From e71d873ff2d0b777ec1eedde23d864220d9cd187 Mon Sep 17 00:00:00 2001
From: Bartok9 <danielrpike9@gmail.com>
Date: Sat, 25 Jul 2026 01:53:13 -0400
Subject: [PATCH] fix(hri): pass base64_audios for AI multimodal to_langchain

AI branch incorrectly re-used base64_images for the audios kwargs, so
image-only AI HRI messages raised MultimodalMessage's audio-not-supported
error. Mirror the human branch and cover images-only vs image+audio.

Signed-off-by: Bartok9 <danielrpike9@gmail.com>
---
 .../rai/communication/hri_connector.py        |  2 +-
 tests/communication/test_hri_message.py       | 32 +++++++++++++------
 2 files changed, 24 insertions(+), 10 deletions(-)

diff --git a/src/rai_core/rai/communication/hri_connector.py b/src/rai_core/rai/communication/hri_connector.py
index 28763766c..f6e352465 100644
--- a/src/rai_core/rai/communication/hri_connector.py
+++ b/src/rai_core/rai/communication/hri_connector.py
@@ -92,7 +92,7 @@ def to_langchain(self) -> LangchainBaseMessage:
                 if self.images == [] and self.audios == []:
                     return AIMessage(content=self.text)
                 return AIMultimodalMessage(
-                    content=self.text, images=base64_images, audios=base64_images
+                    content=self.text, images=base64_images, audios=base64_audios
                 )
             case _:
                 raise ValueError(
diff --git a/tests/communication/test_hri_message.py b/tests/communication/test_hri_message.py
index 8ec939dc1..52abb4482 100644
--- a/tests/communication/test_hri_message.py
+++ b/tests/communication/test_hri_message.py
@@ -82,26 +82,40 @@ def test_to_langchain_human():
     assert langchain_message.content == "Hi there"
 
 
-def test_to_langchain_ai_multimodal(image, audio):
+def test_to_langchain_ai_images_only(image):
+    """AI messages with images (no audio) must map audios correctly and succeed."""
+    from rai.messages.multimodal import AIMultimodalMessage
+
     message = HRIMessage(
         text="Response",
         images=[image],
-        audios=[audio],
+        audios=[],
         message_author="ai",
         communication_id=HRIMessage.generate_communication_id(),
         seq_no=0,
         seq_end=True,
     )
+    langchain_message = message.to_langchain()
+    assert isinstance(langchain_message, AIMultimodalMessage)
+    assert langchain_message.images is not None
+    assert len(langchain_message.images) == 1
+
 
-    with pytest.raises(
-        ValueError
-    ):  # NOTE: update when https://github.com/RobotecAI/rai/issues/370 is resolved
+def test_to_langchain_ai_with_audio_still_unsupported(image, audio):
+    """Audio on MultimodalMessage remains unsupported until issue #370."""
+    message = HRIMessage(
+        text="Response",
+        images=[image],
+        audios=[audio],
+        message_author="ai",
+        communication_id=HRIMessage.generate_communication_id(),
+        seq_no=0,
+        seq_end=True,
+    )
+    with pytest.raises(ValueError, match="Audio is not yet supported"):
         _ = message.to_langchain()
 
-    # assert isinstance(langchain_message, AIMultimodalMessage)
-    # assert langchain_message.content == "Response"
-    # assert langchain_message.images == ["img"]
-    # assert langchain_message.audios == ["audio"]
+
 
 
 def test_from_langchain_human():