Add support for multi-dimensional action spaces to PPO variants

docs/misc/changelog.rst

@@ -3,7 +3,7 @@
 Changelog
 ==========
 
-Release 2.8.0a0 (WIP)
+Release 2.8.0a1 (WIP)
 --------------------------
 
 Breaking Changes:
@@ -19,7 +19,8 @@ New Features:
 
 Bug Fixes:
 ^^^^^^^^^^
-- Do not call ``forward()`` method directly in ``RecurrentPPO``
+- Fix RecurrentPPO and MaskablePPO forward and predict do not reshape action before clip it (@immortal-boy)
+- Do not call ``forward()`` method directly in ``RecurrentPPO`` (@immortal-boy)
 
 Deprecations:
 ^^^^^^^^^^^^^
@@ -657,3 +658,4 @@ Contributors:
 
 @ku2482 @guyk1971 @minhlong94 @ayeright @kronion @glmcdona @cyprienc @sgillen @Gregwar @rnederstigt @qgallouedec
 @mlodel @CppMaster @burakdmb @honglu2875 @ZikangXiong @AlexPasqua @jonasreiher @icheered @Armandpl @danielpalen @corentinlger
+@immortal-boy

sb3_contrib/common/maskable/distributions.py

@@ -138,19 +138,15 @@ def proba_distribution(
         return self
 
     def log_prob(self, actions: th.Tensor) -> th.Tensor:
-        assert self.distribution is not None, "Must set distribution parameters"
         return self.distribution.log_prob(actions)
 
     def entropy(self) -> th.Tensor:
-        assert self.distribution is not None, "Must set distribution parameters"
         return self.distribution.entropy()
 
     def sample(self) -> th.Tensor:
-        assert self.distribution is not None, "Must set distribution parameters"
         return self.distribution.sample()
 
     def mode(self) -> th.Tensor:
-        assert self.distribution is not None, "Must set distribution parameters"
         return th.argmax(self.distribution.probs, dim=1)
 
     def actions_from_params(self, action_logits: th.Tensor, deterministic: bool = False) -> th.Tensor:
@@ -164,7 +160,6 @@ def log_prob_from_params(self, action_logits: th.Tensor) -> tuple[th.Tensor, th.
         return actions, log_prob
 
     def apply_masking(self, masks: MaybeMasks) -> None:
-        assert self.distribution is not None, "Must set distribution parameters"
         self.distribution.apply_masking(masks)
 
 

sb3_contrib/common/maskable/policies.py

@@ -139,6 +139,7 @@ def forward(
             distribution.apply_masking(action_masks)
         actions = distribution.get_actions(deterministic=deterministic)
         log_prob = distribution.log_prob(actions)
+        actions = actions.reshape((-1, *self.action_space.shape))  # type: ignore[misc]
         return actions, values, log_prob
 
     def extract_features(  # type: ignore[override]
@@ -304,7 +305,7 @@ def predict(
         with th.no_grad():
             actions = self._predict(obs_tensor, deterministic=deterministic, action_masks=action_masks)
             # Convert to numpy
-            actions = actions.cpu().numpy()  # type: ignore[assignment]
+            actions = actions.cpu().numpy().reshape((-1, *self.action_space.shape))  # type: ignore[assignment, misc]
 
         if isinstance(self.action_space, spaces.Box):
             if self.squash_output:

sb3_contrib/common/recurrent/policies.py

@@ -253,6 +253,7 @@ def forward(
         distribution = self._get_action_dist_from_latent(latent_pi)
         actions = distribution.get_actions(deterministic=deterministic)
         log_prob = distribution.log_prob(actions)
+        actions = actions.reshape((-1, *self.action_space.shape))  # type: ignore[misc])
         return actions, values, log_prob, RNNStates(lstm_states_pi, lstm_states_vf)
 
     def get_distribution(
@@ -412,7 +413,7 @@ def predict(
             states = (states[0].cpu().numpy(), states[1].cpu().numpy())
 
         # Convert to numpy
-        actions = actions.cpu().numpy()
+        actions = actions.cpu().numpy().reshape((-1, *self.action_space.shape))  # type: ignore[assignment]
 
         if isinstance(self.action_space, spaces.Box):
             if self.squash_output:

sb3_contrib/version.txt

@@ -1 +1 @@
-2.8.0a0
+2.8.0a1

tests/test_distributions.py

@@ -82,19 +82,19 @@ def test_distribution_must_be_initialized(self):
 
         DIMS = 2
         dist = MaskableCategoricalDistribution(DIMS)
-        with pytest.raises(AssertionError):
+        with pytest.raises(AttributeError):
             dist.log_prob(th.randint(DIMS - 1, (1, 3)))
 
-        with pytest.raises(AssertionError):
+        with pytest.raises(AttributeError):
             dist.entropy()
 
-        with pytest.raises(AssertionError):
+        with pytest.raises(AttributeError):
             dist.sample()
 
-        with pytest.raises(AssertionError):
+        with pytest.raises(AttributeError):
             dist.mode()
 
-        with pytest.raises(AssertionError):
+        with pytest.raises(AttributeError):
             dist.apply_masking(None)
 
         # But now we can

tests/test_lstm.py

@@ -244,3 +244,22 @@ def make_env():
     # In CartPole-v1, a non-recurrent policy can easily get >= 450.
     # In CartPoleNoVelEnv, a non-recurrent policy doesn't get more than ~50.
     evaluate_policy(model, env, reward_threshold=450)
+
+
+class MultiDimensionalActionSpaceEnv(gym.Env):
+    def __init__(self):
+        self.observation_space = spaces.Box(low=-1, high=1, shape=(10,), dtype=np.float32)
+        self.action_space = spaces.Box(low=-1, high=1, shape=(2, 2), dtype=np.float32)
+
+    def reset(self, seed=None, options=None):
+        super().reset(seed=seed)
+        return self.observation_space.sample(), {}
+
+    def step(self, action):
+        return self.observation_space.sample(), 1, np.random.rand() > 0.8, False, {}
+
+
+def test_ppo_multi_dimensional_action_space():
+    env = MultiDimensionalActionSpaceEnv()
+    model = RecurrentPPO("MlpLstmPolicy", env, n_steps=64, n_epochs=2).learn(64)
+    evaluate_policy(model, model.get_env())