3.16. Robot SimulationΒΆ

In [1]:

from jyro.simulator import (Robot, Pioneer, Pioneer16Sonars,
                            PioneerFrontLightSensors, Camera,
                            VSimulator, DepthCamera)
import random
import numpy as np

In [2]:

robot = Pioneer("Pioneer", 3.5, 2, 0)
robot.addDevice(Pioneer16Sonars())
robot.addDevice(DepthCamera(4))
light_sensors = PioneerFrontLightSensors(3.0)
light_sensors.lightMode = 'ambient'
robot.addDevice(light_sensors)

Out[2]:
_images/Robot_Simulation_2_0.svg
In [3]:

def worldf(physics):
    physics.addBox(0, 0, 4, 4, fill="backgroundgreen", wallcolor="gray")
    physics.addLight(2, 0.75, 1.0) # increased brightness for new linear version of lights

In [4]:

sim = VSimulator(robot, worldf)

In [5]:

camera = robot.device["camera"]

In [6]:

image = camera.getImage()
image

Out[6]:
_images/Robot_Simulation_6_0.png 
In [7]:

image.size

Out[7]:

(60, 40)

In [8]:

data = camera.getData()
data.shape

Out[8]:

(40, 60, 3)

In [9]:

robot.move(0.50, 0.35)

In [10]:

sim.step()

In [11]:

robot = Pioneer("Pioneer", 3.5, 2, 0)
robot.addDevice(Pioneer16Sonars())
robot.addDevice(Camera())
light_sensors = PioneerFrontLightSensors(3.0)
light_sensors.lightMode = 'ambient'
robot.addDevice(light_sensors)

Out[11]:
_images/Robot_Simulation_11_0.svg
In [12]:

sim = VSimulator(robot, worldf)

In [13]:

camera = robot.device["camera"]
image = camera.getImage()
image

Out[13]:
_images/Robot_Simulation_13_0.png 
In [14]:

data = camera.getData()
data.shape

Out[14]:

(40, 60, 3)

In [15]:

def random_action():
    """Generate a random action from a limited set of possible settings"""
    possible = [-1.0, -0.5, 0.0, 0.5, 1.0]
    return [random.choice(possible), random.choice(possible)]

def get_senses(robot):
    light = robot["light"].getData()
    sonar = [v/3.0 for v in robot["sonar"].getData()]
    camera = robot["camera"].getData()
    return [light, sonar, camera]

In [16]:

senses = get_senses(robot)
list(map(len, senses))

Out[16]:

[2, 16, 40]

In [17]:

robot.history = []

def brain(robot):
    senses = get_senses(robot)
    net.propagate(senses)
    translate, rotate = random_action()
    #self.move(translate, rotate)
    robot.history.append(robot.getPose())
    robot.move(0.50, 0.35)

In [18]:

robot.brain = brain

In [19]:

from conx import Network, Layer, FlattenLayer, SGD, ImageLayer, Conv2DLayer
import numpy as np

Using Theano backend.
conx, version 3.5.4

In [20]:

net = Network("Robot Prediction Network")
net.add(Layer("light", 2))
net.add(Layer("sonar", 16))
net.add(ImageLayer("camera", (40,60), 3))
net.add(FlattenLayer("flatten"))
net.add(Conv2DLayer("conv", 16, (3,3)))
net.add(Layer("hidden", 50, activation="relu"))
net.add(Layer("output1", 2, activation="sigmoid"))
net.add(Layer("hidden2", 5, activation="sigmoid"))
net.add(Layer("hidden3", 10, activation="sigmoid", dropout=0.25))
net.add(Layer("hidden4", 10, activation="sigmoid"))
net.add(Layer("output2", 5, activation="sigmoid"))

In [21]:

net.connect("sonar", "hidden2")
net.connect("light", "hidden")
net.connect("camera", "conv")
net.connect("conv", "flatten")
net.connect("flatten", "hidden2")
net.connect("hidden", "hidden2")
net.connect("hidden2", "hidden3")
##net.connect("hidden2", "output2")
net.connect("hidden3", "output2")
net.connect("hidden3", "hidden4")
net.connect("hidden4", "output1")

In [22]:

net.compile(optimizer="adam", error="mse")
#net.config["hspace"] = 200

In [23]:

net

Out[23]:
Robot Prediction NetworkLayer: output1 (output) shape = (2,) Keras class = Dense activation = sigmoidoutput1Layer: output2 (output) shape = (5,) Keras class = Dense activation = sigmoidoutput2Weights from hidden4 to output1 output1/kernel has shape (10, 2) output1/bias has shape (2,)Layer: hidden4 (hidden) shape = (10,) Keras class = Dense activation = sigmoidhidden4Weights from hidden3 to output2 output2/kernel has shape (10, 5) output2/bias has shape (5,)Weights from hidden3 to output2 output2/kernel has shape (10, 5) output2/bias has shape (5,)Weights from hidden3 to hidden4 hidden4/kernel has shape (10, 10) hidden4/bias has shape (10,)Layer: hidden3 (hidden) shape = (10,) dropout = 0.25 Keras class = Dense activation = sigmoidhidden3Weights from hidden2 to hidden3 hidden3/kernel has shape (5, 10) hidden3/bias has shape (10,)Layer: hidden2 (hidden) shape = (5,) Keras class = Dense activation = sigmoidhidden2Weights from hidden to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from flatten to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: flatten (hidden) Keras class = FlattenflattenWeights from hidden to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: hidden (hidden) shape = (50,) Keras class = Dense activation = reluhiddenWeights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from conv to flattenLayer: conv (hidden) Keras class = Conv2Dconv160Weights from light to hidden hidden/kernel has shape (2, 50) hidden/bias has shape (50,)Layer: light (input) shape = (2,) Keras class = InputlightWeights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: sonar (input) shape = (16,) Keras class = InputsonarWeights from camera to conv conv/kernel has shape (3, 3, 3, 16) conv/bias has shape (16,)Layer: camera (input) shape = (40, 60, 3) Keras class = Inputcamera
In [24]:

matrix = net.propagate_to("conv", get_senses(robot))

In [25]:

net["conv"].feature = 6

In [26]:

net.propagate_to_features("conv", get_senses(robot), scale=3)

Out[26]:

Feature 0

Feature 1

Feature 2

Feature 3

Feature 4

Feature 5

Feature 6

Feature 7

Feature 8

Feature 9

Feature 10

Feature 11

Feature 12

Feature 13

Feature 14

Feature 15
In [27]:

net.dataset.add([[0] * 2, [0] * 16, data], [[0] * 2, [1] + ([0] * 4)])

In [28]:

net.dashboard()

In [29]:

net

Out[29]:
Robot Prediction NetworkLayer: output1 (output) shape = (2,) Keras class = Dense activation = sigmoidoutput1Layer: output2 (output) shape = (5,) Keras class = Dense activation = sigmoidoutput2Weights from hidden4 to output1 output1/kernel has shape (10, 2) output1/bias has shape (2,)Layer: hidden4 (hidden) shape = (10,) Keras class = Dense activation = sigmoidhidden4Weights from hidden3 to output2 output2/kernel has shape (10, 5) output2/bias has shape (5,)Weights from hidden3 to output2 output2/kernel has shape (10, 5) output2/bias has shape (5,)Weights from hidden3 to hidden4 hidden4/kernel has shape (10, 10) hidden4/bias has shape (10,)Layer: hidden3 (hidden) shape = (10,) dropout = 0.25 Keras class = Dense activation = sigmoidhidden3Weights from hidden2 to hidden3 hidden3/kernel has shape (5, 10) hidden3/bias has shape (10,)Layer: hidden2 (hidden) shape = (5,) Keras class = Dense activation = sigmoidhidden2Weights from hidden to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from flatten to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: flatten (hidden) Keras class = FlattenflattenWeights from hidden to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: hidden (hidden) shape = (50,) Keras class = Dense activation = reluhiddenWeights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Weights from conv to flattenLayer: conv (hidden) Keras class = Conv2Dconv166Weights from light to hidden hidden/kernel has shape (2, 50) hidden/bias has shape (50,)Layer: light (input) shape = (2,) Keras class = InputlightWeights from sonar to hidden2 hidden2/kernel has shape (35330, 5) hidden2/bias has shape (5,)Layer: sonar (input) shape = (16,) Keras class = InputsonarWeights from camera to conv conv/kernel has shape (3, 3, 3, 16) conv/bias has shape (16,)Layer: camera (input) shape = (40, 60, 3) Keras class = Inputcamera
In [30]:

net.test()

========================================================
Testing validation dataset with tolerance 0.1...
Total count: 1
      correct: 0
      incorrect: 1
Total percentage correct: 0.0

In [31]:

if net.saved():
    net.load()
    net.plot_results()
else:
    net.train(epochs=100, plot=True)
    net.save()
_images/Robot_Simulation_31_0.png 
In [32]:

net.plot("all")
_images/Robot_Simulation_32_0.png 
In [33]:

net.test(show=True)

========================================================
Testing validation dataset with tolerance 0.1...
# | inputs | targets | outputs | result
---------------------------------------
0 | [[0.00,0.00],[0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00,0.00, 0.00],[[[0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90],  ...,  [0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90]], [[0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90],  ...,  [0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90]], [[0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90],  ...,  [0.68,0.85,0.90],  [0.68,0.85,0.90],  [0.68,0.85,0.90]], ..., [[0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87],  ...,  [0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87]], [[0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87],  ...,  [0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87]], [[0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87],  ...,  [0.93,0.95,0.87],  [0.93,0.95,0.87],  [0.93,0.95,0.87]]]] | [[0.00,0.00],[1.00,0.00,0.00,0.00,0.00]] | [[0.21,0.14],[0.78,0.13,0.37,0.35,0.34]] | X
Total count: 1
      correct: 0
      incorrect: 1
Total percentage correct: 0.0

In [34]:

for i in range(100):
    sim.step()

In [35]:

def function(simulator, index):
    cam_image = simulator.get_image()
    return (simulator.canvas.render("pil"),
            cam_image.resize((cam_image.size[0] * 4,
                              cam_image.size[1] * 4)))

In [36]:

sim.playback(robot.history, function)
_images/Robot_Simulation_36_1.png
_images/Robot_Simulation_36_2.png 
In [37]:

def function(simulator, index):
    cam_image = simulator.get_image()
    return simulator.canvas.render("pil")

In [38]:

sim.movie(robot.history, function, movie_name="sim-robot.gif")

Out[38]: