Merge pull request #9 from Co-Evolve/develop

v1.0.3

moojoco/environment/base.py

@@ -12,8 +12,8 @@
 from mujoco import mjx
 
 import moojoco.environment.mjx_spaces as mjx_spaces
-from moojoco.mjcf.arena import MJCFArena
 from moojoco.environment.renderer import MujocoRenderer
+from moojoco.mjcf.arena import MJCFArena
 from moojoco.mjcf.morphology import MJCFMorphology
 
 
@@ -123,11 +123,15 @@ def observation_space(self) -> SpaceType:
         raise NotImplementedError
 
     @abc.abstractmethod
-    def step(self, state: BaseEnvState, action: chex.Array) -> BaseEnvState:
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
         raise NotImplementedError
 
     @abc.abstractmethod
-    def reset(self, rng: np.random.RandomState | chex.PRNGKey) -> BaseEnvState:
+    def reset(
+        self, rng: np.random.RandomState | chex.PRNGKey, *args, **kwargs
+    ) -> BaseEnvState:
         raise NotImplementedError
 
     @abc.abstractmethod
@@ -225,7 +229,9 @@ def _initialize_mj_model_and_data(self) -> Tuple[mujoco.MjModel, mujoco.MjData]:
         mj_data = mujoco.MjData(mj_model)
         return mj_model, mj_data
 
-    def step(self, state: BaseEnvState, action: chex.Array) -> BaseEnvState:
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
         previous_state = state
         state = self._update_simulation(state=state, ctrl=action)
 
@@ -345,7 +351,9 @@ def _finish_reset(
         raise NotImplementedError
 
     @abc.abstractmethod
-    def reset(self, rng: np.random.RandomState | chex.PRNGKey) -> BaseEnvState:
+    def reset(
+        self, rng: np.random.RandomState | chex.PRNGKey, *args, **kwargs
+    ) -> BaseEnvState:
         raise NotImplementedError
 
     @abc.abstractmethod

moojoco/environment/dual.py

@@ -60,10 +60,14 @@ def actuators(self) -> List[str]:
     def observation_space(self) -> SpaceType:
         return self._env.observation_space
 
-    def step(self, state: BaseEnvState, action: chex.Array) -> BaseEnvState:
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
         return self._env.step(state=state, action=action)
 
-    def reset(self, rng: np.random.RandomState | chex.PRNGKey) -> BaseEnvState:
+    def reset(
+        self, rng: np.random.RandomState | chex.PRNGKey, *args, **kwargs
+    ) -> BaseEnvState:
         return self._env.reset(rng=rng)
 
     def render(self, state: BaseEnvState) -> List[RenderFrame] | None:

moojoco/environment/mjc_env.py

@@ -144,7 +144,7 @@ def _create_observables(self) -> List[MJCObservable]:
         raise NotImplementedError
 
     @abc.abstractmethod
-    def reset(self, rng: np.random.RandomState) -> MJCEnvState:
+    def reset(self, rng: np.random.RandomState, *args, **kwargs) -> MJCEnvState:
         raise NotImplementedError
 
     @abc.abstractmethod
@@ -260,7 +260,9 @@ def actuators(self) -> List[str]:
     def observation_space(self) -> gymnasium.spaces.Space:
         return self._observation_space
 
-    def step(self, state: VectorMJCEnvState, action: np.ndarray) -> VectorMJCEnvState:
+    def step(
+        self, state: VectorMJCEnvState, action: np.ndarray, *args, **kwargs
+    ) -> VectorMJCEnvState:
         self._states = list(
             self._pool.map(
                 lambda env, ste, act: env.step(state=ste, action=act),
@@ -271,7 +273,9 @@ def step(self, state: VectorMJCEnvState, action: np.ndarray) -> VectorMJCEnvStat
         )
         return self._merged_states
 
-    def reset(self, rng: List[np.random.RandomState]) -> VectorMJCEnvState:
+    def reset(
+        self, rng: List[np.random.RandomState], *args, **kwargs
+    ) -> VectorMJCEnvState:
         self._states = list(
             self._pool.map(lambda env, sub_rng: env.reset(sub_rng), self._envs, rng)
         )

moojoco/environment/mjx_env.py

@@ -228,7 +228,7 @@ def _create_observables(self) -> List[MJXObservable]:
         raise NotImplementedError
 
     @abc.abstractmethod
-    def reset(self, rng: jnp.ndarray) -> MJXEnvState:
+    def reset(self, rng: jnp.ndarray, *args, **kwargs) -> MJXEnvState:
         raise NotImplementedError
 
     @abc.abstractmethod

moojoco/environment/wrapper.py

@@ -0,0 +1,118 @@
+import abc
+from typing import List, Tuple
+
+import chex
+import numpy as np
+from gymnasium.core import RenderFrame
+
+from moojoco.environment.base import BaseEnvState, BaseEnvironment, SpaceType
+
+
+class PostInitCaller(type):
+    def __call__(cls, *args, **kwargs):
+        obj = type.__call__(cls, *args, **kwargs)
+        obj.__post__init__()
+        return obj
+
+
+class CombinedMeta(abc.ABCMeta, PostInitCaller):
+    pass
+
+
+class EnvironmentWrapper(BaseEnvironment, metaclass=CombinedMeta):
+    def __init__(self, env: BaseEnvironment) -> None:
+        super().__init__(configuration=env.environment_configuration)
+        self._env = env
+
+        self._observation_space: SpaceType | None = None
+        self._action_space: SpaceType | None = None
+
+    def __post__init__(self) -> None:
+        # Make sure observation space and action space are initialised after creation
+        # noinspection PyStatementEffect
+        self.observation_space
+        # noinspection PyStatementEffect
+        self.action_space
+
+    @property
+    def action_space(self) -> SpaceType:
+        if self._action_space is not None:
+            return self._action_space
+        return self._env.action_space
+
+    @property
+    def actuators(self) -> List[str]:
+        return self._env.actuators
+
+    @property
+    def observation_space(self) -> SpaceType:
+        if self._observation_space is not None:
+            return self._observation_space
+        return self._env.observation_space
+
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
+        return self._env.step(state=state, action=action)
+
+    def reset(
+        self, rng: np.random.RandomState | chex.PRNGKey, *args, **kwargs
+    ) -> BaseEnvState:
+        return self._env.reset(rng=rng)
+
+    def render(self, state: BaseEnvState) -> List[RenderFrame] | None:
+        return self._env.render(state=state)
+
+    def close(self) -> None:
+        return self._env.close()
+
+
+class TransformObservationEnvWrapper(EnvironmentWrapper, abc.ABC):
+    def __init__(self, env: BaseEnvironment) -> None:
+        super().__init__(env=env)
+
+    @property
+    @abc.abstractmethod
+    def observation_space(self) -> SpaceType:
+        raise NotImplementedError
+
+    @abc.abstractmethod
+    def _transform_observations(self, state: BaseEnvState) -> BaseEnvState:
+        raise NotImplementedError
+
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
+        state = self._env.step(state=state, action=action)
+        state = self._transform_observations(state=state)
+        return state
+
+    def reset(
+        self, rng: np.random.RandomState | chex.PRNGKey, *args, **kwargs
+    ) -> BaseEnvState:
+        state = self._env.reset(rng=rng)
+        state = self._transform_observations(state=state)
+        return state
+
+
+class TransformActionEnvWrapper(EnvironmentWrapper, abc.ABC):
+    def __init__(self, env: BaseEnvironment) -> None:
+        super().__init__(env=env)
+
+    @property
+    @abc.abstractmethod
+    def action_space(self) -> SpaceType:
+        raise NotImplementedError
+
+    @abc.abstractmethod
+    def _transform_action(
+        self, action: chex.Array, state: BaseEnvState
+    ) -> Tuple[chex.Array, BaseEnvState]:
+        raise NotImplementedError
+
+    def step(
+        self, state: BaseEnvState, action: chex.Array, *args, **kwargs
+    ) -> BaseEnvState:
+        action, state = self._transform_action(action=action, state=state)
+        state = self._env.step(state=state, action=action)
+        return state

moojoco/mjcf/component.py

@@ -7,6 +7,7 @@
 import numpy as np
 from dm_control import mjcf
 from dm_control.mjcf import export_with_assets
+from dm_control.mjcf.element import _AttachmentFrame
 from scipy.spatial.transform import Rotation
 
 
@@ -35,7 +36,7 @@ def attach(
         position: Optional[np.ndarray] = None,
         euler: Optional[np.ndarray] = None,
         free_joint: bool = False,
-    ) -> None:
+    ) -> _AttachmentFrame:
         attachment_site = self.mjcf_body.add(
             "site",
             name=f"{self.base_name}_attachment_{other.base_name}",

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "moojoco"
-version = "1.0.2"
+version = "1.0.3"
 authors = [
     { name = "Dries Marzougui", email = "[email protected]" },
 ]

setup.py

@@ -11,7 +11,7 @@
 
 setup(
         name='moojoco',
-        version='1.0.2',
+        version='1.0.3',
         description='A unified framework for implementing and interfacing with MuJoCo and MuJoCo-XLA simulation '
                     'environments.',
         long_description=readme,