[gym_jiminy/common] Fix stack length computation from horizon.

python/gym_jiminy/common/gym_jiminy/common/bases/compositions.py

@@ -59,10 +59,19 @@ def __new__(cls,
         self = super(partial_hashable, cls).__new__(cls, func, *args, **kwargs)
 
         # Pre-compute normalized arguments once and for all
-        sig = inspect.signature(self.func)
-        bound = sig.bind_partial(*self.args, **(self.keywords or {}))
-        bound.apply_defaults()
-        self._normalized_args = tuple(bound.arguments.values())
+        try:
+            sig = inspect.signature(self.func)
+            bound = sig.bind_partial(*self.args, **(self.keywords or {}))
+            bound.apply_defaults()
+            self._normalized_args = tuple(bound.arguments.values())
+        except ValueError as e:
+            # Impossible to get signature from Python bindings. Keyword-only is
+            # enforced to ensure that equality check can be implemented.
+            if self.args:
+                raise ValueError(
+                    "Specifying position arguments is not supported for "
+                    "methods whose signature cannot be inspected.") from e
+            _, self._normalized_args = zip(*sorted(self.keywords.items()))
 
         return self
 

python/gym_jiminy/common/gym_jiminy/common/compositions/generic.py

@@ -430,7 +430,8 @@ def __init__(self,
         self.op = op
 
         # Convert horizon in stack length, assuming constant env timestep
-        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1)
+        assert horizon >= env.step_dt
+        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1) + 1
 
         # Define drift of quantity
         stack_creator = lambda mode: (StackedQuantity, dict(  # noqa: E731

python/gym_jiminy/common/gym_jiminy/common/quantities/generic.py

@@ -1862,7 +1862,7 @@ def __init__(
         :param mode: Desired mode of evaluation for this quantity.
         """
         # Convert horizon in stack length, assuming constant env timestep
-        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1)
+        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1) + 1
 
         # Backup some of the user-arguments
         self.max_stack = max_stack

python/gym_jiminy/common/gym_jiminy/common/quantities/locomotion.py

@@ -1569,7 +1569,8 @@ def __init__(self,
                      Optional: 'QuantityEvalMode.TRUE' by default.
         """
         # Convert horizon in stack length, assuming constant env timestep
-        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1)
+        assert horizon >= env.step_dt
+        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1) + 1
 
         # Backup some of the user-arguments
         self.max_stack = max_stack
@@ -1675,7 +1676,8 @@ def __init__(self,
                      Optional: 'QuantityEvalMode.TRUE' by default.
         """
         # Convert horizon in stack length, assuming constant env timestep
-        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1)
+        assert horizon >= env.step_dt
+        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1) + 1
 
         # Backup some of the user-arguments
         self.max_stack = max_stack

python/gym_jiminy/common/gym_jiminy/common/quantities/transform.py

@@ -219,6 +219,8 @@ def refresh(self) -> OtherValueT:
                 raise RuntimeError(
                     "Previous step missing in the stack. Please reset the "
                     "environment after adding this quantity.")
+        else:
+            must_refresh = False
 
         # Extract contiguous slice of (future) available data if necessary
         if self.as_array:
@@ -227,18 +229,18 @@ def refresh(self) -> OtherValueT:
             if num_stack < self.max_stack:
                 data = self._data[..., :num_stack]
 
-        # Get current index if wrapping around
-        if self.is_wrapping:
-            index = num_steps % self.max_stack
-
         # Append current value of the quantity to the history buffer or update
         # aggregated continuous array directly if necessary.
-        is_stack_full = num_steps >= self.max_stack
         if must_refresh:
             # Get the current value of the quantity
             value = self.quantity.get()
 
+            # Get current index if wrapping around
+            if self.is_wrapping:
+                index = num_steps % self.max_stack
+
             # Append value to the history or aggregate data directly
+            is_stack_full = num_steps >= self.max_stack
             if self.as_array:
                 if self.is_wrapping:
                     array_copyto(data[..., index], value)
@@ -798,7 +800,7 @@ def __init__(self,
                             Optional: True by default.
         """
         # Convert horizon in stack length, assuming constant env timestep
-        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1)
+        max_stack = max(int(np.ceil(horizon / env.step_dt)), 1) + 1
 
         # Backup some of the user-arguments
         self.op = op