def simulate():
# Animate the resulting policy.
builder = DiagramBuilder()
plant = builder.AddSystem(DoubleIntegrator())
vi_policy = builder.AddSystem(policy)
builder.Connect(plant.get_output_port(0), vi_policy.get_input_port(0))
builder.Connect(vi_policy.get_output_port(0), plant.get_input_port(0))
diagram = builder.Build()
simulator = Simulator(diagram)
logger = LogOutput(plant.get_output_port(0), builder)
logger.set_name("logger")
simulator.get_mutable_context().SetContinuousState(env.reset())
simulator.AdvanceTo(10. if get_ipython() is not None else 5.)
return logger
def __init__(self):
#import pdb; pdb.set_trace()
# Create a block diagram containing our system.
builder = DiagramBuilder()
# add the two decoupled plants (x(s)=u/s^2; y(s)=u/s^2)
plant_x = builder.AddSystem(DoubleIntegrator())
plant_y = builder.AddSystem(DoubleIntegrator())
plant_x.set_name("double_integrator_x")
plant_y.set_name("double_integrator_y")
# add the controller, lead compensator for now just to stablize it
controller_x = builder.AddSystem(
Controller(tau_num=2., tau_den=.2, k=1.))
controller_y = builder.AddSystem(
Controller(tau_num=2., tau_den=.2, k=1.))
controller_x.set_name("controller_x")
controller_y.set_name("controller_y")
# the perception's "Beam" model with the four sources of noise
x_meas_fb = builder.AddSystem(Adder(num_inputs=4, size=1))
x_meas_fb.set_name("x_fb")
y_meas_fb = builder.AddSystem(Adder(num_inputs=4, size=1))
y_meas_fb.set_name("y_fb")
y_perception = builder.AddSystem(Adder(num_inputs=2, size=1))
y_perception.set_name("range_measurment_y")
neg_y_meas = builder.AddSystem(Gain(k=-1., size=1))
neg_y_meas.set_name("neg_y_meas")
wall_position = builder.AddSystem(ConstantVectorSource([Y_wall]))
p_hit = builder.AddSystem(RandomSource(distribution=RandomDistribution.kGaussian, num_outputs=2,\
sampling_interval_sec=0.01))
p_hit.set_name("GaussionNoise(0,1)")
p_short = builder.AddSystem(RandomSource(distribution=RandomDistribution.kExponential, num_outputs=2,\
sampling_interval_sec=0.01))
p_short.set_name("ExponentialNoise(1)")
#p_max = builder.AddSystem(RandomSource(distribution=RandomDistribution.kUniform, num_outputs=1,\
# sampling_interval_sec=0.01))
p_rand = builder.AddSystem(RandomSource(distribution=RandomDistribution.kUniform, num_outputs=2,\
sampling_interval_sec=0.01))
p_rand.set_name("UniformNoise(0,1)")
#import pdb; pdb.set_trace()
p_hit_Dx = builder.AddSystem(Demultiplexer(size=2))
p_hit_Dx.set_name('Dmux1')
p_short_Dx = builder.AddSystem(Demultiplexer(size=2))
p_short_Dx.set_name('Dmux2')
p_rand_Dx = builder.AddSystem(Demultiplexer(size=2))
p_rand_Dx.set_name('Dmux3')
normgain_x = builder.AddSystem(Gain(k=normal_coeff, size=1))
normgain_x.set_name("Sigma_x")
normgain_y = builder.AddSystem(Gain(k=normal_coeff, size=1))
normgain_y.set_name("Sigma_y")
expgain_x = builder.AddSystem(Gain(k=exp_coeff, size=1))
expgain_x.set_name("Exp_x")
expgain_y = builder.AddSystem(Gain(k=exp_coeff, size=1))
expgain_y.set_name("Exp_y")
randgain_x = builder.AddSystem(Gain(k=rand_coeff, size=1))
randgain_x.set_name("Rand_x")
randgain_y = builder.AddSystem(Gain(k=rand_coeff, size=1))
randgain_y.set_name("Rand_y")
#maxgain_x = builder.AddSystem(Adder(num_inputs=2, size=1))
#maxgain_x.set_name("Max_x")
#maxgain_y = builder.AddSystem(Adder(num_inputs=2, size=1))
#maxgain_y.set_name("Max_y")
# the summation to get the error (closing the loop)
summ_x = builder.AddSystem(Adder(num_inputs=2, size=1))
summ_y = builder.AddSystem(Adder(num_inputs=2, size=1))
summ_x.set_name("summation_x")
summ_y.set_name("summation_y")
neg_x = builder.AddSystem(Gain(k=-1., size=1))
neg_y = builder.AddSystem(Gain(k=-1., size=1))
neg_uy = builder.AddSystem(Gain(k=-1., size=1))
neg_x.set_name("neg_x")
neg_y.set_name("neg_y")
neg_uy.set_name("neg_uy")
# wire up all the blocks (summation to the controller to the plant ...)
builder.Connect(summ_x.get_output_port(0),
controller_x.get_input_port(0)) # e_x
builder.Connect(summ_y.get_output_port(0),
controller_y.get_input_port(0)) # e_y
builder.Connect(controller_x.get_output_port(0),
plant_x.get_input_port(0)) # u_x
builder.Connect(
controller_y.get_output_port(0),
neg_uy.get_input_port(0)) # u_y (to deal with directions)
builder.Connect(neg_uy.get_output_port(0),
plant_y.get_input_port(0)) # u_y
builder.Connect(
plant_x.get_output_port(0),
x_meas_fb.get_input_port(0)) # perception, nominal state meas
builder.Connect(wall_position.get_output_port(0),
y_perception.get_input_port(0)) # perception
builder.Connect(plant_y.get_output_port(0),
neg_y_meas.get_input_port(0)) # perception
builder.Connect(neg_y_meas.get_output_port(0),
y_perception.get_input_port(1)) # perception, z meas
builder.Connect(
y_perception.get_output_port(0),
y_meas_fb.get_input_port(0)) # perception, nominal state meas
builder.Connect(p_hit.get_output_port(0),
p_hit_Dx.get_input_port(0)) # demux the noise
builder.Connect(p_hit_Dx.get_output_port(0),
normgain_x.get_input_port(0)) # normalize Normal dist
builder.Connect(normgain_x.get_output_port(0),
x_meas_fb.get_input_port(1)) # Normal dist
builder.Connect(p_hit_Dx.get_output_port(1),
normgain_y.get_input_port(0)) # normalize Normal dist
builder.Connect(normgain_y.get_output_port(0),
y_meas_fb.get_input_port(1)) # Normal dist
builder.Connect(p_short.get_output_port(0),
p_short_Dx.get_input_port(0)) # demux the noise
builder.Connect(p_short_Dx.get_output_port(0),
expgain_x.get_input_port(0)) # normalize Exp dist
builder.Connect(expgain_x.get_output_port(0),
x_meas_fb.get_input_port(2)) # Exp dist
builder.Connect(p_short_Dx.get_output_port(1),
expgain_y.get_input_port(0)) # normalize Exp dist
builder.Connect(expgain_y.get_output_port(0),
y_meas_fb.get_input_port(2)) # Exp dist
#builder.Connect(p_max.get_output_port(0), x_meas_fb.get_input_port(3)) # Uniform dist
#builder.Connect(p_max.get_output_port(0), y_meas_fb.get_input_port(3)) # Uniform dist
builder.Connect(p_rand.get_output_port(0),
p_rand_Dx.get_input_port(0)) # normalize Uniform dist
builder.Connect(p_rand_Dx.get_output_port(0),
randgain_x.get_input_port(0)) # normalize Uniform dist
builder.Connect(randgain_x.get_output_port(0),
x_meas_fb.get_input_port(3)) # Uniform dist
builder.Connect(p_rand_Dx.get_output_port(1),
randgain_y.get_input_port(0)) # normalize Uniform dist
builder.Connect(randgain_y.get_output_port(0),
y_meas_fb.get_input_port(3)) # Uniform dist
builder.Connect(x_meas_fb.get_output_port(0),
neg_x.get_input_port(0)) # -x
builder.Connect(y_meas_fb.get_output_port(0),
neg_y.get_input_port(0)) # -y
builder.Connect(neg_x.get_output_port(0),
summ_x.get_input_port(1)) # r-x
builder.Connect(neg_y.get_output_port(0),
summ_y.get_input_port(1)) # r-y
# Make the desired_state input of the controller, an input to the diagram.
builder.ExportInput(summ_x.get_input_port(0))
builder.ExportInput(summ_y.get_input_port(0))
# get telemetry
logger_x = LogOutput(plant_x.get_output_port(0), builder)
logger_noise_x = LogOutput(x_meas_fb.get_output_port(0), builder)
logger_y = LogOutput(plant_y.get_output_port(0), builder)
logger_noise_y = LogOutput(y_meas_fb.get_output_port(0), builder)
logger_x.set_name("logger_x_state")
logger_y.set_name("logger_y_state")
logger_noise_x.set_name("x_state_meas")
logger_noise_y.set_name("y_meas")
# compile the system
diagram = builder.Build()
diagram.set_name("diagram")
# Create the simulator, and simulate for 10 seconds.
simulator = Simulator(diagram)
context = simulator.get_mutable_context()
# First we extract the subsystem context for the plant
plant_context_x = diagram.GetMutableSubsystemContext(plant_x, context)
plant_context_y = diagram.GetMutableSubsystemContext(plant_y, context)
# Then we can set the pendulum state, which is (x, y).
z0 = X_0
plant_context_x.get_mutable_continuous_state_vector().SetFromVector(z0)
z0 = Y_0
plant_context_y.get_mutable_continuous_state_vector().SetFromVector(z0)
# The diagram has a single input port (port index 0), which is the desired_state.
context.FixInputPort(
0,
[X_d]) # here we assume no perception, closing feedback on state X
context.FixInputPort(
1, [Z_d]) #Z_y, keep 3m away, basically we want to be at Y=0
# run the simulation
simulator.AdvanceTo(10)
t = logger_x.sample_times()
#import pdb; pdb.set_trace()
# Plot the results.
plt.figure()
plot_system_graphviz(diagram, max_depth=2)
plt.figure()
plt.plot(t, logger_noise_x.data().transpose(), label='x_state_meas')
plt.plot(t, logger_noise_y.data().transpose(), label='y_meas (z(y))')
plt.plot(t, logger_x.data().transpose(), label='x_state')
plt.plot(t, logger_y.data().transpose(), label='y_state')
plt.xlabel('t')
plt.ylabel('x(t),y(t)')
plt.legend()
plt.show(block=True)