diff --git a/image_0.png b/image_0.png
index 7fd439e1d..874f8f32f 100644
Binary files a/image_0.png and b/image_0.png differ
diff --git a/image_1.png b/image_1.png
index bdabc7479..303729689 100644
Binary files a/image_1.png and b/image_1.png differ
diff --git a/image_2.png b/image_2.png
index ad118a751..7d386f85f 100644
Binary files a/image_2.png and b/image_2.png differ
diff --git a/image_3.png b/image_3.png
index 2d6dd6a89..07385ca6d 100644
Binary files a/image_3.png and b/image_3.png differ
diff --git a/image_4.png b/image_4.png
index b3e3b8b76..40b36bfea 100644
Binary files a/image_4.png and b/image_4.png differ
diff --git a/image_5.png b/image_5.png
index 9b227df31..de471a98a 100644
Binary files a/image_5.png and b/image_5.png differ
diff --git a/predicators/perception/spot_perceiver.py b/predicators/perception/spot_perceiver.py
index db98b07ac..1b498e4ab 100644
--- a/predicators/perception/spot_perceiver.py
+++ b/predicators/perception/spot_perceiver.py
@@ -4,11 +4,13 @@
 import time
 from pathlib import Path
 from typing import Dict, List, Optional, Set, Collection
+import json
 
 import imageio.v2 as iio
 import numpy as np
 from bosdyn.client import math_helpers
 from matplotlib import pyplot as plt
+import PIL.Image
 
 from predicators import utils
 from predicators.envs import BaseEnv, get_or_create_env
@@ -569,8 +571,7 @@ def _create_goal(self, state: State,
                     GroundAtom(Inside, [cup, plastic_bin]),
                 }
             object_name_to_object = {}
-            self._parse_vlm_goal_from_state(state, goal_description, object_name_to_object)
-            import ipdb; ipdb.set_trace()
+            return self._parse_vlm_goal_from_state(state, goal_description, object_name_to_object)
         if goal_description == "know container as empty":
             cup = Object("cup", _container_type)
             ContainingFoodKnown = pred_name_to_pred["ContainingFoodKnown"]
@@ -722,9 +723,8 @@ def _parse_vlm_goal_from_state(
         object_names = set(id_to_obj)
         prompt_prefix = self._get_language_goal_prompt_prefix(object_names)
         prompt = prompt_prefix + f"\n# {language_goal}"
-        import ipdb; ipdb.set_trace()
         image_list = [
-            PIL.Image.fromarray(v.rotated_rgb) for _, v in rgbds.items()
+            PIL.Image.fromarray(v.rotated_rgb) for _, v in state.camera_images.items()
         ]
         responses = vlm.sample_completions(
                 prompt=prompt,
@@ -734,10 +734,13 @@ def _parse_vlm_goal_from_state(
                 num_completions=1,
             )
         response = responses[0]
-        import ipdb; ipdb.set_trace()
         # Currently assumes that the LLM is perfect. In the future, will need
         # to handle various errors and perhaps query the LLM for multiple
         # responses until we find one that can be parsed.
-        goal_spec = json.loads(response)
+        try:
+            goal_spec = json.loads(response)
+        except json.JSONDecodeError as e:
+            goal_spec = json.loads(response.replace('`', '').replace('json', ''))
+        
         return self._curr_env._parse_goal_from_json(goal_spec, id_to_obj)
 
diff --git a/predicators/spot_utils/perception/object_perception.py b/predicators/spot_utils/perception/object_perception.py
index fc9f4e1fe..448aa03b5 100644
--- a/predicators/spot_utils/perception/object_perception.py
+++ b/predicators/spot_utils/perception/object_perception.py
@@ -216,7 +216,6 @@ def query_vlm(full_prompt, image_list):
         votes = [result[i] for result in results]
         final_results.append(votes.count(True) > votes.count(False))
 
-    import ipdb; ipdb.set_trace()
     return final_results