def makeSim(hallwayColors,
legalInputs,
obsNoise,
dynamics,
transNoise,
title='sim',
initialDist=None):
"""
Make an instance of the simulator with noisy motion and sensing models.
@param hallwayColors: A list of the colors of the rooms in the
hallway, from left to right.
@param legalInputs: A list of the possible action commands
@param obsNoise: conditional distribution specifying the
probability of observing a color given the actual color of the room
@param dynamics: function that takes the robot's current location,
action, and hallwaylength, and returns its nominal new location
@param transNoise: P(actualResultingLocation | nominalResultingLoc)
represented as a function from ideal location to the actual
location the robot will end up in
@param title: String specifying title for simulator window
"""
n = len(hallwayColors)
if not initialDist:
initialDist = dist.UniformDist(range(n))
worldSM = ssm.StochasticSM(initialDist,
makeTransitionModel(dynamics, transNoise, n),
makeObservationModel(hallwayColors, obsNoise))
return makeSESwithGUI(worldSM,
hallwayColors,
legalInputs,
verbose=True,
title=title,
initBelief=initialDist)
def transitionGivenState(s):
# A uniform distribution we mix in to handle teleportation
transUniform = dist.UniformDist(range(numStates))
return dist.MixtureDist(dist.triangleDist(\
util.clip(s+a, 0, numStates-1),
transDiscTriangleWidth,
0, numStates-1),
transUniform, 1 - teleportProb)
def hallSE(hallwayColors,
legalInputs,
obsNoise,
dynamics,
transNoise,
initialDist=None,
verbose=True):
n = len(hallwayColors)
if not initialDist:
initialDist = dist.UniformDist(range(n))
worldSM = ssm.StochasticSM(initialDist,
makeTransitionModel(dynamics, transNoise, n),
makeObservationModel(hallwayColors, obsNoise))
return se.StateEstimator(worldSM, verbose)
def wrapWindowUI(m,
worldColors,
legalInputs,
windowName='Belief',
initBelief=None):
"""
@param m: A machine created by applying
C{se.makeStateEstimationSimulation} to a hallway world, which
take movement commands as input and generates as output structures
of the form C{(b, (o, a))}, where C{b} is a belief state, C{a} is
the action command, and C{o} is the observable output generated by
the world.
@param worldColors: A list of the colors of the rooms in the
hallway, from left to right.
@returns: A composite machine that prompts the user for input to, and
graphically displays the output of C{m} on each step.
"""
def drawWorld(size):
y = 0
for x in range(size):
window.drawRect((x, y), (x + 1, y + 1), color=worldColors[x])
def processOutput(stuff):
(dist, (o, a)) = stuff
drawWorld(dim)
drawBelief(dist, window, dim)
return (o, a)
def processInput(stuff):
if stuff == 'quit':
print 'Taking action 0 before quitting'
return 0
else:
return int(stuff)
dim = len(worldColors)
if not initBelief:
initBelief = dist.UniformDist(range(dim))
ydim = 1
window = DrawingWindow(dim * 50, ydim * 50 + 10, -0.2, dim + 0.2, -0.2,
ydim + 0.2, windowName)
drawWorld(dim)
drawBelief(initBelief, window, dim)
return sm.Cascade(
sm.Cascade(
sm.Cascade(TextInputSM(legalInputs),
sm.PureFunction(processInput)), m),
sm.PureFunction(processOutput))
def makeRobotNavModel(ideal, xMin, xMax, numStates, numObservations):
#! startDistribution = None # redefine this
"""
Create a model of a robot navigating in a 1 dimensional world with
a single sonar.
@param ideal: list of ideal sonar readings
@param xMin: minimum x coordinate for center of robot
@param xMax: maximum x coordinate for center of robot
@param numStates: number of discrete states into which to divide
the range of x coordinates
@param numObservations: number of discrete observations into which to
divide the range of good sonar observations, between 0 and C{goodSonarRange}
@returns: an instance of {\tt ssm.StochasticSM} that describes the
dynamics of the world
"""
# make initial distribution over states
startDistribution = dist.UniformDist(range(numStates))
######################################################################
### Define observation model
######################################################################
# real width of triangle distribution, in meters
# obsTriangleWidth = sonarStDev * 3
obsTriangleWidth = .05
obsDiscTriangleWidth = max(2,
int(obsTriangleWidth * \
(numObservations / sonarDist.sonarMax)))
# Part of distribution common to all observations
obsBackground = dist.MixtureDist(dist.UniformDist(range(numObservations)),
dist.DeltaDist(numObservations - 1), 0.5)
#!
def observationModel(ix):
# ix is a discrete location of the robot
# return a distribution over observations in that state
#! pass
return dist.MixtureDist(
dist.triangleDist(ideal[ix], obsDiscTriangleWidth, 0,
numObservations), obsBackground, 0.9)
######################################################################
### Define transition model
######################################################################
# real width of triangle distribution, in meters
transTriangleWidth = odoStDev * 3
transDiscTriangleWidth = max(
2, int(transTriangleWidth * (numStates / (xMax - xMin))))
transDiscTriangleWidth = 2
teleportProb = 0
#!
def transitionModel(a):
# a is a discrete action
# returns a conditional probability distribution on the next state
# given the previous state
#! pass
def transitionGivenState(s):
# A uniform distribution we mix in to handle teleportation
transUniform = dist.UniformDist(range(numStates))
return dist.MixtureDist(dist.triangleDist(\
util.clip(s+a, 0, numStates-1),
transDiscTriangleWidth,
0, numStates-1),
transUniform, 1 - teleportProb)
return transitionGivenState
######################################################################
### Create and return SSM
######################################################################
#!
return ssm.StochasticSM(startDistribution, transitionModel,
observationModel)
