diff --git a/y2012/control_loops/python/BUILD b/y2012/control_loops/python/BUILD
new file mode 100644
index 0000000..7e2b657
--- /dev/null
+++ b/y2012/control_loops/python/BUILD
@@ -0,0 +1,39 @@
+package(default_visibility = ['//y2012:__subpackages__'])
+
+py_binary(
+  name = 'drivetrain',
+  srcs = [
+    'drivetrain.py',
+  ],
+  deps = [
+    '//external:python-gflags',
+    '//external:python-glog',
+    '//frc971/control_loops/python:controls',
+  ],
+)
+
+py_binary(
+  name = 'polydrivetrain',
+  srcs = [
+    'polydrivetrain.py',
+    'drivetrain.py',
+  ],
+  deps = [
+    '//external:python-gflags',
+    '//external:python-glog',
+    '//frc971/control_loops/python:controls',
+  ],
+)
+
+py_library(
+  name = 'polydrivetrain_lib',
+  srcs = [
+    'polydrivetrain.py',
+    'drivetrain.py',
+  ],
+  deps = [
+    '//external:python-gflags',
+    '//external:python-glog',
+    '//frc971/control_loops/python:controls',
+  ],
+)
diff --git a/y2012/control_loops/python/drivetrain.py b/y2012/control_loops/python/drivetrain.py
new file mode 100755
index 0000000..291fe14
--- /dev/null
+++ b/y2012/control_loops/python/drivetrain.py
@@ -0,0 +1,351 @@
+#!/usr/bin/python
+
+from frc971.control_loops.python import control_loop
+from frc971.control_loops.python import controls
+import numpy
+import sys
+import argparse
+from matplotlib import pylab
+
+import gflags
+import glog
+
+FLAGS = gflags.FLAGS
+
+gflags.DEFINE_bool('plot', False, 'If true, plot the loop response.')
+
+class CIM(control_loop.ControlLoop):
+  def __init__(self):
+    super(CIM, self).__init__("CIM")
+    # Stall Torque in N m
+    self.stall_torque = 2.42
+    # Stall Current in Amps
+    self.stall_current = 133
+    # Free Speed in RPM
+    self.free_speed = 4650.0
+    # Free Current in Amps
+    self.free_current = 2.7
+    # Moment of inertia of the CIM in kg m^2
+    self.J = 0.0001
+    # Resistance of the motor, divided by 2 to account for the 2 motors
+    self.resistance = 12.0 / self.stall_current
+    # Motor velocity constant
+    self.Kv = ((self.free_speed / 60.0 * 2.0 * numpy.pi) /
+              (12.0 - self.resistance * self.free_current))
+    # Torque constant
+    self.Kt = self.stall_torque / self.stall_current
+    # Control loop time step
+    self.dt = 0.005
+
+    # State feedback matrices
+    self.A_continuous = numpy.matrix(
+        [[-self.Kt / self.Kv / (self.J * self.resistance)]])
+    self.B_continuous = numpy.matrix(
+        [[self.Kt / (self.J * self.resistance)]])
+    self.C = numpy.matrix([[1]])
+    self.D = numpy.matrix([[0]])
+
+    self.A, self.B = self.ContinuousToDiscrete(self.A_continuous,
+                                               self.B_continuous, self.dt)
+
+    self.PlaceControllerPoles([0.01])
+    self.PlaceObserverPoles([0.01])
+
+    self.U_max = numpy.matrix([[12.0]])
+    self.U_min = numpy.matrix([[-12.0]])
+
+    self.InitializeState()
+
+
+class Drivetrain(control_loop.ControlLoop):
+  def __init__(self, name="Drivetrain", left_low=True, right_low=True):
+    super(Drivetrain, self).__init__(name)
+    # Number of motors per side
+    self.num_motors = 2
+    # Stall Torque in N m
+    self.stall_torque = 2.42 * self.num_motors * 0.60
+    # Stall Current in Amps
+    self.stall_current = 133.0 * self.num_motors
+    # Free Speed in RPM. Used number from last year.
+    self.free_speed = 5500.0
+    # Free Current in Amps
+    self.free_current = 4.7 * self.num_motors
+    # Moment of inertia of the drivetrain in kg m^2
+    self.J = 1.5
+    # Mass of the robot, in kg.
+    self.m = 30
+    # Radius of the robot, in meters (from last year).
+    self.rb = 0.647998644 / 2.0
+    # Radius of the wheels, in meters.
+    self.r = .04445
+    # Resistance of the motor, divided by the number of motors.
+    self.resistance = 12.0 / self.stall_current
+    # Motor velocity constant
+    self.Kv = ((self.free_speed / 60.0 * 2.0 * numpy.pi) /
+               (12.0 - self.resistance * self.free_current))
+    # Torque constant
+    self.Kt = self.stall_torque / self.stall_current
+    # Gear ratios
+    self.G_low = 15.0 / 60.0 * 15.0 / 50.0
+    self.G_high = 30.0 / 45.0 * 15.0 / 50.0
+    if left_low:
+      self.Gl = self.G_low
+    else:
+      self.Gl = self.G_high
+    if right_low:
+      self.Gr = self.G_low
+    else:
+      self.Gr = self.G_high
+
+    # Control loop time step
+    self.dt = 0.005
+
+    # These describe the way that a given side of a robot will be influenced
+    # by the other side. Units of 1 / kg.
+    self.msp = 1.0 / self.m + self.rb * self.rb / self.J
+    self.msn = 1.0 / self.m - self.rb * self.rb / self.J
+    # The calculations which we will need for A and B.
+    self.tcl = -self.Kt / self.Kv / (self.Gl * self.Gl * self.resistance * self.r * self.r)
+    self.tcr = -self.Kt / self.Kv / (self.Gr * self.Gr * self.resistance * self.r * self.r)
+    self.mpl = self.Kt / (self.Gl * self.resistance * self.r)
+    self.mpr = self.Kt / (self.Gr * self.resistance * self.r)
+
+    # State feedback matrices
+    # X will be of the format
+    # [[positionl], [velocityl], [positionr], velocityr]]
+    self.A_continuous = numpy.matrix(
+        [[0, 1, 0, 0],
+         [0, self.msp * self.tcl, 0, self.msn * self.tcr],
+         [0, 0, 0, 1],
+         [0, self.msn * self.tcl, 0, self.msp * self.tcr]])
+    self.B_continuous = numpy.matrix(
+        [[0, 0],
+         [self.msp * self.mpl, self.msn * self.mpr],
+         [0, 0],
+         [self.msn * self.mpl, self.msp * self.mpr]])
+    self.C = numpy.matrix([[1, 0, 0, 0],
+                           [0, 0, 1, 0]])
+    self.D = numpy.matrix([[0, 0],
+                           [0, 0]])
+
+    self.A, self.B = self.ContinuousToDiscrete(
+        self.A_continuous, self.B_continuous, self.dt)
+
+    q_pos = 0.12
+    q_vel = 1.0
+    self.Q = numpy.matrix([[(1.0 / (q_pos ** 2.0)), 0.0, 0.0, 0.0],
+                           [0.0, (1.0 / (q_vel ** 2.0)), 0.0, 0.0],
+                           [0.0, 0.0, (1.0 / (q_pos ** 2.0)), 0.0],
+                           [0.0, 0.0, 0.0, (1.0 / (q_vel ** 2.0))]])
+
+    self.R = numpy.matrix([[(1.0 / (12.0 ** 2.0)), 0.0],
+                           [0.0, (1.0 / (12.0 ** 2.0))]])
+    self.K = controls.dlqr(self.A, self.B, self.Q, self.R)
+
+    glog.debug('DT K %s', name)
+    glog.debug(str(self.K))
+    glog.debug(str(numpy.linalg.eig(self.A - self.B * self.K)[0]))
+
+    self.hlp = 0.3
+    self.llp = 0.4
+    self.PlaceObserverPoles([self.hlp, self.hlp, self.llp, self.llp])
+
+    self.U_max = numpy.matrix([[12.0], [12.0]])
+    self.U_min = numpy.matrix([[-12.0], [-12.0]])
+    self.InitializeState()
+
+
+class KFDrivetrain(Drivetrain):
+  def __init__(self, name="KFDrivetrain", left_low=True, right_low=True):
+    super(KFDrivetrain, self).__init__(name, left_low, right_low)
+
+    self.unaugmented_A_continuous = self.A_continuous
+    self.unaugmented_B_continuous = self.B_continuous
+
+    # The states are
+    # The practical voltage applied to the wheels is
+    #   V_left = U_left + left_voltage_error
+    #
+    # [left position, left velocity, right position, right velocity,
+    #  left voltage error, right voltage error, angular_error]
+    self.A_continuous = numpy.matrix(numpy.zeros((7, 7)))
+    self.B_continuous = numpy.matrix(numpy.zeros((7, 2)))
+    self.A_continuous[0:4,0:4] = self.unaugmented_A_continuous
+    self.A_continuous[0:4,4:6] = self.unaugmented_B_continuous
+    self.B_continuous[0:4,0:2] = self.unaugmented_B_continuous
+    self.A_continuous[0,6] = 1
+    self.A_continuous[2,6] = -1
+
+    self.A, self.B = self.ContinuousToDiscrete(
+        self.A_continuous, self.B_continuous, self.dt)
+
+    self.C = numpy.matrix([[1, 0, 0, 0, 0, 0, 0],
+                           [0, 0, 1, 0, 0, 0, 0],
+                           [0, -0.5 / self.rb, 0, 0.5 / self.rb, 0, 0, 0]])
+
+    self.D = numpy.matrix([[0, 0],
+                           [0, 0],
+                           [0, 0]])
+
+    q_pos = 0.05
+    q_vel = 1.00
+    q_voltage = 10.0
+    q_encoder_uncertainty = 2.00
+
+    self.Q = numpy.matrix([[(q_pos ** 2.0), 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
+                           [0.0, (q_vel ** 2.0), 0.0, 0.0, 0.0, 0.0, 0.0],
+                           [0.0, 0.0, (q_pos ** 2.0), 0.0, 0.0, 0.0, 0.0],
+                           [0.0, 0.0, 0.0, (q_vel ** 2.0), 0.0, 0.0, 0.0],
+                           [0.0, 0.0, 0.0, 0.0, (q_voltage ** 2.0), 0.0, 0.0],
+                           [0.0, 0.0, 0.0, 0.0, 0.0, (q_voltage ** 2.0), 0.0],
+                           [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, (q_encoder_uncertainty ** 2.0)]])
+
+    r_pos =  0.0001
+    r_gyro = 0.000001
+    self.R = numpy.matrix([[(r_pos ** 2.0), 0.0, 0.0],
+                           [0.0, (r_pos ** 2.0), 0.0],
+                           [0.0, 0.0, (r_gyro ** 2.0)]])
+
+    # Solving for kf gains.
+    self.KalmanGain, self.Q_steady = controls.kalman(
+        A=self.A, B=self.B, C=self.C, Q=self.Q, R=self.R)
+
+    self.L = self.A * self.KalmanGain
+
+    # We need a nothing controller for the autogen code to be happy.
+    self.K = numpy.matrix(numpy.zeros((self.B.shape[1], self.A.shape[0])))
+
+
+def main(argv):
+  argv = FLAGS(argv)
+
+  # Simulate the response of the system to a step input.
+  drivetrain = Drivetrain()
+  simulated_left = []
+  simulated_right = []
+  for _ in xrange(100):
+    drivetrain.Update(numpy.matrix([[12.0], [12.0]]))
+    simulated_left.append(drivetrain.X[0, 0])
+    simulated_right.append(drivetrain.X[2, 0])
+
+  if FLAGS.plot:
+    pylab.plot(range(100), simulated_left)
+    pylab.plot(range(100), simulated_right)
+    pylab.show()
+
+  # Simulate forwards motion.
+  drivetrain = Drivetrain()
+  close_loop_left = []
+  close_loop_right = []
+  R = numpy.matrix([[1.0], [0.0], [1.0], [0.0]])
+  for _ in xrange(100):
+    U = numpy.clip(drivetrain.K * (R - drivetrain.X_hat),
+                   drivetrain.U_min, drivetrain.U_max)
+    drivetrain.UpdateObserver(U)
+    drivetrain.Update(U)
+    close_loop_left.append(drivetrain.X[0, 0])
+    close_loop_right.append(drivetrain.X[2, 0])
+
+  if FLAGS.plot:
+    pylab.plot(range(100), close_loop_left)
+    pylab.plot(range(100), close_loop_right)
+    pylab.show()
+
+  # Try turning in place
+  drivetrain = Drivetrain()
+  close_loop_left = []
+  close_loop_right = []
+  R = numpy.matrix([[-1.0], [0.0], [1.0], [0.0]])
+  for _ in xrange(100):
+    U = numpy.clip(drivetrain.K * (R - drivetrain.X_hat),
+                   drivetrain.U_min, drivetrain.U_max)
+    drivetrain.UpdateObserver(U)
+    drivetrain.Update(U)
+    close_loop_left.append(drivetrain.X[0, 0])
+    close_loop_right.append(drivetrain.X[2, 0])
+
+  if FLAGS.plot:
+    pylab.plot(range(100), close_loop_left)
+    pylab.plot(range(100), close_loop_right)
+    pylab.show()
+
+  # Try turning just one side.
+  drivetrain = Drivetrain()
+  close_loop_left = []
+  close_loop_right = []
+  R = numpy.matrix([[0.0], [0.0], [1.0], [0.0]])
+  for _ in xrange(100):
+    U = numpy.clip(drivetrain.K * (R - drivetrain.X_hat),
+                   drivetrain.U_min, drivetrain.U_max)
+    drivetrain.UpdateObserver(U)
+    drivetrain.Update(U)
+    close_loop_left.append(drivetrain.X[0, 0])
+    close_loop_right.append(drivetrain.X[2, 0])
+
+  if FLAGS.plot:
+    pylab.plot(range(100), close_loop_left)
+    pylab.plot(range(100), close_loop_right)
+    pylab.show()
+
+  # Write the generated constants out to a file.
+  drivetrain_low_low = Drivetrain(
+      name="DrivetrainLowLow", left_low=True, right_low=True)
+  drivetrain_low_high = Drivetrain(
+      name="DrivetrainLowHigh", left_low=True, right_low=False)
+  drivetrain_high_low = Drivetrain(
+      name="DrivetrainHighLow", left_low=False, right_low=True)
+  drivetrain_high_high = Drivetrain(
+      name="DrivetrainHighHigh", left_low=False, right_low=False)
+
+  kf_drivetrain_low_low = KFDrivetrain(
+      name="KFDrivetrainLowLow", left_low=True, right_low=True)
+  kf_drivetrain_low_high = KFDrivetrain(
+      name="KFDrivetrainLowHigh", left_low=True, right_low=False)
+  kf_drivetrain_high_low = KFDrivetrain(
+      name="KFDrivetrainHighLow", left_low=False, right_low=True)
+  kf_drivetrain_high_high = KFDrivetrain(
+      name="KFDrivetrainHighHigh", left_low=False, right_low=False)
+
+  if len(argv) != 5:
+    print "Expected .h file name and .cc file name"
+  else:
+    namespaces = ['y2012', 'control_loops', 'drivetrain']
+    dog_loop_writer = control_loop.ControlLoopWriter(
+        "Drivetrain", [drivetrain_low_low, drivetrain_low_high,
+                       drivetrain_high_low, drivetrain_high_high],
+        namespaces = namespaces)
+    dog_loop_writer.AddConstant(control_loop.Constant("kDt", "%f",
+          drivetrain_low_low.dt))
+    dog_loop_writer.AddConstant(control_loop.Constant("kStallTorque", "%f",
+          drivetrain_low_low.stall_torque))
+    dog_loop_writer.AddConstant(control_loop.Constant("kStallCurrent", "%f",
+          drivetrain_low_low.stall_current))
+    dog_loop_writer.AddConstant(control_loop.Constant("kFreeSpeedRPM", "%f",
+          drivetrain_low_low.free_speed))
+    dog_loop_writer.AddConstant(control_loop.Constant("kFreeCurrent", "%f",
+          drivetrain_low_low.free_current))
+    dog_loop_writer.AddConstant(control_loop.Constant("kJ", "%f",
+          drivetrain_low_low.J))
+    dog_loop_writer.AddConstant(control_loop.Constant("kMass", "%f",
+          drivetrain_low_low.m))
+    dog_loop_writer.AddConstant(control_loop.Constant("kRobotRadius", "%f",
+          drivetrain_low_low.rb))
+    dog_loop_writer.AddConstant(control_loop.Constant("kWheelRadius", "%f",
+          drivetrain_low_low.r))
+    dog_loop_writer.AddConstant(control_loop.Constant("kR", "%f",
+          drivetrain_low_low.resistance))
+    dog_loop_writer.AddConstant(control_loop.Constant("kV", "%f",
+          drivetrain_low_low.Kv))
+    dog_loop_writer.AddConstant(control_loop.Constant("kT", "%f",
+          drivetrain_low_low.Kt))
+
+    dog_loop_writer.Write(argv[1], argv[2])
+
+    kf_loop_writer = control_loop.ControlLoopWriter(
+        "KFDrivetrain", [kf_drivetrain_low_low, kf_drivetrain_low_high,
+                         kf_drivetrain_high_low, kf_drivetrain_high_high],
+        namespaces = namespaces)
+    kf_loop_writer.Write(argv[3], argv[4])
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv))
diff --git a/y2012/control_loops/python/polydrivetrain.py b/y2012/control_loops/python/polydrivetrain.py
new file mode 100755
index 0000000..9948ff2
--- /dev/null
+++ b/y2012/control_loops/python/polydrivetrain.py
@@ -0,0 +1,512 @@
+#!/usr/bin/python
+
+import numpy
+import sys
+from frc971.control_loops.python import polytope
+from y2012.control_loops.python import drivetrain
+from frc971.control_loops.python import control_loop
+from frc971.control_loops.python import controls
+from matplotlib import pylab
+
+import gflags
+import glog
+
+__author__ = 'Austin Schuh (austin.linux@gmail.com)'
+
+FLAGS = gflags.FLAGS
+
+try:
+  gflags.DEFINE_bool('plot', False, 'If true, plot the loop response.')
+except gflags.DuplicateFlagError:
+  pass
+
+def CoerceGoal(region, K, w, R):
+  """Intersects a line with a region, and finds the closest point to R.
+
+  Finds a point that is closest to R inside the region, and on the line
+  defined by K X = w.  If it is not possible to find a point on the line,
+  finds a point that is inside the region and closest to the line.  This
+  function assumes that
+
+  Args:
+    region: HPolytope, the valid goal region.
+    K: numpy.matrix (2 x 1), the matrix for the equation [K1, K2] [x1; x2] = w
+    w: float, the offset in the equation above.
+    R: numpy.matrix (2 x 1), the point to be closest to.
+
+  Returns:
+    numpy.matrix (2 x 1), the point.
+  """
+  return DoCoerceGoal(region, K, w, R)[0]
+
+def DoCoerceGoal(region, K, w, R):
+  if region.IsInside(R):
+    return (R, True)
+
+  perpendicular_vector = K.T / numpy.linalg.norm(K)
+  parallel_vector = numpy.matrix([[perpendicular_vector[1, 0]],
+                                  [-perpendicular_vector[0, 0]]])
+
+  # We want to impose the constraint K * X = w on the polytope H * X <= k.
+  # We do this by breaking X up into parallel and perpendicular components to
+  # the half plane.  This gives us the following equation.
+  #
+  #  parallel * (parallel.T \dot X) + perpendicular * (perpendicular \dot X)) = X
+  #
+  # Then, substitute this into the polytope.
+  #
+  #  H * (parallel * (parallel.T \dot X) + perpendicular * (perpendicular \dot X)) <= k
+  #
+  # Substitute K * X = w
+  #
+  # H * parallel * (parallel.T \dot X) + H * perpendicular * w <= k
+  #
+  # Move all the knowns to the right side.
+  #
+  # H * parallel * ([parallel1 parallel2] * X) <= k - H * perpendicular * w
+  #
+  # Let t = parallel.T \dot X, the component parallel to the surface.
+  #
+  # H * parallel * t <= k - H * perpendicular * w
+  #
+  # This is a polytope which we can solve, and use to figure out the range of X
+  # that we care about!
+
+  t_poly = polytope.HPolytope(
+      region.H * parallel_vector,
+      region.k - region.H * perpendicular_vector * w)
+
+  vertices = t_poly.Vertices()
+
+  if vertices.shape[0]:
+    # The region exists!
+    # Find the closest vertex
+    min_distance = numpy.infty
+    closest_point = None
+    for vertex in vertices:
+      point = parallel_vector * vertex + perpendicular_vector * w
+      length = numpy.linalg.norm(R - point)
+      if length < min_distance:
+        min_distance = length
+        closest_point = point
+
+    return (closest_point, True)
+  else:
+    # Find the vertex of the space that is closest to the line.
+    region_vertices = region.Vertices()
+    min_distance = numpy.infty
+    closest_point = None
+    for vertex in region_vertices:
+      point = vertex.T
+      length = numpy.abs((perpendicular_vector.T * point)[0, 0])
+      if length < min_distance:
+        min_distance = length
+        closest_point = point
+
+    return (closest_point, False)
+
+
+class VelocityDrivetrainModel(control_loop.ControlLoop):
+  def __init__(self, left_low=True, right_low=True, name="VelocityDrivetrainModel"):
+    super(VelocityDrivetrainModel, self).__init__(name)
+    self._drivetrain = drivetrain.Drivetrain(left_low=left_low,
+                                             right_low=right_low)
+    self.dt = 0.005
+    self.A_continuous = numpy.matrix(
+        [[self._drivetrain.A_continuous[1, 1], self._drivetrain.A_continuous[1, 3]],
+         [self._drivetrain.A_continuous[3, 1], self._drivetrain.A_continuous[3, 3]]])
+
+    self.B_continuous = numpy.matrix(
+        [[self._drivetrain.B_continuous[1, 0], self._drivetrain.B_continuous[1, 1]],
+         [self._drivetrain.B_continuous[3, 0], self._drivetrain.B_continuous[3, 1]]])
+    self.C = numpy.matrix(numpy.eye(2))
+    self.D = numpy.matrix(numpy.zeros((2, 2)))
+
+    self.A, self.B = self.ContinuousToDiscrete(self.A_continuous,
+                                               self.B_continuous, self.dt)
+
+    # FF * X = U (steady state)
+    self.FF = self.B.I * (numpy.eye(2) - self.A)
+
+    self.PlaceControllerPoles([0.8, 0.8])
+    self.PlaceObserverPoles([0.02, 0.02])
+
+    self.G_high = self._drivetrain.G_high
+    self.G_low = self._drivetrain.G_low
+    self.R = self._drivetrain.R
+    self.r = self._drivetrain.r
+    self.Kv = self._drivetrain.Kv
+    self.Kt = self._drivetrain.Kt
+
+    self.U_max = self._drivetrain.U_max
+    self.U_min = self._drivetrain.U_min
+
+
+class VelocityDrivetrain(object):
+  HIGH = 'high'
+  LOW = 'low'
+  SHIFTING_UP = 'up'
+  SHIFTING_DOWN = 'down'
+
+  def __init__(self):
+    self.drivetrain_low_low = VelocityDrivetrainModel(
+        left_low=True, right_low=True, name='VelocityDrivetrainLowLow')
+    self.drivetrain_low_high = VelocityDrivetrainModel(left_low=True, right_low=False, name='VelocityDrivetrainLowHigh')
+    self.drivetrain_high_low = VelocityDrivetrainModel(left_low=False, right_low=True, name = 'VelocityDrivetrainHighLow')
+    self.drivetrain_high_high = VelocityDrivetrainModel(left_low=False, right_low=False, name = 'VelocityDrivetrainHighHigh')
+
+    # X is [lvel, rvel]
+    self.X = numpy.matrix(
+        [[0.0],
+         [0.0]])
+
+    self.U_poly = polytope.HPolytope(
+        numpy.matrix([[1, 0],
+                      [-1, 0],
+                      [0, 1],
+                      [0, -1]]),
+        numpy.matrix([[12],
+                      [12],
+                      [12],
+                      [12]]))
+
+    self.U_max = numpy.matrix(
+        [[12.0],
+         [12.0]])
+    self.U_min = numpy.matrix(
+        [[-12.0000000000],
+         [-12.0000000000]])
+
+    self.dt = 0.005
+
+    self.R = numpy.matrix(
+        [[0.0],
+         [0.0]])
+
+    # ttrust is the comprimise between having full throttle negative inertia,
+    # and having no throttle negative inertia.  A value of 0 is full throttle
+    # inertia.  A value of 1 is no throttle negative inertia.
+    self.ttrust = 1.0
+
+    self.left_gear = VelocityDrivetrain.LOW
+    self.right_gear = VelocityDrivetrain.LOW
+    self.left_shifter_position = 0.0
+    self.right_shifter_position = 0.0
+    self.left_cim = drivetrain.CIM()
+    self.right_cim = drivetrain.CIM()
+
+  def IsInGear(self, gear):
+    return gear is VelocityDrivetrain.HIGH or gear is VelocityDrivetrain.LOW
+
+  def MotorRPM(self, shifter_position, velocity):
+    if shifter_position > 0.5:
+      return (velocity / self.CurrentDrivetrain().G_high /
+              self.CurrentDrivetrain().r)
+    else:
+      return (velocity / self.CurrentDrivetrain().G_low /
+              self.CurrentDrivetrain().r)
+
+  def CurrentDrivetrain(self):
+    if self.left_shifter_position > 0.5:
+      if self.right_shifter_position > 0.5:
+        return self.drivetrain_high_high
+      else:
+        return self.drivetrain_high_low
+    else:
+      if self.right_shifter_position > 0.5:
+        return self.drivetrain_low_high
+      else:
+        return self.drivetrain_low_low
+
+  def SimShifter(self, gear, shifter_position):
+    if gear is VelocityDrivetrain.HIGH or gear is VelocityDrivetrain.SHIFTING_UP:
+      shifter_position = min(shifter_position + 0.5, 1.0)
+    else:
+      shifter_position = max(shifter_position - 0.5, 0.0)
+
+    if shifter_position == 1.0:
+      gear = VelocityDrivetrain.HIGH
+    elif shifter_position == 0.0:
+      gear = VelocityDrivetrain.LOW
+
+    return gear, shifter_position
+
+  def ComputeGear(self, wheel_velocity, should_print=False, current_gear=False, gear_name=None):
+    high_omega = (wheel_velocity / self.CurrentDrivetrain().G_high /
+                  self.CurrentDrivetrain().r)
+    low_omega = (wheel_velocity / self.CurrentDrivetrain().G_low /
+                 self.CurrentDrivetrain().r)
+    #print gear_name, "Motor Energy Difference.", 0.5 * 0.000140032647 * (low_omega * low_omega - high_omega * high_omega), "joules"
+    high_torque = ((12.0 - high_omega / self.CurrentDrivetrain().Kv) *
+                   self.CurrentDrivetrain().Kt / self.CurrentDrivetrain().R)
+    low_torque = ((12.0 - low_omega / self.CurrentDrivetrain().Kv) *
+                  self.CurrentDrivetrain().Kt / self.CurrentDrivetrain().R)
+    high_power = high_torque * high_omega
+    low_power = low_torque * low_omega
+    #if should_print:
+    #  print gear_name, "High omega", high_omega, "Low omega", low_omega
+    #  print gear_name, "High torque", high_torque, "Low torque", low_torque
+    #  print gear_name, "High power", high_power, "Low power", low_power
+
+    # Shift algorithm improvements.
+    # TODO(aschuh):
+    # It takes time to shift.  Shifting down for 1 cycle doesn't make sense
+    # because you will end up slower than without shifting.  Figure out how
+    # to include that info.
+    # If the driver is still in high gear, but isn't asking for the extra power
+    # from low gear, don't shift until he asks for it.
+    goal_gear_is_high = high_power > low_power
+    #goal_gear_is_high = True
+
+    if not self.IsInGear(current_gear):
+      glog.debug('%s Not in gear.', gear_name)
+      return current_gear
+    else:
+      is_high = current_gear is VelocityDrivetrain.HIGH
+      if is_high != goal_gear_is_high:
+        if goal_gear_is_high:
+          glog.debug('%s Shifting up.', gear_name)
+          return VelocityDrivetrain.SHIFTING_UP
+        else:
+          glog.debug('%s Shifting down.', gear_name)
+          return VelocityDrivetrain.SHIFTING_DOWN
+      else:
+        return current_gear
+
+  def FilterVelocity(self, throttle):
+    # Invert the plant to figure out how the velocity filter would have to work
+    # out in order to filter out the forwards negative inertia.
+    # This math assumes that the left and right power and velocity are equal.
+
+    # The throttle filter should filter such that the motor in the highest gear
+    # should be controlling the time constant.
+    # Do this by finding the index of FF that has the lowest value, and computing
+    # the sums using that index.
+    FF_sum = self.CurrentDrivetrain().FF.sum(axis=1)
+    min_FF_sum_index = numpy.argmin(FF_sum)
+    min_FF_sum = FF_sum[min_FF_sum_index, 0]
+    min_K_sum = self.CurrentDrivetrain().K[min_FF_sum_index, :].sum()
+    # Compute the FF sum for high gear.
+    high_min_FF_sum = self.drivetrain_high_high.FF[0, :].sum()
+
+    # U = self.K[0, :].sum() * (R - x_avg) + self.FF[0, :].sum() * R
+    # throttle * 12.0 = (self.K[0, :].sum() + self.FF[0, :].sum()) * R
+    #                   - self.K[0, :].sum() * x_avg
+
+    # R = (throttle * 12.0 + self.K[0, :].sum() * x_avg) /
+    #     (self.K[0, :].sum() + self.FF[0, :].sum())
+
+    # U = (K + FF) * R - K * X
+    # (K + FF) ^-1 * (U + K * X) = R
+
+    # Scale throttle by min_FF_sum / high_min_FF_sum.  This will make low gear
+    # have the same velocity goal as high gear, and so that the robot will hold
+    # the same speed for the same throttle for all gears.
+    adjusted_ff_voltage = numpy.clip(throttle * 12.0 * min_FF_sum / high_min_FF_sum, -12.0, 12.0)
+    return ((adjusted_ff_voltage + self.ttrust * min_K_sum * (self.X[0, 0] + self.X[1, 0]) / 2.0)
+            / (self.ttrust * min_K_sum + min_FF_sum))
+
+  def Update(self, throttle, steering):
+    # Shift into the gear which sends the most power to the floor.
+    # This is the same as sending the most torque down to the floor at the
+    # wheel.
+
+    self.left_gear = self.right_gear = True
+    if True:
+      self.left_gear = self.ComputeGear(self.X[0, 0], should_print=True,
+                                        current_gear=self.left_gear,
+                                        gear_name="left")
+      self.right_gear = self.ComputeGear(self.X[1, 0], should_print=True,
+                                         current_gear=self.right_gear,
+                                         gear_name="right")
+      if self.IsInGear(self.left_gear):
+        self.left_cim.X[0, 0] = self.MotorRPM(self.left_shifter_position, self.X[0, 0])
+
+      if self.IsInGear(self.right_gear):
+        self.right_cim.X[0, 0] = self.MotorRPM(self.right_shifter_position, self.X[0, 0])
+
+    if self.IsInGear(self.left_gear) and self.IsInGear(self.right_gear):
+      # Filter the throttle to provide a nicer response.
+      fvel = self.FilterVelocity(throttle)
+
+      # Constant radius means that angualar_velocity / linear_velocity = constant.
+      # Compute the left and right velocities.
+      steering_velocity = numpy.abs(fvel) * steering
+      left_velocity = fvel - steering_velocity
+      right_velocity = fvel + steering_velocity
+
+      # Write this constraint in the form of K * R = w
+      # angular velocity / linear velocity = constant
+      # (left - right) / (left + right) = constant
+      # left - right = constant * left + constant * right
+
+      # (fvel - steering * numpy.abs(fvel) - fvel - steering * numpy.abs(fvel)) /
+      #  (fvel - steering * numpy.abs(fvel) + fvel + steering * numpy.abs(fvel)) =
+      #       constant
+      # (- 2 * steering * numpy.abs(fvel)) / (2 * fvel) = constant
+      # (-steering * sign(fvel)) = constant
+      # (-steering * sign(fvel)) * (left + right) = left - right
+      # (steering * sign(fvel) + 1) * left + (steering * sign(fvel) - 1) * right = 0
+
+      equality_k = numpy.matrix(
+          [[1 + steering * numpy.sign(fvel), -(1 - steering * numpy.sign(fvel))]])
+      equality_w = 0.0
+
+      self.R[0, 0] = left_velocity
+      self.R[1, 0] = right_velocity
+
+      # Construct a constraint on R by manipulating the constraint on U
+      # Start out with H * U <= k
+      # U = FF * R + K * (R - X)
+      # H * (FF * R + K * R - K * X) <= k
+      # H * (FF + K) * R <= k + H * K * X
+      R_poly = polytope.HPolytope(
+          self.U_poly.H * (self.CurrentDrivetrain().K + self.CurrentDrivetrain().FF),
+          self.U_poly.k + self.U_poly.H * self.CurrentDrivetrain().K * self.X)
+
+      # Limit R back inside the box.
+      self.boxed_R = CoerceGoal(R_poly, equality_k, equality_w, self.R)
+
+      FF_volts = self.CurrentDrivetrain().FF * self.boxed_R
+      self.U_ideal = self.CurrentDrivetrain().K * (self.boxed_R - self.X) + FF_volts
+    else:
+      glog.debug('Not all in gear')
+      if not self.IsInGear(self.left_gear) and not self.IsInGear(self.right_gear):
+        # TODO(austin): Use battery volts here.
+        R_left = self.MotorRPM(self.left_shifter_position, self.X[0, 0])
+        self.U_ideal[0, 0] = numpy.clip(
+            self.left_cim.K * (R_left - self.left_cim.X) + R_left / self.left_cim.Kv,
+            self.left_cim.U_min, self.left_cim.U_max)
+        self.left_cim.Update(self.U_ideal[0, 0])
+
+        R_right = self.MotorRPM(self.right_shifter_position, self.X[1, 0])
+        self.U_ideal[1, 0] = numpy.clip(
+            self.right_cim.K * (R_right - self.right_cim.X) + R_right / self.right_cim.Kv,
+            self.right_cim.U_min, self.right_cim.U_max)
+        self.right_cim.Update(self.U_ideal[1, 0])
+      else:
+        assert False
+
+    self.U = numpy.clip(self.U_ideal, self.U_min, self.U_max)
+
+    # TODO(austin): Model the robot as not accelerating when you shift...
+    # This hack only works when you shift at the same time.
+    if self.IsInGear(self.left_gear) and self.IsInGear(self.right_gear):
+      self.X = self.CurrentDrivetrain().A * self.X + self.CurrentDrivetrain().B * self.U
+
+    self.left_gear, self.left_shifter_position = self.SimShifter(
+        self.left_gear, self.left_shifter_position)
+    self.right_gear, self.right_shifter_position = self.SimShifter(
+        self.right_gear, self.right_shifter_position)
+
+    glog.debug('U is %s %s', str(self.U[0, 0]), str(self.U[1, 0]))
+    glog.debug('Left shifter %s %d Right shifter %s %d',
+               self.left_gear, self.left_shifter_position,
+               self.right_gear, self.right_shifter_position)
+
+
+def main(argv):
+  argv = FLAGS(argv)
+
+  vdrivetrain = VelocityDrivetrain()
+
+  if len(argv) != 5:
+    glog.fatal('Expected .h file name and .cc file name')
+  else:
+    namespaces = ['y2012', 'control_loops', 'drivetrain']
+    dog_loop_writer = control_loop.ControlLoopWriter(
+        "VelocityDrivetrain", [vdrivetrain.drivetrain_low_low,
+                       vdrivetrain.drivetrain_low_high,
+                       vdrivetrain.drivetrain_high_low,
+                       vdrivetrain.drivetrain_high_high],
+                       namespaces=namespaces)
+
+    dog_loop_writer.Write(argv[1], argv[2])
+
+    cim_writer = control_loop.ControlLoopWriter(
+        "CIM", [drivetrain.CIM()])
+
+    cim_writer.Write(argv[3], argv[4])
+    return
+
+  vl_plot = []
+  vr_plot = []
+  ul_plot = []
+  ur_plot = []
+  radius_plot = []
+  t_plot = []
+  left_gear_plot = []
+  right_gear_plot = []
+  vdrivetrain.left_shifter_position = 0.0
+  vdrivetrain.right_shifter_position = 0.0
+  vdrivetrain.left_gear = VelocityDrivetrain.LOW
+  vdrivetrain.right_gear = VelocityDrivetrain.LOW
+
+  glog.debug('K is %s', str(vdrivetrain.CurrentDrivetrain().K))
+
+  if vdrivetrain.left_gear is VelocityDrivetrain.HIGH:
+    glog.debug('Left is high')
+  else:
+    glog.debug('Left is low')
+  if vdrivetrain.right_gear is VelocityDrivetrain.HIGH:
+    glog.debug('Right is high')
+  else:
+    glog.debug('Right is low')
+
+  for t in numpy.arange(0, 1.7, vdrivetrain.dt):
+    if t < 0.5:
+      vdrivetrain.Update(throttle=0.00, steering=1.0)
+    elif t < 1.2:
+      vdrivetrain.Update(throttle=0.5, steering=1.0)
+    else:
+      vdrivetrain.Update(throttle=0.00, steering=1.0)
+    t_plot.append(t)
+    vl_plot.append(vdrivetrain.X[0, 0])
+    vr_plot.append(vdrivetrain.X[1, 0])
+    ul_plot.append(vdrivetrain.U[0, 0])
+    ur_plot.append(vdrivetrain.U[1, 0])
+    left_gear_plot.append((vdrivetrain.left_gear is VelocityDrivetrain.HIGH) * 2.0 - 10.0)
+    right_gear_plot.append((vdrivetrain.right_gear is VelocityDrivetrain.HIGH) * 2.0 - 10.0)
+
+    fwd_velocity = (vdrivetrain.X[1, 0] + vdrivetrain.X[0, 0]) / 2
+    turn_velocity = (vdrivetrain.X[1, 0] - vdrivetrain.X[0, 0])
+    if abs(fwd_velocity) < 0.0000001:
+      radius_plot.append(turn_velocity)
+    else:
+      radius_plot.append(turn_velocity / fwd_velocity)
+
+  cim_velocity_plot = []
+  cim_voltage_plot = []
+  cim_time = []
+  cim = drivetrain.CIM()
+  R = numpy.matrix([[300]])
+  for t in numpy.arange(0, 0.5, cim.dt):
+    U = numpy.clip(cim.K * (R - cim.X) + R / cim.Kv, cim.U_min, cim.U_max)
+    cim.Update(U)
+    cim_velocity_plot.append(cim.X[0, 0])
+    cim_voltage_plot.append(U[0, 0] * 10)
+    cim_time.append(t)
+  pylab.plot(cim_time, cim_velocity_plot, label='cim spinup')
+  pylab.plot(cim_time, cim_voltage_plot, label='cim voltage')
+  pylab.legend()
+  pylab.show()
+
+  # TODO(austin):
+  # Shifting compensation.
+
+  # Tighten the turn.
+  # Closed loop drive.
+
+  pylab.plot(t_plot, vl_plot, label='left velocity')
+  pylab.plot(t_plot, vr_plot, label='right velocity')
+  pylab.plot(t_plot, ul_plot, label='left voltage')
+  pylab.plot(t_plot, ur_plot, label='right voltage')
+  pylab.plot(t_plot, radius_plot, label='radius')
+  pylab.plot(t_plot, left_gear_plot, label='left gear high')
+  pylab.plot(t_plot, right_gear_plot, label='right gear high')
+  pylab.legend()
+  pylab.show()
+  return 0
+
+if __name__ == '__main__':
+  sys.exit(main(sys.argv))