renaming to return_intermediate_predictions

src/lerobot/policies/pretrained.py

@@ -94,7 +94,7 @@ def _build_card_context(
 
 class ActionSelectKwargs(TypedDict, total=False):
     noise: Tensor | None
-    return_extra: bool
+    return_intermediate_predictions: bool
 
 
 class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
@@ -281,9 +281,10 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
         Child classes using action chunking should use this method within `select_action` to form the action chunk
         cached for selection.
 
-        By default returns just the action `Tensor`. If `return_extra=True`, returns `(action, extra)`
+        By default returns just the action `Tensor`. If `return_intermediate_predictions=True`,
-        where `extra` is a (possibly empty) `dict[str, Tensor]` of auxiliary outputs a policy may
+        returns `(action, predictions)` where `predictions` is a (possibly empty) `dict[str, Tensor]`
-        expose (e.g. world-model predictions). Policies that produce nothing extra may ignore the kwarg.
+        of additional model predictions a policy may expose (e.g. world-model predicted frames).
+        Policies that produce nothing extra may ignore the kwarg.
         """
         raise NotImplementedError
 
@@ -296,9 +297,10 @@ class PreTrainedPolicy(nn.Module, HubMixin, abc.ABC):
         When the model uses a history of observations, or outputs a sequence of actions, this method deals
         with caching.
 
-        By default returns just the action `Tensor`. If `return_extra=True`, returns `(action, extra)`
+        By default returns just the action `Tensor`. If `return_intermediate_predictions=True`,
-        where `extra` is a (possibly empty) `dict[str, Tensor]` of auxiliary outputs a policy may
+        returns `(action, predictions)` where `predictions` is a (possibly empty) `dict[str, Tensor]`
-        expose (e.g. world-model predictions). Policies that produce nothing extra may ignore the kwarg.
+        of additional model predictions a policy may expose (e.g. world-model predicted frames).
+        Policies that produce nothing extra may ignore the kwarg.
         """
         raise NotImplementedError