diff --git a/robowaiter/behavior_tree/dataset/__init__.py b/robowaiter/behavior_tree/dataset/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/robowaiter/behavior_tree/dataset/goal_states_generation.py b/robowaiter/behavior_tree/dataset/goal_states_generation.py
new file mode 100644
index 0000000..550d86f
--- /dev/null
+++ b/robowaiter/behavior_tree/dataset/goal_states_generation.py
@@ -0,0 +1,66 @@
+# the empty string '' represents robot holds nothing
+Object = ['Coffee', 'Water', 'Dessert', 'Softdrink', 'BottledDrink', 'Yogurt', 'ADMilk', 'MilkDrink', 'Milk',
+          'VacuumCup', '']
+
+Place = ['Bar', 'WaterTable', 'CoffeeTable', 'Bar2', 'Table1', 'Table2', 'Table3']
+
+Entity = ['Robot', 'Customer']
+
+Operable = ['AC', 'ACTemperature', 'HallLight', 'TubeLight', 'Curtain']
+
+import random
+
+
+def single_predict_generation(oplist_1, oplist_2, predict_pattern) -> str:
+    index_1 = random.randint(0, len(oplist_1) - 1)
+    if oplist_2:
+        index_2 = random.randint(0, len(oplist_2) - 1)
+
+    match predict_pattern:
+        case 'at':
+            return f'At({oplist_1[index_1]}, {oplist_2[index_2]})'
+        case 'is':
+            return f'Is({oplist_1[index_1]}, {oplist_2[index_2]})'
+        case 'hold':
+            return f'Holding({oplist_1[index_1]})'
+        case 'on':
+            return f'On({oplist_1[index_1]}, {oplist_2[index_2]})'
+        case _:
+            raise RuntimeError('Incorrect predict pattern!')
+
+
+def generate_goal_states(vln_num: int, vlm_num: int, opentask_num: int):
+    # res stores lists of sets, while each state represent in set.
+    res = []
+
+    # goal states for VLN
+    for i in range(vln_num):
+        res.append({single_predict_generation(['Robot'], Place, 'at')})
+
+    # goal states for VLM
+    for i in range(int(vlm_num)):
+        for j in range(int(vlm_num)):
+            res.append(
+                {
+                    single_predict_generation(['Robot'], Place, 'at'),
+                    single_predict_generation(Operable, ['0', '1'], 'is')
+                }
+            )
+
+    # goal states for Open-task-1
+    for i in range(int(opentask_num)):
+        for j in range(int(opentask_num)):
+            res.append(
+                {
+                    single_predict_generation(['Robot'], Place, 'at'),
+                    single_predict_generation(Object, Place, 'on')
+                }
+            )
+
+    # print(res)
+    # print(len(res))
+
+    return res
+
+
+generate_goal_states(30, 6, 6)