diff --git a/donkey/models.py b/donkey/models.py
index b02e9ed17..6252329e6 100644
--- a/donkey/models.py
+++ b/donkey/models.py
@@ -1,8 +1,8 @@
 """
 Keras model constructors.
 
-All models accept 120x160x3 images and output a 
-single floating point number (ie steering angle)  
+All models accept 120x160x3 images and output a
+single floating point number (ie steering angle)
 
 """
 
@@ -11,6 +11,7 @@
 from keras.models import Sequential
 from keras.layers import Convolution2D, MaxPooling2D, SimpleRNN, Reshape, BatchNormalization
 from keras.layers import Activation, Dropout, Flatten, Dense
+from keras.regularizers import l2
 
 def cnn3_full1():
 
@@ -81,7 +82,7 @@ def cnn1_full1():
 
     img_in = Input(shape=(120, 160, 3), name='img_in')
     angle_in = Input(shape=(1,), name='angle_in')
-    
+
     x = Convolution2D(1, 3, 3)(img_in)
     x = Activation('relu')(x)
     x = MaxPooling2D(pool_size=(2, 2))(x)
@@ -131,7 +132,7 @@ def norm_cnn3_full1():
 
 
 def vision_2D(dropout_frac=.2):
-    ''' 
+    '''
     Network with 4 convolutions, 2 residual shortcuts to predict angle.
     '''
     img_in = Input(shape=(120, 160, 3), name='img_in')
@@ -144,7 +145,7 @@ def vision_2D(dropout_frac=.2):
 
     #Create residual to shortcut
     aux1 = Flatten(name='aux1_flat')(net)
-    aux1 = Dense(64, name='aux1_dense')(aux1) 
+    aux1 = Dense(64, name='aux1_dense')(aux1)
 
     net =  Convolution2D(128, 3, 3, subsample=(2,2), border_mode='same', name='conv2')(net)
     net =  Dropout(dropout_frac)(net)
@@ -168,4 +169,39 @@ def vision_2D(dropout_frac=.2):
     angle_out = Dense(1, name='angle_out')(net)
     model = Model(input=[img_in], output=[angle_out])
     model.compile(optimizer='adam', loss='mean_squared_error')
-    return model 
\ No newline at end of file
+    return model
+
+
+def regularized_cnn4():
+    reg = l2(0.005)
+
+    img_in = Input(shape=(120, 160,3), name='img_in')
+    angle_in = Input(shape=(1,), name='angle_in')
+
+    x = img_in
+    x = Convolution2D(4, 3, 3,W_regularizer=reg)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D(pool_size=(2, 2))(x)
+
+    x = Convolution2D(8, 3, 3, W_regularizer=reg)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D(pool_size=(2, 2))(x)
+
+    x = Convolution2D(16, 3, 3, W_regularizer=reg)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D(pool_size=(2, 2))(x)
+
+    x = Convolution2D(32, 3, 3, W_regularizer=reg)(x)
+    x = Activation('relu')(x)
+    x = MaxPooling2D(pool_size=(2, 2))(x)
+    x = Flatten()(x)
+
+    x = Dense(128, W_regularizer=reg)(x)
+    x = Activation('linear')(x)
+    x = Dropout(.2)(x)
+
+    angle_out = Dense(1, name='angle_out')(x)
+
+    model = Model(input=[img_in], output=[angle_out])
+    model.compile(optimizer='adam', loss='mean_squared_error')
+    return model