def relative_time_streams(fname, lr, freq, single_camera):
"""
return start-end time, number of time samples, and streams for each sensor
"""
c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs = load_data(fname, lr, single_camera)
dt =1./freq
if c2_ts.any():
tmin = min(np.min(c1_ts), np.min(c2_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(c2_ts), np.max(hy_ts))
else:
tmin = min(np.min(c1_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(hy_ts))
## calculate the number of time-steps for the kalman filter.
nsteps = int(math.ceil((tmax-tmin)/dt))
## get rid of absolute time, put the three data-streams on the same time-scale
c1_ts -= tmin
c2_ts -= tmin
hy_ts -= tmin
ar1_strm = streamize(c1_tfs, c1_ts, freq, avg_transform)
ar2_strm = streamize(c2_tfs, c2_ts, freq, avg_transform)
hy_strm = streamize(hy_tfs, hy_ts, freq, avg_transform)
return tmin, tmax, nsteps, ar1_strm, ar2_strm, hy_strm
def setup_kalman(fname, lr, freq, single_camera):
c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs = load_data(fname, lr, single_camera)
motion_var, ar_var, hydra_var, hydra_vvar = load_covariances()
dt = 1./freq
if c2_ts.any():
tmin = min(np.min(c1_ts), np.min(c2_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(c2_ts), np.max(hy_ts))
else:
tmin = min(np.min(c1_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(hy_ts))
## calculate the number of time-steps for the kalman filter.
nsteps = int(math.ceil((tmax-tmin)/dt))
## get rid of absolute time, put the three data-streams on the same time-scale
c1_ts -= tmin
c2_ts -= tmin
hy_ts -= tmin
# initialize KF:
x0, S0 = get_first_state(dt, c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs)
KF = kalman()
## ===> assumes that the variance for ar1 and ar2 are the same!!!
KF.init_filter(-dt, x0, S0, motion_var, hydra_var, hydra_vvar, ar_var, ar_var)
ar1_strm = streamize(c1_tfs, c1_ts, freq, avg_transform)
ar2_strm = streamize(c2_tfs, c2_ts, freq, avg_transform)
hy_strm = streamize(hy_tfs, hy_ts, freq, avg_transform)
return (KF, nsteps, tmin, ar1_strm, ar2_strm, hy_strm)
def setup_kalman(fname, freq=30.):
c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs = load_data(fname)
motion_var, ar_var, hydra_var = load_covariances()
dt = 1/freq
tmin = min(np.min(c1_ts), np.min(c2_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(c2_ts), np.max(hy_ts))
## calculate the number of time-steps for the kalman filter.
nsteps = int(math.ceil((tmax-tmin)/dt))
## get rid of absolute time, put the three data-streams on the same time-scale
c1_ts -= tmin
c2_ts -= tmin
hy_ts -= tmin
# initialize KF:
x0, S0 = get_first_state(dt, c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs)
KF = kalman()
## ===> assumes that the variance for ar1 and ar2 are the same!!!
hydra_var_vel = np.zeros((6, 6))
hydra_var_vel[3:6, 3:6] = hydra_var[3:6, 3:6]
hydra_var_vel[0:3, 0:3] = 25e-8 * np.eye(3)
KF.init_filter(0,x0, S0, motion_var, hydra_var_vel, ar_var, ar_var)
ar1_strm = streamize(c1_tfs, c1_ts, freq, avg_transform)
ar2_strm = streamize(c2_tfs, c2_ts, freq, avg_transform)
hy_strm = streamize(hy_tfs, hy_ts, freq, avg_transform)
return (KF, nsteps, tmin, ar1_strm, ar2_strm, hy_strm)
def load_data(data_file, lr, freq=30.0):
with open(data_file, 'r') as f:
dat = cp.load(f)
demo_dir = osp.dirname(data_file)
cam_types = get_cam_types(demo_dir)
T_cam2hbase = dat['T_cam2hbase']
cam_tmin = float('inf')
cam_info = {}
for kname in dat[lr].keys():
if 'cam' in kname:
if kname != 'camera1': continue
tfs = [tt[0] for tt in dat[lr][kname]]
ts = [tt[1] for tt in dat[lr][kname]]
#ctype_name = int(kname[-1])
## don't append any empty-streams:
if len(ts) > 0:
cam_strm = streamize(tfs, ts, freq, avg_transform)#, tstart=-1./freq)
cam_info[kname] = {'type' : cam_types[kname],
'stream' : cam_strm}
cam_tmin = min(ts[0], cam_tmin)
## hydra data:
hydra_tfs = [tt[0] for tt in dat[lr]['hydra']]
hydra_ts = np.array([tt[1] for tt in dat[lr]['hydra']])
if abs(cam_tmin - hydra_ts[0]) > 1000:
hydra_ts += (cam_tmin - hydra_ts[0])
if len(hydra_ts) <= 0:
redprint("ERROR : No hydra data found in : %s"%(osp.basename(data_file)))
sys.exit(-1)
hydra_strm = streamize(hydra_tfs, hydra_ts, freq, avg_transform)#, tstart=-1./freq)
## potentiometer angles:
pot_vals = np.array([tt[0] for tt in dat[lr]['pot_angles']])
pot_ts = np.array([tt[1] for tt in dat[lr]['pot_angles']])
print cam_tmin, pot_ts[0]
if abs(cam_tmin - pot_ts[0]) > 1000:
pot_ts += (cam_tmin - pot_ts[0])
if len(pot_ts) <= 0:
redprint("ERROR : No potentiometer data found in : %s"%(osp.basename(data_file)))
sys.exit(-1)
pot_strm = streamize(pot_vals, pot_ts, freq, np.mean)#, tstart=-1./freq)
print hydra_ts[0], pot_ts[0]
return (T_cam2hbase, cam_info, hydra_strm, pot_strm)
def load_data(data_file, lr, freq=30.0, speed=1.0, hydra_only=False):
"""
Changed slightly from the one in demo_data_prep to include speed.
"""
with open(data_file, 'r') as f:
dat = cp.load(f)
demo_dir = osp.dirname(data_file)
with open(osp.join(demo_dir, demo_names.camera_types_name),'r') as fh: cam_types = yaml.load(fh)
T_cam2hbase = dat['T_cam2hbase']
if not hydra_only:
cam_info = {}
for kname in dat[lr].keys():
if 'cam' in kname:
tfs = [tt[0] for tt in dat[lr][kname]]
ts = [tt[1] for tt in dat[lr][kname]]
ctype_name = int(kname[-1])
## we will add empty stream
cam_strm = streamize(tfs, ts, freq, avg_transform, speed=speed)#, tstart=-1./freq)
cam_info[ctype_name] = {'type' : cam_types[ctype_name],
'stream' : cam_strm}
## hydra data:
hydra_tfs = [tt[0] for tt in dat[lr]['hydra']]
hydra_ts = np.array([tt[1] for tt in dat[lr]['hydra']])
if len(hydra_ts) <= 0:
redprint("ERROR : No hydra data found in : %s"%(osp.basename(data_file)))
sys.exit(-1)
hydra_strm = streamize(hydra_tfs, hydra_ts, freq, avg_transform, speed=speed)#, tstart=-1./freq)
## potentiometer angles:
pot_vals = np.array([tt[0] for tt in dat[lr]['pot_angles']])
pot_ts = np.array([tt[1] for tt in dat[lr]['pot_angles']])
if len(pot_ts) <= 0:
redprint("ERROR : No potentiometer data found in : %s"%(osp.basename(data_file)))
sys.exit(-1)
pot_strm = streamize(pot_vals, pot_ts, freq, np.mean, speed=speed)#, tstart=-1./freq)
if not hydra_only:
return (T_cam2hbase, cam_info, hydra_strm, pot_strm)
else:
return (T_cam2hbase, hydra_strm, pot_strm)
def relative_time_streams(fname, freq):
c1_ts, c1_tfs, c2_ts, c2_tfs, hy_ts, hy_tfs = load_data(fname)
dt =1/freq
tmin = min(np.min(c1_ts), np.min(c2_ts), np.min(hy_ts))
tmax = max(np.max(c1_ts), np.max(c2_ts), np.max(hy_ts))
## calculate the number of time-steps for the kalman filter.
nsteps = int(math.ceil((tmax-tmin)/dt))
## get rid of absolute time, put the three data-streams on the same time-scale
c1_ts -= tmin
c2_ts -= tmin
hy_ts -= tmin
ar1_strm = streamize(c1_tfs, c1_ts, freq, avg_transform)
ar2_strm = streamize(c2_tfs, c2_ts, freq, avg_transform)
hy_strm = streamize(hy_tfs, hy_ts, freq, avg_transform)
return tmin, tmax, nsteps, ar1_strm, ar2_strm, hy_strm
def reject_outliers_tf_stream(strm, n_window=10, v_th=[0.35,0.35,0.25]):
"""
Rejects transforms from a stream of TF based on
outliers in the x,y or z coordinates.
Rejects based on threshold on the median velocity computed over a window.
STRM : The stream of transforms to filter
N_WINDOW: The number of neighbors of a data-point to consider
(note the neighbors can be far away in time, if the stream is sparse)
V_TH : The threshold velocity. A data-point is considered an outlier, if v > vth
V_TH is specified for x,y,z
Note: N_WINDOW/2 number of data-points at the beginning and
the end of the stream are not filtered.
"""
s_window = int(n_window/2)
tfs, ts = strm.get_data()
N = len(tfs)
Xs = np.empty((N,3))
for i in xrange(N):
Xs[i,:] = tfs[i][0:3,3]
n = N-2*s_window
## when stream is very short, just return the original stream
if n <= 0:
return strm
v = np.empty((n, 3, 2*s_window))
X0 = Xs[s_window:s_window+n,:]
t0 = ts[s_window:s_window+n]
shifts = np.arange(-s_window, s_window+1)
shifts = shifts[np.nonzero(shifts)]
for i,di in enumerate(shifts):
dx = Xs[s_window+di:s_window+n+di,:] - X0
dt = ts[s_window+di: s_window+n+di] - t0
v[:,:,i] = np.abs(dx/dt[:,None])
v_med = np.median(v,axis=2)
inliers = s_window + np.arange(n)[np.all(v_med <= v_th, axis=1)]
inliers = np.r_[np.arange(s_window), inliers, np.arange(n+s_window,N)]
tfs_in = []
for i in inliers:
tfs_in.append(tfs[i])
ts_in = ts[inliers]
return streamize(tfs_in, ts_in, 1./strm.dt, strm.favg, strm.tstart)
def fit_spline_to_tf_stream(strm, new_freq, deg=3):
"""
Interpolates a stream of transforms using splines, such that
there is no "NONE" when sampling the stream at NEW_FREQ frequency.
Returns a stream of transforms.
"""
tfs, ts = strm.get_data()
tstart = strm.get_start_time()
tmax = ts[-1]
ndt = 1./new_freq
new_ts = np.arange(tstart, tmax+ndt/4., ndt/2.)
## get data in xyzrpy format (6xn) matrix:
N = len(tfs)
tf_dat = np.empty((6, N))
tf_dat[0:3,0] = tfs[0][0:3,3]
tf_dat[3:6,0] = tfms.euler_from_matrix(tfs[0])
for i in xrange(1,N):
now_tf = tfs[i]
tf_dat[0:3,i] = now_tf[0:3,3]
prev_rpy = tf_dat[3:6,i-1]
now_rpy = tfms.euler_from_matrix(now_tf)
now_rpy = closer_angle(now_rpy, prev_rpy)
tf_dat[3:6,i] = now_rpy
blueprint("\t fitting spline to data (scipy) ..")
s = N*.001**2
(tck, _) = si.splprep(tf_dat, s=s, u=ts, k=deg)
blueprint("\t evaluating spline at new time-stamps ..")
interp_xyzrpys = np.r_[si.splev(new_ts, tck)].T
smooth_tfms = []
for xyzrpy in interp_xyzrpys:
tfm = tfms.euler_matrix(*xyzrpy[3:6])
tfm[0:3,3] = xyzrpy[0:3]
smooth_tfms.append(tfm)
return streamize(smooth_tfms, new_ts, new_freq, strm.favg, tstart)
def view_tracking_on_rviz(demo_type, demo_name, tps_model_fname, freq=30.0, speed=1.0, use_smoother=True, prompt=False, verbose=False):
"""
Visualizes demo after kalman tracking/smoothing on rviz.
@demo_type, @demo_name: demo identification.
@freq: basically measure of fine-ness of timesteps.
@speed: how fast to replay demo.
@main: which sensor to display the marker for
@prompt: does the user hit enter after each time step?
"""
demo_dir = osp.join(demo_files_dir, demo_type, demo_name)
bag_file = osp.join(demo_dir, demo_names.bag_name)
traj_file = osp.join(demo_dir, demo_names.traj_name)
calib_file = osp.join(demo_dir, demo_names.calib_name)
with open(osp.join(demo_dir, demo_names.camera_types_name),'r') as fh: cam_types = yaml.load(fh)
if not osp.isfile(traj_file):
yellowprint("%s does not exist for this demo. Running kalman filter/smoother now with default args."%demo_names.traj_name)
data_file = osp.join(demo_dir, demo_names.data_name)
if not osp.isfile(data_file):
yellowprint("%s does not exist for this demo. Extracting now."%demo_names.data_name)
ed.save_observations_rgbd(demo_type, demo_name)
filter_traj(demo_dir, tps_model_fname=tps_model_fname, save_tps=True, do_smooth=True, plot='', block=False)
with open(traj_file, 'r') as fh: traj = cp.load(fh)
# get grippers used
grippers = traj.keys()
if rospy.get_name() == "/unnamed":
rospy.init_node("visualize_demo")
# data
rgbd_dirs = {cam:osp.join(demo_dir,demo_names.video_dir%cam) for cam in cam_types if cam_types[cam] == 'rgbd'}
pc_pubs = {cam:rospy.Publisher('/point_cloud%i'%cam, PointCloud2) for cam in rgbd_dirs}
cam_frames = {cam:'/camera%i_rgb_optical_frame'%cam for cam in rgbd_dirs}
cam_tfms = get_cam_transforms (calib_file, len(cam_types))
for cam in rgbd_dirs:
if cam != 1:
publish_static_tfm(cam_frames[1], cam_frames[cam], cam_tfms[cam])
# Remove segment "done", it is just a single frame
segs = sorted(traj[grippers[0]].keys())
if 'done' in segs: segs.remove('done')
sleeper = rospy.Rate(freq)
T_far = np.eye(4)
T_far[0:3,3] = [10,10,10]
for seg in segs:
if prompt:
raw_input("Press enter for segment %s."%seg)
else:
yellowprint("Segment %s beginning."%seg)
time.sleep(1)
# Initializing data streams:
traj_strms = {}
pot_strms = {}
tfms_key = 'tfms_s' if use_smoother else 'tfms'
for lr in grippers:
traj_strms[lr] = streamize(traj[lr][seg][tfms_key], traj[lr][seg]['stamps'], freq, avg_transform, speed=speed)
# HACK
pot_strms[lr] = streamize(traj[lr][seg]['pot_angles'][:len(traj[lr][seg]['stamps'])], traj[lr][seg]['stamps'], freq, np.mean, speed=speed)
tmin, tmax, nsteps = relative_time_streams(traj_strms.values() + pot_strms.values(), freq, speed)
pc_strms = {cam:streamize_rgbd_pc(rgbd_dirs[cam], cam_frames[cam], freq, tstart=tmin, tend=tmax,speed=speed,verbose=verbose) for cam in rgbd_dirs}
prev_ang = {'l': 0, 'r': 0}
dat_snext = {lr:{} for lr in grippers}
for lr in grippers:
dat_snext[lr]['traj'] = stream_soft_next(traj_strms[lr])
dat_snext[lr]['pot'] = stream_soft_next(pot_strms[lr])
for i in xrange(nsteps):
if prompt:
raw_input("Hit enter when ready.")
if verbose:
print "Time stamp: ", tmin+(0.0+i*speed)/freq
## show the point-cloud:
found_pc = False
for cam in pc_strms:
try:
pc = pc_strms[cam].next()
if pc is not None:
if verbose:
print "pc%i ts:"%cam, pc.header.stamp.to_sec()
pc.header.stamp = rospy.Time.now()
pc_pubs[cam].publish(pc)
found_pc = True
else:
if verbose:
print "pc%i ts:"%cam,None
except StopIteration:
pass
tfms = []
ang_vals = []
for lr in grippers:
if lr == 'r':
continue
tfm = dat_snext[lr]['traj']()
ang_val = dat_snext[lr]['pot']()
if ang_val != None and not np.isnan(ang_val):
prev_ang[lr] = ang_val
ang_val = ang_val
else:
ang_val = prev_ang[lr]
ang_val *= 2
if tfm is None:
tfms.append(T_far)
else:
tfms.append(tfm)
ang_vals.append(rad_angle(ang_val))
handles = draw_trajectory(cam_frames[cam], tfms, color=(1,0,0,1), open_fracs=ang_vals)
time.sleep(1.0/freq)
def traj_kalman_lr(data_file, calib_file, lr, freq, use_spline, customized_shift, single_camera):
'''
input: data_file
lr: 'l' or 'r'
freq
use_spline
customized_shift: custimized shift between smoother and filter: to compensate for the lag of smoother
return trajectory data
'''
_, _, _, ar1_strm, ar2_strm, hy_strm = relative_time_streams(data_file, lr, freq, single_camera)
## run kalman filter:
nsteps, tmin, F_means, S, A, R = run_kalman_filter(data_file, lr, freq, use_spline, True, single_camera)
## run kalman smoother:
S_means, _ = smoother(A, R, F_means, S)
X_kf = np.array(F_means)
X_kf = np.reshape(X_kf, (X_kf.shape[0], X_kf.shape[1])).T
X_ks = np.array(S_means)
X_ks = np.reshape(X_ks, (X_ks.shape[0], X_ks.shape[1])).T
# Shifting
if customized_shift != None:
shift = customized_shift
else:
shift = correlation_shift(X_kf, X_ks)
X_ks = np.roll(X_ks,shift,axis=1)
X_ks[:,:shift] = X_ks[:,shift][:,None]
T_filter = state_to_hmat(list(X_ks.T))
T_smoother = state_to_hmat(list(X_kf.T))
## load the potentiometer-angle stream:
pot_data = cp.load(open(data_file))[lr]['pot_angles']
ang_ts = np.array([tt[1] for tt in pot_data]) ## time-stamps
ang_vals = [open_frac(tt[0]) for tt in pot_data] ## angles
ang_strm = streamize(ang_vals, ang_ts, freq, lambda x : x[-1], tmin)
ang_strm_vals = []
prev_ang = 0
for i in xrange(nsteps):
ang_val = soft_next(ang_strm)
if ang_val != None:
prev_ang = ang_val
ang_val = [ang_val]
else:
ang_val = [prev_ang]
ang_strm_vals.append(ang_val)
stamps = []
for i in xrange(nsteps):
stamps.append(tmin + i * 1.0 / freq)
traj = {"tfms": T_filter, "tfms_s": T_smoother, "pot_angles": ang_strm_vals, "stamps": stamps}
return traj
def rviz_kalman(demo_dir, bag_file, data_file, calib_file, freq, use_rgbd, use_smoother, use_spline, customized_shift, single_camera):
'''
For rgbd, data_file and bag_file are redundant
Otherwise, demo_dir is redundant
'''
if use_rgbd:
bag_file = osp.join(demo_dir, 'demo.bag')
rgbd1_dir = osp.join(demo_dir, 'camera_#1')
rgbd2_dir = osp.join(demo_dir, 'camera_#2')
data_file = osp.join(demo_dir, 'demo.data')
bag = rosbag.Bag(bag_file)
else:
bag = rosbag.Bag(bag_file)
dat = cp.load(open(data_file))
grippers = dat.keys()
pub = rospy.Publisher('/point_cloud1', PointCloud2)
pub2= rospy.Publisher('/point_cloud2', PointCloud2)
c1_tfm_pub = {}
c2_tfm_pub = {}
hydra_tfm_pub = {}
T_filt = {}
ang_strm = {}
ar1_strm = {}
ar2_strm = {}
hy_strm = {}
smooth_hy = {}
## publishers for unfiltered-data:
for lr in grippers:
c1_tfm_pub[lr] = rospy.Publisher('/%s_ar1_estimate'%(lr), PoseStamped)
c2_tfm_pub[lr] = rospy.Publisher('/%s_ar2_estimate'%(lr), PoseStamped)
hydra_tfm_pub[lr] = rospy.Publisher('/%s_hydra_estimate'%(lr), PoseStamped)
_, _, _, ar1_strm[lr], ar2_strm[lr], hy_strm = relative_time_streams(data_file, lr, freq, single_camera)
## run the kalman filter:
nsteps, tmin, F_means, S,A,R = run_kalman_filter(data_file, lr, freq, use_spline, True, single_camera)
## run the kalman smoother:
S_means, _ = smoother(A, R, F_means, S)
X_kf = np.array(F_means)
X_kf = np.reshape(X_kf, (X_kf.shape[0], X_kf.shape[1])).T
X_ks = np.array(S_means)
X_ks = np.reshape(X_ks, (X_ks.shape[0], X_ks.shape[1])).T
# Shifting between filter and smoother:
if customized_shift != None:
shift = customized_shift
else:
shift = correlation_shift(X_kf, X_ks)
X_ks = np.roll(X_ks,shift,axis=1)
X_ks[:,:shift] = X_ks[:,shift][:,None]
if use_smoother:
T_filt[lr] = state_to_hmat(list(X_ks.T))
else:
T_filt[lr] = state_to_hmat(list(X_kf.T))
## load the potentiometer-angle stream:
pot_data = cp.load(open(data_file))[lr]['pot_angles']
ang_ts = np.array([tt[1] for tt in pot_data]) ## time-stamps
ang_vals = [tt[0] for tt in pot_data] ## angles
# plt.plot(ang_vals)
# plt.show()
ang_vals = [0*open_frac(x) for x in ang_vals]
ang_strm[lr] = streamize(ang_vals, ang_ts, freq, lambda x : x[-1], tmin)
if use_spline:
smooth_hy[lr] = (t for t in fit_spline_to_stream(hy_strm, nsteps))
else:
smooth_hy[lr] = hy_strm
## get the point-cloud stream
cam1_frame_id = '/camera1_rgb_optical_frame'
cam2_frame_id = '/camera2_rgb_optical_frame'
if use_rgbd:
pc1_strm = streamize_rgbd_pc(rgbd1_dir, cam1_frame_id, freq, tmin)
if single_camera:
pc2_strm = streamize_rgbd_pc(None, cam2_frame_id, freq, tmin)
else:
pc2_strm = streamize_rgbd_pc(rgbd2_dir, cam2_frame_id, freq, tmin)
else:
pc1_strm = streamize_pc(bag, '/camera1/depth_registered/points', freq, tmin)
if single_camera:
pc2_strm = streamize_pc(bag, None, freq, tmin)
else:
pc2_strm = streamize_pc(bag, '/camera2/depth_registered/points', freq, tmin)
## get the relative-transforms between the cameras:
cam_tfm = get_cam_transform(calib_file)
publish_static_tfm(cam1_frame_id, cam2_frame_id, cam_tfm)
## frame of the filter estimate:
sleeper = rospy.Rate(freq)
T_far = np.eye(4)
T_far[0:3,3] = [10,10,10]
handles = []
prev_ang = {'l': 0, 'r': 0}
for i in xrange(nsteps):
#raw_input("Hit next when ready.")
print "Kalman ts: ", tmin+(0.0+i)/freq
## show the point-cloud:
found_pc = False
try:
pc = pc1_strm.next()
if pc is not None:
print "pc1 ts:", pc.header.stamp.to_sec()
pc.header.stamp = rospy.Time.now()
pub.publish(pc)
found_pc = True
else:
print "pc1 ts:",None
except StopIteration:
print "pc1 ts: finished"
#print "no more point-clouds"
pass
try:
pc2 = pc2_strm.next()
if pc2 is not None:
#print "pc2 not none"
print "pc2 ts:", pc2.header.stamp.to_sec()
pc2.header.stamp = rospy.Time.now()
pub2.publish(pc2)
found_pc = True
else:
print "pc2 ts:", None
except StopIteration:
print "pc2 ts: finished"
pass
ang_vals = []
T_filt_lr = []
for lr in grippers:
ang_val = soft_next(ang_strm[lr])
if ang_val != None:
prev_ang[lr] = ang_val
ang_val = ang_val
else:
ang_val = prev_ang[lr]
ang_vals.append(ang_val)
T_filt_lr.append(T_filt[lr][i])
handles = draw_trajectory(cam1_frame_id, T_filt_lr, color=(1,1,0,1), open_fracs=ang_vals)
# draw un-filtered estimates:
for lr in grippers:
ar1_est = soft_next(ar1_strm[lr])
if ar1_est != None:
c1_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(ar1_est), cam1_frame_id))
else:
c1_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(T_far), cam1_frame_id))
ar2_est = soft_next(ar2_strm[lr])
if ar2_est != None:
c2_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(ar2_est), cam1_frame_id))
else:
c2_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(T_far), cam1_frame_id))
hy_est = soft_next(smooth_hy[lr])
if hy_est != None:
hydra_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(hy_est), cam1_frame_id))
else:
hydra_tfm_pub[lr].publish(pose_to_stamped_pose(hmat_to_pose(T_far), cam1_frame_id))
sleeper.sleep()
def run_KF(KF, nsteps, freq, hydra_strm, cam_dat, hydra_covar, cam_covars, do_smooth, plot, plot_title, block):
"""
Runs the Kalman filter/smoother for NSTEPS using
{hydra/rgb/rgbd}_covar as the measurement covariances.
HYDRA_STRM : TF stream from hydra
CAM_DAT : Dictionary of camera streams
do_smooth: default true
plot: default true
"""
dt = 1./freq
## place holders for kalman filter's output:
xs_kf, covars_kf, ts_kf = [KF.x_filt],[KF.S_filt],[KF.t_filt]
hydra_snext = stream_soft_next(hydra_strm)
cam_strms = [cinfo['stream'] for cinfo in cam_dat.values()]
cam_names = cam_dat.keys()
cam_types = [cinfo['type'] for cinfo in cam_dat.values()]
cam_snext = [stream_soft_next(cstrm) for cstrm in cam_strms]
for i in xrange(nsteps):
KF.register_tf_observation(hydra_snext(), hydra_covar, do_control_update=True)
for i in xrange(len(cam_names)):
cam_covar = cam_covars[cam_names[i]]
KF.register_tf_observation(cam_snext[i](), cam_covar, do_control_update=False)
xs_kf.append(KF.x_filt)
covars_kf.append(KF.S_filt)
ts_kf.append(KF.t_filt)
hydra_strm.reset()
for cstrm in cam_strms:
cstrm.reset()
if do_smooth:
'''
# UNCOMMENT BELOW TO RUN KALMAN SMOOTHER:
A,_ = KF.get_motion_mats(dt)
R = get_smoother_motion_covariance(freq)
xs_smthr, covars_smthr = smoother(A, R, xs_kf, covars_kf)
'''
### do low-pass filtering for smoothing:
xs_kf_xyz = np.array(xs_kf)[:,0:3]
xs_kf_rpy = np.squeeze(np.array(xs_kf)[:,6:9])
xs_lp_xyz = low_pass(xs_kf_xyz, freq)
xs_lp_rpy = low_pass(unbreak(xs_kf_rpy), freq)
xs_smthr = np.c_[xs_lp_xyz, np.squeeze(np.array(xs_kf))[:,3:6], xs_lp_rpy, np.squeeze(np.array(xs_kf))[:,9:12]].tolist()
covars_smthr = None
if 's' in plot:
kf_strm = streamize(state_to_hmat(xs_kf), np.array(ts_kf), 1./hydra_strm.dt, hydra_strm.favg)
sm_strm = streamize(state_to_hmat(xs_smthr), np.array(ts_kf), 1./hydra_strm.dt, hydra_strm.favg)
plot_tf_streams([kf_strm, sm_strm, hydra_strm]+cam_strms, ['kf', 'smoother', 'hydra']+cam_dat.keys(), styles=['-','-','.']+['.' for i in xrange(len(cam_strms))], title=plot_title, block=block)
#plot_tf_streams([hydra_strm]+cam_strms, ['hydra']+cam_dat.keys(), block=block)
return (ts_kf, xs_kf, covars_kf, xs_smthr, covars_smthr)
else:
if 's' in plot:
kf_strm = streamize(state_to_hmat(xs_kf), np.array(ts_kf), 1./hydra_strm.dt, hydra_strm.favg)
plot_tf_streams([kf_strm, hydra_strm]+cam_strms, ['kf', 'hydra']+cam_dat.keys(), styles=['-','-','.']+['.' for i in xrange(len(cam_strms))], title=plot_title, block=block)
#plot_tf_streams([hydra_strm]+cam_strms, ['hydra']+cam_dat.keys(), block=block)
return (ts_kf, xs_kf, covars_kf, None, None)
def load_demo_data(demo_dir, freq, rem_outliers, tps_correct, tps_model_fname, plot, block):
cam_dat = {}
hydra_dat = {}
pot_dat = {}
dt = 1./freq
lr_full = {'l': 'left', 'r':'right'}
data_file = osp.join(demo_dir, demo_names.data_name)
T_cam2hbase, cam_dat['l'], hydra_dat['l'], pot_dat['l'] = load_data(data_file, 'l', freq)
_, cam_dat['r'], hydra_dat['r'], pot_dat['r'] = load_data(data_file, 'r', freq)
## collect all camera streams in a list:
all_cam_names = set()
for lr in 'lr':
for cam in cam_dat[lr].keys():
all_cam_names.add(cam)
## time-align all tf-streams (wrt their respective hydra-streams):
## NOTE THE STREAMS ARE MODIFIED IN PLACE (the time-stamps are changed)
## and also the tstart
blueprint("Getting rid of absolute time..")
all_cam_strms = []
for lr in 'lr':
for cam in cam_dat[lr].keys():
all_cam_strms.append(cam_dat[lr][cam]['stream'])
tmin, _, _ = relative_time_streams(hydra_dat.values() + pot_dat.values() + all_cam_strms, freq)
tshift1 = -tmin
## remove outliers in the camera streams
# plot command: o
if rem_outliers:
blueprint("Rejecting TF outliers in camera-data..")
for lr in 'lr':
for cam in cam_dat[lr].keys():
strm_in = reject_outliers_tf_stream(cam_dat[lr][cam]['stream'])
if 'o' in plot:
blueprint("\t Plotting outlier rejection..")
cam_name = cam+'_'+lr
plot_tf_streams([cam_dat[lr][cam]['stream'], strm_in], strm_labels=[cam_name, cam_name+'_in'], styles=['.','-'], block=block)
cam_dat[lr][cam]['stream'] = strm_in
## do time-alignment (with the respective l/r hydra-streams):
blueprint("Calculating TF stream time-shifts..")
time_shifts = {} ## dictionary : maps a camera-name to the time-shift wrt hydra
for cam in all_cam_names:
ndat = {'r':-1, 'l':-1} ## compare whether left/right has more data-points
for lr in 'lr':
if cam in cam_dat[lr].keys():
ndat[lr] = len(cam_dat[lr][cam]['stream'].get_data()[0])
lr_align = ndat.keys()[np.argmax(ndat.values())]
time_shifts[cam] = dt* align_tf_streams(hydra_dat[lr_align], cam_dat[lr_align][cam]['stream'])
greenprint("Time-shifts found : %s"%str(time_shifts))
## time-shift the streams:
# plot command: t
all_cam_strms = []
blueprint("Time-aligning TF streams..")
for lr in 'lr':
redprint("\t Alignment for : %s"%lr_full[lr])
aligned_cam_strms = []
for cam in cam_dat[lr].keys():
aligned_cam_strms.append( time_shift_stream(cam_dat[lr][cam]['stream'], time_shifts[cam]) )
if 't' in plot:
unaligned_cam_streams = []
for cam in cam_dat[lr].values():
unaligned_cam_streams.append(cam['stream'])
blueprint("\t plotting unaligned TF streams...")
plot_tf_streams(unaligned_cam_streams + [hydra_dat[lr]], cam_dat[lr].keys()+['hydra'], title='UNALIGNED CAMERA-streams (%s)'%lr_full[lr], block=block)
blueprint("\t plotting aligned TF streams...")
plot_tf_streams(aligned_cam_strms+[hydra_dat[lr]], cam_dat[lr].keys()+['hydra'], title='ALIGNED CAMERA-streams (%s)'%lr_full[lr], block=block)
for i,cam in enumerate(cam_dat[lr].keys()):
cam_dat[lr][cam]['stream'] = aligned_cam_strms[i]
all_cam_strms.append(cam_dat[lr][cam]['stream']) ###### SIBI'S CHANGE
## put the aligned-streams again on the same time-scale:
blueprint("Re-aligning the TF-streams after TF based time-shifts..")
tmin, tmax, nsteps = relative_time_streams(hydra_dat.values() + pot_dat.values() + all_cam_strms, freq)
tshift2 = -tmin
greenprint("TOTAL TIME-SHIFT : (%0.3f + %0.3f) = %0.3f"%(tshift1, tshift2, tshift1+tshift2))
## TPS-correct the hydra-data:
if tps_correct:
tps_models = {'l':None, 'r':None}
if tps_model_fname == '':
blueprint("\t Fitting TPS model to demo data..")
for lr in 'lr':
blueprint("\t TPS fitting for %s"%lr_full[lr])
if 'camera1' not in cam_dat[lr].keys():
redprint("\t\t camera1 not in the data for %s gripper -- Cannot do tps-fit. Skipping.."%lr_full[lr])
continue
_, hydra_tfs, cam1_tfs = get_corresponding_data(hydra_dat[lr], cam_dat[lr]['camera1']['stream'])
f_tps = fit_tps_on_tf_data(hydra_tfs, cam1_tfs, plot=False) ## <<< mayavi plotter has some issues with multiple insantiations..
T_cam2hbase_train = T_cam2hbase
tps_models[lr] = f_tps
tps_save_fname = osp.join(demo_dir, 'tps_models.cp')
with open(tps_save_fname, 'w') as f:
cp.dump([tps_models, T_cam2hbase], f)
else:
blueprint("\t Getting TPS-model from : %s"%tps_model_fname)
with open(tps_model_fname, 'r') as f:
tps_models, T_cam2hbase_train = cp.load(f)
blueprint("TPS-correcting hydra-data..")
# plot command: p
for lr in 'lr':
if tps_models[lr] != None:
hydra_tfs, hydra_ts = hydra_dat[lr].get_data()
hydra_tfs_aligned = correct_hydra(hydra_tfs, T_cam2hbase, tps_models[lr], T_cam2hbase_train)
hydra_strm_aligned = streamize(hydra_tfs_aligned, hydra_ts, 1./hydra_dat[lr].dt, hydra_dat[lr].favg, hydra_dat[lr].tstart)
if 'p' in plot:
if tps_model_fname!=None:
plot_tf_streams([hydra_dat[lr], hydra_strm_aligned], ['hydra-old', 'hydra-corr'], title='hydra-correction %s'%lr_full[lr], block=block)
else:
plot_tf_streams([hydra_dat[lr], hydra_strm_aligned, cam_dat[lr]['camera1']['stream']], ['hydra-old', 'hydra-corr', 'cam1'], title='hydra-correction', block=block)
hydra_dat[lr] = hydra_strm_aligned
## return the data:
filter_data = {'demo_dir': demo_dir,
'nsteps' : nsteps,
'tmin' : tmin,
'tmax' : tmax,
'pot_dat': pot_dat,
'hydra_dat': hydra_dat,
'cam_dat': cam_dat,
't_shift': tshift1+tshift2,
'cam_shifts': time_shifts}
return filter_data