer.viewer
         if viewer:
             clear_viewer(viewer)
             add_velocity_marker(