Go to the documentation of this file.00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00025
00026
00027
00028
00029
00030
00031
00032
00033
00034
00035
00036
00037
00038
00039
00040
00041
00042
00043
00044
00045 import numpy
00046 from numpy import reshape, array, zeros, diag, matrix, real
00047 import roslib; roslib.load_manifest('calibration_estimation')
00048 import rospy
00049 from calibration_estimation.full_chain import FullChainRobotParams
00050 from sensor_msgs.msg import JointState
00051
00052 class ChainBundler:
00053 def __init__(self, valid_configs):
00054 self._valid_configs = valid_configs
00055
00056
00057 def build_blocks(self, M_robot):
00058 sensors = []
00059 for cur_config in self._valid_configs:
00060 if cur_config["sensor_id"] == M_robot.chain_id and \
00061 cur_config["sensor_id"] in [ x.chain_id for x in M_robot.M_chain ]:
00062 M_chain = [x for x in M_robot.M_chain if cur_config["sensor_id"] == x.chain_id][0]
00063 cur_sensor = ChainSensor(cur_config, M_chain, M_robot.target_id)
00064 sensors.append(cur_sensor)
00065 else:
00066 rospy.logdebug(" Didn't find block")
00067 return sensors
00068
00069 class ChainSensor:
00070 def __init__(self, config_dict, M_chain, target_id):
00071
00072 self.sensor_type = "chain"
00073 self.sensor_id = config_dict["sensor_id"]
00074
00075 self._config_dict = config_dict
00076 self._M_chain = M_chain
00077 self._target_id = target_id
00078
00079 self._full_chain = FullChainRobotParams(self.sensor_id, self.sensor_id+"_cb_link")
00080
00081 self.terms_per_sample = 3
00082
00083 def update_config(self, robot_params):
00084 self._full_chain.update_config(robot_params)
00085 self._checkerboard = robot_params.checkerboards[ self._target_id ]
00086
00087 def compute_residual(self, target_pts):
00088 h_mat = self.compute_expected(target_pts)
00089 z_mat = self.get_measurement()
00090 assert(h_mat.shape == z_mat.shape)
00091 assert(h_mat.shape[0] == 4)
00092 r_mat = h_mat[0:3,:] - z_mat[0:3,:]
00093 r = array(reshape(r_mat.T, [-1,1]))[:,0]
00094 return r
00095
00096 def compute_residual_scaled(self, target_pts):
00097 r = self.compute_residual(target_pts)
00098 gamma_sqrt = self.compute_marginal_gamma_sqrt(target_pts)
00099 r_scaled = gamma_sqrt * matrix(r).T
00100 return array(r_scaled.T)[0]
00101
00102 def compute_marginal_gamma_sqrt(self, target_pts):
00103 import scipy.linalg
00104
00105 cov = self.compute_cov(target_pts)
00106 gamma = matrix(zeros(cov.shape))
00107 num_pts = self.get_residual_length()/3
00108
00109 for k in range(num_pts):
00110
00111 first = 3*k
00112 last = 3*k+3
00113 sub_cov = matrix(cov[first:last, first:last])
00114 sub_gamma_sqrt_full = matrix(scipy.linalg.sqrtm(sub_cov.I))
00115 sub_gamma_sqrt = real(sub_gamma_sqrt_full)
00116 assert(scipy.linalg.norm(sub_gamma_sqrt_full - sub_gamma_sqrt) < 1e-6)
00117 gamma[first:last, first:last] = sub_gamma_sqrt
00118 return gamma
00119
00120 def compute_cov(self, target_pts):
00121 epsilon = 1e-8
00122
00123 num_joints = len(self._M_chain.chain_state.position)
00124 Jt = zeros([num_joints, self.get_residual_length()])
00125
00126 x = JointState()
00127 x.position = self._M_chain.chain_state.position[:]
00128
00129 f0 = reshape(array(self._calc_fk_target_pts(x)[0:3,:].T), [-1])
00130 for i in range(num_joints):
00131 x.position = list(self._M_chain.chain_state.position[:])
00132 x.position[i] += epsilon
00133 fTest = reshape(array(self._calc_fk_target_pts(x)[0:3,:].T), [-1])
00134 Jt[i] = (fTest - f0)/epsilon
00135 cov_angles = [x*x for x in self._full_chain.calc_block._chain._cov_dict['joint_angles']]
00136 cov = matrix(Jt).T * matrix(diag(cov_angles)) * matrix(Jt)
00137
00138 if ( self._full_chain.calc_block._chain._cov_dict.has_key('translation') ):
00139 translation_var = self._full_chain.calc_block._chain._cov_dict['translation'];
00140 translation_cov = numpy.diag(translation_var*(self.get_residual_length()/3))
00141 cov = cov + translation_cov
00142
00143 return cov
00144
00145 def get_residual_length(self):
00146 pts = self._checkerboard.generate_points()
00147 N = pts.shape[1]
00148 return N*3
00149
00150 def get_measurement(self):
00151 '''
00152 Returns a 4xN matrix with the locations of the checkerboard points in homogenous coords,
00153 as per the forward kinematics of the chain
00154 '''
00155 return self._calc_fk_target_pts(self._M_chain.chain_state)
00156
00157 def _calc_fk_target_pts(self, chain_state):
00158
00159 target_pts_tip = self._checkerboard.generate_points()
00160
00161
00162 target_pose_root = self._full_chain.calc_block.fk(chain_state)
00163
00164
00165 target_pts_root = target_pose_root * target_pts_tip
00166
00167 return target_pts_root
00168
00169 def compute_expected(self, target_pts):
00170 return target_pts
00171
00172
00173 def build_sparsity_dict(self):
00174 sparsity = self._full_chain.build_sparsity_dict()
00175 sparsity['checkerboards'] = {}
00176 sparsity['checkerboards'][self._target_id] = { 'spacing_x': 1, 'spacing_y': 1 }
00177 return sparsity
00178