示例#1
0
def makeLineLocalizer(numObservations, numStates, ideal, xMin, xMax, robotY):
    #!    pass
    """
    Create behavior controlling robot to move in a line and to
                      localize itself in one dimension

    @param numObservations: number of discrete observations into which to
    divide the range of good sonar observations, between 0 and C{goodSonarRange}
    @param numStates: number of discrete states into which to divide
    the range of x coordinates
    @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 robotY: constant y coordinate for center of robot

    @returns: an instance of {\tt sm.SM} that implements the behavior
    """

    # Size of a state in meters
    stateWidthMeters = (xMax - xMin) / float(numStates)

    preprocessor = PreProcess(numObservations, stateWidthMeters)
    navModel = makeRobotNavModel(ideal, xMin, xMax, numStates, numObservations)
    estimator = seGraphics.StateEstimator(navModel)
    driver = move.MoveToFixedPose(util.Pose(xMax, robotY, 0.0), maxVel=0.5)

    return sm.Cascade(sm.Parallel(sm.Cascade(preprocessor, estimator), driver),
                      sm.Select(1))
示例#2
0
def wrapTextUI(m):
    """
    @param m: An instance of C{sm.SM}
    @returns: A composite machine that prompts the user for input to, and
    prints the output of C{m} on each step.
    """
    return sm.Cascade(sm.Cascade(TextInputSM(), m),
                      sm.PureFunction(textOutput))
示例#3
0
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))
示例#4
0
def systemSM(circuit, inComponents, motorNodes, initState):

    # circuit takes input which is a dictionary of values (or list of them)
    # output is a le.Solution instance.
    cSM = CircuitSM(circuit.components, inComponents, circuit.groundNode)
    # motor input is Solution instance, output is (vel, angle)
    motorSM = sm.Cascade(MotorAccel(initState, motorNodes), sm.R(initState))
    # world has Solution as input, ((vel, angle), sol) as output
    wSM = sm.Parallel(motorSM, sm.Wire())

    if inComponents:
        # there's another input besides motor feedback
        return Feedback2(sm.Cascade(cSM,wSM), MotorFeedback())
    else:
        # only motor feedback
        return Feedback(sm.Cascade(cSM,wSM), MotorFeedback())
示例#5
0
def makeStateEstimationSimulation(worldSM, verbose = False):
    """
    Make a machine that simulates the state estimation process.  It
    takes a state machine representing the world, at construction
    time.  Let i be an input to the world machine.  The input is fed
    into the world machine, generating (stochastically) an output, o.
    The (o, i) pair is fed into a state-estimator using worldSM as its
    model.  The output of the state estimator is a belief state, b.
    The output of this entire composite machine is (b, (o, i)).

    @param worldSM: an instance of C{ssm.StochasticSM}
    @returns: a state machine that simulates the world and executes
    the state estimation process.
    """
    return sm.Cascade(sm.Parallel(worldSM, sm.Wire()),
                      sm.Parallel(StateEstimator(worldSM, verbose = verbose),
                                  sm.Wire()))
示例#6
0
 def spireFlow(incr):
     return sm.Cascade(SpyroGyra(incr), XYDriver())