diff --git a/src/lerobot/policies/groot/processor_groot.py b/src/lerobot/policies/groot/processor_groot.py
index 62ad34094..a695c7d8e 100644
--- a/src/lerobot/policies/groot/processor_groot.py
+++ b/src/lerobot/policies/groot/processor_groot.py
@@ -469,6 +469,7 @@ def make_groot_pre_post_processors_from_pretrained(
         # lerobot-eval or the policy server) to the freshly built steps.
         _apply_groot_step_overrides(preprocessor, preprocessor_overrides)
         _apply_groot_step_overrides(postprocessor, postprocessor_overrides)
+        _apply_groot_action_decode_transform(postprocessor, config.action_decode_transform)
         return preprocessor, postprocessor
 
     preprocessor, postprocessor = _load_groot_processor_pipelines(
@@ -480,6 +481,7 @@ def make_groot_pre_post_processors_from_pretrained(
     )
     _reconnect_groot_relative_absolute_steps(preprocessor, postprocessor)
     _reconnect_groot_n1_7_pack_decode_steps(preprocessor, postprocessor)
+    _apply_groot_action_decode_transform(postprocessor, config.action_decode_transform)
     return preprocessor, postprocessor
 
 
@@ -552,6 +554,20 @@ def _reconnect_groot_n1_7_pack_decode_steps(
             step.pack_step = pack_step
 
 
+def _apply_groot_action_decode_transform(
+    postprocessor: PolicyProcessorPipeline,
+    action_decode_transform: str | None,
+) -> None:
+    use_libero_transform = action_decode_transform == GROOT_ACTION_DECODE_TRANSFORM_LIBERO
+
+    for step in postprocessor.steps:
+        if isinstance(step, GrootN17ActionDecodeStep):
+            step.action_decode_transform = action_decode_transform
+        elif isinstance(step, GrootActionUnpackUnnormalizeStep):
+            step.libero_gripper_action = use_libero_transform
+            if use_libero_transform:
+                step.libero_gripper_binarize = True
+
 
 def _resolve_feature_names_from_dataset_meta(dataset_meta: Any | None, feature_key: str) -> list[str] | None:
     features = getattr(dataset_meta, "features", {}) or {}
@@ -1198,6 +1214,7 @@ def make_groot_pre_post_processors(
             stats=padded_stats,
             normalize_min_max=True,
             clip_normalized_action=True,
+            libero_gripper_action=config.action_decode_transform == GROOT_ACTION_DECODE_TRANSFORM_LIBERO,
         )
     else:
         action_decode_step = GrootN17ActionDecodeStep(
diff --git a/tests/policies/groot/test_groot_n1_7.py b/tests/policies/groot/test_groot_n1_7.py
index 6f31b5695..f915f985c 100644
--- a/tests/policies/groot/test_groot_n1_7.py
+++ b/tests/policies/groot/test_groot_n1_7.py
@@ -1391,6 +1391,80 @@ def test_groot_n1_7_action_decode_requires_gripper_key_for_libero_transform():
         step({TransitionKey.ACTION: torch.zeros(1, 1, 1)})
 
 
+def test_groot_n1_7_fallback_processors_wire_libero_transform_to_postprocessor():
+    config = _groot_config()
+    dataset_stats = {
+        OBS_STATE: {
+            "min": torch.zeros(8),
+            "max": torch.ones(8),
+        },
+        ACTION: {
+            "min": torch.zeros(7),
+            "max": torch.ones(7),
+        },
+    }
+
+    _, postprocessor = make_groot_pre_post_processors(config, dataset_stats=dataset_stats)
+
+    action_decode_step = next(
+        step for step in postprocessor.steps if isinstance(step, GrootActionUnpackUnnormalizeStep)
+    )
+    assert action_decode_step.libero_gripper_action is True
+
+
+def test_groot_n1_7_loaded_fallback_postprocessor_honors_config_action_decode_transform(tmp_path):
+    input_features, output_features = _groot_features(state_dim=8, action_dim=7)
+    dataset_stats = {
+        OBS_STATE: {
+            "min": torch.zeros(8),
+            "max": torch.ones(8),
+        },
+        ACTION: {
+            "min": torch.zeros(7),
+            "max": torch.ones(7),
+        },
+    }
+    disabled_config = GrootConfig(
+        input_features=input_features,
+        output_features=output_features,
+        device="cpu",
+        use_bf16=False,
+        action_decode_transform=None,
+    )
+    preprocessor, postprocessor = make_groot_pre_post_processors(
+        disabled_config,
+        dataset_stats=dataset_stats,
+    )
+    save_dir = tmp_path / "saved_fallback_processors"
+    disabled_config.save_pretrained(save_dir)
+    preprocessor.save_pretrained(save_dir)
+    postprocessor.save_pretrained(save_dir)
+
+    saved_postprocessor = json.loads((save_dir / "policy_postprocessor.json").read_text())
+    saved_decode_config = next(
+        step["config"]
+        for step in saved_postprocessor["steps"]
+        if step["registry_name"] == "groot_action_unpack_unnormalize_v2"
+    )
+    assert saved_decode_config["libero_gripper_action"] is False
+
+    enabled_config = GrootConfig(
+        input_features=input_features,
+        output_features=output_features,
+        device="cpu",
+        use_bf16=False,
+        action_decode_transform=GROOT_ACTION_DECODE_TRANSFORM_LIBERO,
+    )
+    _, loaded_postprocessor = make_pre_post_processors(enabled_config, pretrained_path=str(save_dir))
+    action_decode_step = next(
+        step for step in loaded_postprocessor.steps if isinstance(step, GrootActionUnpackUnnormalizeStep)
+    )
+
+    assert action_decode_step.libero_gripper_action is True
+    output = action_decode_step({TransitionKey.ACTION: torch.tensor([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0]])})
+    torch.testing.assert_close(output[TransitionKey.ACTION][0, -1], torch.tensor(1.0))
+
+
 def test_groot_n1_7_postprocessor_converts_libero_gripper_convention():
     step = GrootActionUnpackUnnormalizeStep(
         env_action_dim=7,