def __init__(self, sflist, c_map, case):
'''
Constructor
'''
self.sflist = sflist
self.noise_model = [case.noise_turn, case.noise_velocity]
self.c_map = c_map
self.goal_graph = case.goal_graph
# Command algo
self.cmd_algo = CmdBiased(case, c_map)
# self.cmd_algo = CmdRandom(case, c_map)
self.deltick = case.deltick
# Initialize particle filters for each SensorFly dir
self.pfilters = {}
for sf in self.sflist:
pf = RobotPF(sf, case.num_particles, case.init_p_wt, \
self.noise_model, case.deltick)
self.pfilters[sf.name] = pf
sf.pf_estimated_xy = np.array(sf.xy, np.float32)
# record the actual position without pf correction
self.dr_estimated_pos = {}
for sf in sflist:
self.dr_estimated_pos[sf.name] = [sf.xy, sf.dir]
sf.dr_estimated_xy = np.array(sf.xy, np.float32)
# Iniitialize a landmark database
self.landmark_db = LandmarkDB(case)
class DrunkWalk(object):
'''
DrunkWalk deployment algorithm
'''
def __init__(self, sflist, c_map, case):
'''
Constructor
'''
self.sflist = sflist
self.noise_model = [case.noise_turn, case.noise_velocity]
self.c_map = c_map
self.goal_graph = case.goal_graph
# Command algo
self.cmd_algo = CmdBiased(case, c_map)
# self.cmd_algo = CmdRandom(case, c_map)
self.deltick = case.deltick
# Initialize particle filters for each SensorFly dir
self.pfilters = {}
for sf in self.sflist:
pf = RobotPF(sf, case.num_particles, case.init_p_wt, \
self.noise_model, case.deltick)
self.pfilters[sf.name] = pf
sf.pf_estimated_xy = np.array(sf.xy, np.float32)
# record the actual position without pf correction
self.dr_estimated_pos = {}
for sf in sflist:
self.dr_estimated_pos[sf.name] = [sf.xy, sf.dir]
sf.dr_estimated_xy = np.array(sf.xy, np.float32)
# Iniitialize a landmark database
self.landmark_db = LandmarkDB(case)
def command(self):
'''
Command as per the planning algorithm
'''
for sf in self.sflist:
cmd = self.cmd_algo.getCommand(sf)
if cmd:
sf.setMoveCommand(cmd)
def update(self):
'''
Update the estimates as per command
'''
for sf in self.sflist:
# Collision back-off strategy
if sf.has_collided:
if sf.is_backing_off:
sf.backoff_time_cnt = 0
else:
sf.backoff_time_cnt = sf.backoff_time_cnt + 2 if sf.backoff_time_cnt < 5 else 5
else:
if not sf.is_backing_off:
sf.backoff_time_cnt = sf.backoff_time_cnt - 1 if sf.backoff_time_cnt > 1 else 0
# Particle filter update
if not sf.has_collided:
pf = self.pfilters[sf.name]
# Predict
pf.predict(sf.command)
# Correct
pf.correct(self.landmark_db)
# Resample
pf.resample_fast()
# Record estimated location
sf.pf_estimated_xy = pf.getEstimate()
sf.pf_particles_xy = pf.particles_xy
# Dead reckoning update
if not sf.has_collided:
pos = self.dr_estimated_pos[sf.name]
# Get direction from magnetometer
if sf.has_turned:
# Add noise to turn
pos[1] = (pos[1] + sf.command.turn) % 360
# Compute new poses
# pos[1] = sf.mag_dir
# Update location on all ticks
pos[0][0] = pos[0][0] + (sf.command.velocity * self.deltick) * math.cos(math.radians(pos[1]))
pos[0][1] = pos[0][1] + (sf.command.velocity * self.deltick) * math.sin(math.radians(pos[1]))
self.dr_estimated_pos[sf.name] = pos
sf.dr_estimated_xy = np.array(pos[0], np.float32)