diff --git a/selfdrive/modeld/modeld.py b/selfdrive/modeld/modeld.py
index 006eeef6f5ed56..db43f6ac91264d 100755
--- a/selfdrive/modeld/modeld.py
+++ b/selfdrive/modeld/modeld.py
@@ -43,11 +43,11 @@
 LAT_SMOOTH_SECONDS = 0.1
 LONG_SMOOTH_SECONDS = 0.3
 MIN_LAT_CONTROL_SPEED = 0.3
-
+RECOVERY_POWER = 1.0 # The higher this number the more aggressively the model will recover to lanecenter, too high and it will ping-pong
 
 def get_action_from_model(model_output: dict[str, np.ndarray], prev_action: log.ModelDataV2.Action,
                           lat_action_t: float, long_action_t: float, v_ego: float) -> log.ModelDataV2.Action:
-    plan = model_output['plan'][0]
+    plan = model_output['plan'][0] + RECOVERY_POWER*model_output['planplus'][0]
     desired_accel, should_stop = get_accel_from_plan(plan[:,Plan.VELOCITY][:,0],
                                                      plan[:,Plan.ACCELERATION][:,0],
                                                      ModelConstants.T_IDXS,
diff --git a/selfdrive/modeld/models/driving_policy.onnx b/selfdrive/modeld/models/driving_policy.onnx
index 1e764af9ba0680..7bf5d9aed4f462 100644
--- a/selfdrive/modeld/models/driving_policy.onnx
+++ b/selfdrive/modeld/models/driving_policy.onnx
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5a1f0655ddf266ed42ad1980389d96f47cc5e756da1fa3ca1477a920bb9b157
-size 13926324
+oid sha256:718f75921ab33a2c7d1acce6633f758c39202ed88ce28952a988323afaa77f90
+size 27812192
diff --git a/selfdrive/modeld/parse_model_outputs.py b/selfdrive/modeld/parse_model_outputs.py
index 038f51ca9cf2f9..0881a1ff9fa021 100644
--- a/selfdrive/modeld/parse_model_outputs.py
+++ b/selfdrive/modeld/parse_model_outputs.py
@@ -113,6 +113,7 @@ def parse_policy_outputs(self, outs: dict[str, np.ndarray]) -> dict[str, np.ndar
     plan_mhp = self.is_mhp(outs, 'plan',  ModelConstants.IDX_N * ModelConstants.PLAN_WIDTH)
     plan_in_N, plan_out_N = (ModelConstants.PLAN_MHP_N, ModelConstants.PLAN_MHP_SELECTION) if plan_mhp else (0, 0)
     self.parse_mdn('plan', outs, in_N=plan_in_N, out_N=plan_out_N, out_shape=(ModelConstants.IDX_N, ModelConstants.PLAN_WIDTH))
+    self.parse_mdn('planplus', outs, in_N=plan_in_N, out_N=plan_out_N, out_shape=(ModelConstants.IDX_N, ModelConstants.PLAN_WIDTH))
     self.parse_categorical_crossentropy('desire_state', outs, out_shape=(ModelConstants.DESIRE_PRED_WIDTH,))
     return outs