Пример #1
0
    def measurement_callback(self, t, dt, navdata):
        from plot import plot, plot_trajectory
        #plot("x_true", self.simulator.drone.x[0])
        #plot("y_true", self.simulator.drone.x[1])
        #plot("z_true", self.simulator.drone.x[2])
    
        rot = self.rotation_to_world(navdata.rotZ)
        
        updated_state = self.copy_state(self.state[-1]);
        updated_state.velocity = np.dot(rot, np.array([[navdata.vx], [navdata.vy], [navdata.vz]])) + self.noise()
        updated_state.position += dt * updated_state.velocity
        self.state.append(updated_state);
        
        current = self.state.pop(0)

        self.update_setpoint(t)
        
        plot_trajectory("setpoint", self.state_desired.position)
        plot_trajectory("integrated", current.position)
        lin_vel = self.user.compute_control_command(t, dt, current, self.copy_state(self.state_desired))
        
        ad = current.position - self.state_desired.position
        al = math.sqrt(ad.item(0)*ad.item(0) + ad.item(1)*ad.item(1) + ad.item(2)*ad.item(2))
        if (al < 0.15):
            self.state_desired.position[1] = 3

        self.simulator.set_input_world(lin_vel, 0, self.state_desired)
Пример #2
0
    def measurement_callback(self, t, dt, navdata):        
        import plot
        
        rot = self.rotation_to_world(navdata.rotZ)
        self.state.velocity = np.dot(rot, np.array([[navdata.vx], [navdata.vy], [navdata.vz]]))# + self.noise()
        self.state.position += dt * self.state.velocity


        self.update_setpoint(t)
        self.kalman_state = State();
        self.kalman_state.position[0:2] = self.user.x[0:2];
        self.kalman_state.velocity[0:2] = self.user.x[2:4];
        lin_vel = self.controller.compute_control_command(t, dt, self.kalman_state, self.state_desired) + self.noise();
        self.simulator.set_input_world(lin_vel, 0)
        self.user.state_callback()

        vis_setpoint = self.state_desired.position;
        vis_setpoint[2] += 0.01
        plot.plot_trajectory("setpoint", vis_setpoint);

        if(self.counter % 15 == 0):
            measurement = self.state.position + self.noise2();
            self.user.measurement_callback(measurement[0:2])
    
        self.counter +=1
Пример #3
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # Translation coordinates x and y
     posX = np.multiply(navdata.vx, dt)
     posY = np.multiply(navdata.vy, dt)
     
     # Rotation about Z axis
     rotPsi = navdata.rotZ
     
     # Matrix of Quadcopter's local position
     localPos = np.array([[posX],
                          [posY]])
     
     # Transformation matrix to transform local to global coordinates
     rotMatrix = np.array([[cos(rotPsi), -sin(rotPsi)], 
                        [sin(rotPsi), cos(rotPsi)]])
     
     # Transform from local to global
     globalPos = np.dot(rotMatrix, localPos)
     
     # Quadcopter Global coordinates position update
     self.position += globalPos
     
     
     # TODO: update self.position by integrating measurements contained in navdata
     plot_trajectory("odometry", self.position)
Пример #4
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        # Translation coordinates x and y
        posX = np.multiply(navdata.vx, dt)
        posY = np.multiply(navdata.vy, dt)

        # Rotation about Z axis
        rotPsi = navdata.rotZ

        # Matrix of Quadcopter's local position
        localPos = np.array([[posX], [posY]])

        # Transformation matrix to transform local to global coordinates
        rotMatrix = np.array([[cos(rotPsi), -sin(rotPsi)],
                              [sin(rotPsi), cos(rotPsi)]])

        # Transform from local to global
        globalPos = np.dot(rotMatrix, localPos)

        # Quadcopter Global coordinates position update
        self.position += globalPos

        # TODO: update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)
Пример #5
0
    def measurement_callback(self, t, dt, navdata):
        import plot

        rot = self.rotation_to_world(navdata.rotZ)
        self.state.velocity = np.dot(rot,
                                     np.array([[navdata.vx], [navdata.vy],
                                               [navdata.vz]
                                               ]))  # + self.noise()
        self.state.position += dt * self.state.velocity

        self.update_setpoint(t)
        self.kalman_state = State()
        self.kalman_state.position[0:2] = self.user.x[0:2]
        self.kalman_state.velocity[0:2] = self.user.x[2:4]
        lin_vel = self.controller.compute_control_command(
            t, dt, self.kalman_state, self.state_desired) + self.noise()
        self.simulator.set_input_world(lin_vel, 0)
        self.user.state_callback()

        vis_setpoint = self.state_desired.position
        vis_setpoint[2] += 0.01
        plot.plot_trajectory("setpoint", vis_setpoint)

        if (self.counter % 15 == 0):
            measurement = self.state.position + self.noise2()
            self.user.measurement_callback(measurement[0:2])

        self.counter += 1
Пример #6
0
Файл: hw01.py Проект: z80/hw
 def state_callback(self):
     self.x = self.predictState(self.A, self.x)
     self.sigma = self.predictCovariance(self.A, self.sigma, self.Q)
     
     # visualize position state
     plot_trajectory("kalman", self.x[0:2])
     plot_covariance_2d("kalman", self.sigma[0:2,0:2])
Пример #7
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        #self.position[0] += cos(navdata.rotZ) * navdata.vx * dt - sin(navdata.rotZ) * navdata.vy * dt
        #self.position[1] += sin(navdata.rotZ) * navdata.vx * dt + cos(navdata.rotZ) * navdata.vy * dt

        cs = cos(navdata.rotZ)
        sn = sin(navdata.rotZ)
        R = np.array([[cs, -sn], [sn, cs]])
        v = np.array([[navdata.vx, navdata.vy]]).transpose()
        move = np.dot(R, v) * dt
        self.position += move

        # TODO: update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)


#Navdata
#The navdata contains the following values:
#
#rotX - roll angle in radians (rotation around x axis)
#rotY - pitch angle in radians (rotation around y axis)
#rotZ - yaw angle in radians (rotation around z axis)
#vx - velocity along the x axis (local)
#vy - velocity along the y axis (local)
    def state_callback(self):
        self.x = self.predictState(self.A, self.x)
        self.sigma = self.predictCovariance(self.A, self.sigma, self.Q)

        # visualize position state
        plot_trajectory("kalman", self.x[0:2])
        plot_covariance_2d("kalman", self.sigma[0:2, 0:2])
Пример #9
0
    def measurement_callback(self, t, dt, navdata):
        # t: time since simulation start
        # dt : time since last call to measurement_callback
        # navdata : measurements of the quadrotor

        self.position = self.position + dt * np.dot(self.rotation_to_world(navdata.rotZ),
                                                    np.array([[navdata.vx], [navdata.vy]]))
        plot_trajectory("odometry", self.position)
Пример #10
0
    def measurement_callback(self, marker_position_world, marker_yaw_world,
                             marker_position_relative, marker_yaw_relative):
        '''
        :param measurement: vector of measured coordinates
        '''

        while (marker_yaw_relative > math.pi):
            marker_yaw_relative -= 2 * math.pi
        while (marker_yaw_relative < -math.pi):
            marker_yaw_relative += 2 * math.pi
        from math import sin, cos
        #meas = Pose2D(marker_yaw_world, marker_position_world) * Pose2D(marker_yaw_relative, marker_position_relative).inv();
        measurement = np.array([[
            marker_position_relative[0], marker_position_relative[1],
            marker_yaw_relative
        ]]).T

        s_psi = sin(self.x[2])
        c_psi = cos(self.x[2])

        #TODO: Enter the Jacobian derived in the previous assignment
        dx = marker_position_world[0] - self.x[0]
        dy = marker_position_world[1] - self.x[1]
        self.H = np.array([[-c_psi, -s_psi, s_psi * dx - c_psi * dy],
                           [s_psi, -c_psi, c_psi * dx + s_psi * dy],
                           [0, 0, -1]])
        I = np.zeros((3, 3))
        self.H = np.hstack((self.H, I))

        predicted_meas = Pose2D(self.x[2], self.x[0:2]).inv() * Pose2D(
            marker_yaw_world, marker_position_world)
        predicted_measurement = np.array([[
            predicted_meas.translation[0], predicted_meas.translation[1],
            predicted_meas.yaw()
        ]]).T

        while (predicted_measurement[2] > math.pi):
            predicted_measurement[2] -= 2 * math.pi
        while (predicted_measurement[2] < -math.pi):
            predicted_measurement[2] += 2 * math.pi

        print "z_predicted", predicted_measurement.T
        print "z", measurement.T

        # visualize measurement
        plot_point("gps", measurement[0:2])

        k = self.calculateKalmanGain(self.sigma, self.H, self.R)
        #print "kalman", k
        #print "before", self.x.T
        self.x = self.correctState(measurement, self.x, k,
                                   predicted_measurement)
        self.sigma = self.correctCovariance(self.sigma, k, self.H)
        #print "after", self.x.T
        # visualize position state
        p = self.x[0:2]
        plot_trajectory("kalman", p)
        plot_covariance_2d("kalman", self.sigma[0:2, 0:2])
Пример #11
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # TODO: update self.position by integrating measurements contained in navdata
     plot_trajectory("odometry", self.position)
Пример #12
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     self.position = self.position + dt * np.dot(self.rotation_to_world(navdata.rotZ), np.array([[navdata.vx], [navdata.vy]]))
     plot_trajectory("odometry", self.position)
Пример #13
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        # TODO: update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)
Пример #14
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # TODO: update self.position by integrating measurements contained in navdata
     self.position += dt * np.dot(self.rotation_to_world(navdata.rotZ), np.array([[navdata.vx], [navdata.vy]]))
     plot_trajectory("odometry", self.position)
Пример #15
0
def main():
    args = parse_args()
    header, counts = read_counts(args.counts)
    counts = pd.DataFrame(counts, columns=header)
    counts = sort_counts(counts)

    num_genes = int(args.num_genes)-1
    out_dir = args.out_dir

    # plot pca
    print('Generating PCA plot ...')
    plot.plot_pca(out_dir, counts)

    # plot heatmap
    print('Generating Gene Expression Heatmap ...')
    plot.plot_heatmap(out_dir, counts, num_genes)

    # Plot gene expression tragectory
    print('Generating Gene Expression Trajectory ...')
    plot.plot_trajectory(out_dir, counts, num_genes)

    # Save formatted and sorted count file
    counts.to_csv('data/counts_clust.txt', index=False, sep='\t')

    # Call clust to cluster the genes
    print('Clustering Genes (this might take awhile) ...')
    subprocess.call(['mkdir', 'clust_out'])
    if args.kmeans is not None:
        subprocess.call(['clust', 'data/counts_clust.txt',
                         '-o', './clust_out', '-K', args.kmeans])
    else:
        subprocess.call(['clust', 'data/counts_clust.txt',
                        '-o', './clust_out'])

    # Make sure everythin is all set for motif enrichment
    clust_out = args.input_genes
    ref_bed = args.bedfile
    ref_fa = args.genome
    ref_motif = args.motif

    try:
        open(clust_out)
    except FileNotFoundError:
        print('Clustering failed, '
              + 'Clusters_Objects.tsv not found in the output directory')

    print('Motif Enrichment Analysis (this might take awhile) ...')
    try:
        enrichment.run_motif_enrichment(clust_out,
                                        ref_bed,
                                        ref_fa, ref_motif,
                                        args.ame,
                                        100, 100)
    except NameError:
        print('AME is not installed! Please install AME and try again')
Пример #16
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     cur_mat = rot_trans_mat(self.position[0], self.position[1], navdata.rotZ)
     new_pos = dot(cur_mat, array([[navdata.vx * dt], [navdata.vy * dt], [1.0]]))
     #print new_pos
     self.position = array([[new_pos[0]], [new_pos[1]]])
     plot_trajectory("odometry", self.position)
Пример #17
0
 def measurement_callback(self, t, dt, navdata):        
     import plot
     from math import hypot
     
     yaw_vel = (navdata.rotZ - self.last_yaw) / dt
     self.last_yaw = navdata.rotZ
     
     noise_lin_vel = np.random.normal(0.0, 0.1, (2, 1))
     noise_yaw_vel = np.random.normal(0.0, 0.1, (1, 1))
     
     u = self.user.state_callback(t, dt, np.array([[navdata.vx, navdata.vy]]).T + noise_lin_vel, yaw_vel + noise_yaw_vel[0])
     
     #for beacon in self.user.beacons: #should the beacons be in user or internal? the grading safely happens in javascript right?
     #    if(hypot(beacon.position[0] - self.simulator.drone.x[0], beacon.position[1] - self.simulator.drone.x[1]) <= 0.5):
     #        self.user.beacon_callback(beacon)
 
     if(self.counter % 7 == 0): # approx. 30Hz
         
         for i in range(len(self.markers)):
             visible, rel_pos, rel_yaw = self.is_marker_visible(self.markers[i])
             
             if visible:
                 plot.plot_trajectory("test", self.simulator.drone.x);
                 plot.plot_trajectory("test", np.dot(self.rotation_to_world(self.simulator.drone.theta[2]), rel_pos) + self.simulator.drone.x);
                 
                 noise_pos = np.random.normal(0, 0.005, (2, 1));
                 noise_yaw = np.random.normal(0, 0.03, (1, 1));
                 
                 marker_pos_world = np.array([[self.markers[i][0], self.markers[i][1]]]).T
                 marker_pos_quadrotor = rel_pos[0:2] + noise_pos
                 
                 marker_yaw_world = self.markers[i][2]
                 marker_yaw_quadrotor = rel_yaw + noise_yaw[0]
                 
                 self.user.measurement_callback(marker_pos_world, marker_yaw_world, marker_pos_quadrotor, marker_yaw_quadrotor)
                 break;
     
     self.counter += 1
     
     lin_vel = np.zeros((2,1))
     yaw_vel = 0
     
     if type(u) is tuple and len(u) == 2:
         noise_lin_vel = np.random.normal(0.0, 0.15, (2, 1))
         noise_yaw_vel = np.random.normal(0.1, 0.1, (1, 1))
         
         lin_vel = u[0] + noise_lin_vel
         yaw_vel = u[1] + noise_yaw_vel[0]
     else:
         print "state_callback(...) should return a tuple of linear velocity control command (2x1 numpy.array) and yaw velocity command (float)!"
     
     u_internal = [lin_vel[0], lin_vel[1], yaw_vel, 0];
     
     self.simulator.set_input(u_internal)
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        # TODO: update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)
        distance = navdata.vx * dt
        dx = distance * cos(navdata.rotZ)
        dy = distance * sin(navdata.rotZ)
        self.position = self.position + np.array([[dx], [dy]])
Пример #19
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     cur_mat = rot_trans_mat(self.position[0], self.position[1],
                             navdata.rotZ)
     new_pos = dot(cur_mat,
                   array([[navdata.vx * dt], [navdata.vy * dt], [1.0]]))
     #print new_pos
     self.position = array([[new_pos[0]], [new_pos[1]]])
     plot_trajectory("odometry", self.position)
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # TODO: update self.position by integrating measurements contained in navdata
     plot_trajectory("odometry", self.position)
     distance = navdata.vx*dt
     dx = distance*cos(navdata.rotZ)
     dy = distance*sin(navdata.rotZ)
     self.position = self.position + np.array([[dx],[dy]])
Пример #21
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        d = navdata.vx * dt
        a = navdata.rotZ
        t = np.array([[d * math.cos(a)], [d * math.sin(a)]])
        self.position = self.position + t

        # update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)
Пример #22
0
    def measurement_callback(self, measurement):
        '''
        param measurement: vector of measured coordinates
        '''

        # Visualize measurement
        plot_point("gps", measurement)
        k = self.calculateKalmanGain(self.sigma, self.H, self.R)
        self.x = self.correctState(measurement, self.x, k, self.H)
        self.sigma = self.correctCovariance(self.sigma, k, self.H)

        # Visualize position state
        plot_trajectory("kalman", self.x[0:2])
        plot_covariance_2d("kalman", self.sigma[0:2, 0:2])
Пример #23
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     local_velocity = np.array([[navdata.vx], [navdata.vy]])
     prev_position = self.position
     angle = navdata.rotZ
     rotation_to_global = np.array([[cos(angle), -sin(angle)],
                                    [sin(angle), cos(angle)]])
     velocity = np.array([[navdata.vx], [navdata.vy]])
     self.position += dt * np.dot(rotation_to_global, velocity)
     # TODO: update self.position by integrating measurements contained in navdata
     plot_trajectory("odometry", self.position)
Пример #24
0
Файл: hw01.py Проект: z80/hw
 def measurement_callback(self, measurement):
     '''
     :param measurement: vector of measured coordinates
     '''
     
     # visualize measurement
     plot_point("gps", measurement)
     
     k = self.calculateKalmanGain(self.sigma, self.H, self.R)
     
     self.x = self.correctState(measurement, self.x, k, self.H)
     self.sigma = self.correctCovariance(self.sigma, k, self.H)
     
     # visualize position state
     plot_trajectory("kalman", self.x[0:2])
     plot_covariance_2d("kalman", self.sigma[0:2,0:2])
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # TODO: update self.position by integrating measurements contained in navdata
     #mat = np.array([[cos(navdata.rotZ), -sin(navdata.rotZ), navdata.vx], [sin(navdata.rotZ), cos(navdata.rotZ), navdata.vy], [0, 0, 1]])
     #vec = mat * np.array([self.position[0,0], self.position[1, 0], 0])
     #self.position = np.array([[vec[0,0]], [vec[1, 0]]])
     pos_local=np.array([[navdata.vx*dt],[navdata.vy*dt]])
     R=np.array([[cos(navdata.rotZ),-1.0*sin(navdata.rotZ)],[sin(navdata.rotZ),cos(navdata.rotZ)]])
     pos_global=np.dot(R,pos_local)
     self.position +=pos_global
     plot_trajectory("odometry", self.position)
Пример #26
0
def main():
    # common style arguments for plotting
    style = {
        'border': {
            'color': 'red',
            'linewidth': 0.5
        },
    }

    # set-up mdp
    world, reward, terminal = setup_mdp()

    # show our original reward
    ax = plt.figure(num='Original Reward').add_subplot(111)
    P.plot_state_values(ax, world, reward, **style)
    plt.draw()

    # generate "expert" trajectories
    trajectories, expert_policy = generate_trajectories(
        world, reward, terminal)

    # show our expert policies
    ax = plt.figure(num='Expert Trajectories and Policy').add_subplot(111)
    P.plot_stochastic_policy(ax, world, expert_policy, **style)

    for t in trajectories:
        P.plot_trajectory(ax, world, t, lw=5, color='white', alpha=0.025)

    plt.draw()

    # maximum entropy reinforcement learning (non-causal)
    reward_maxent = maxent(world, terminal, trajectories)

    # show the computed reward
    ax = plt.figure(num='MaxEnt Reward').add_subplot(111)
    P.plot_state_values(ax, world, reward_maxent, **style)
    plt.draw()

    # maximum casal entropy reinforcement learning (non-causal)
    reward_maxcausal = maxent_causal(world, terminal, trajectories)

    # show the computed reward
    ax = plt.figure(num='MaxEnt Reward (Causal)').add_subplot(111)
    P.plot_state_values(ax, world, reward_maxcausal, **style)
    plt.draw()

    plt.show()
Пример #27
0
 def measurement_callback(self, t, dt, navdata):
     '''
     :param t: time since simulation start
     :param dt: time since last call to measurement_callback
     :param navdata: measurements of the quadrotor
     '''
     
     # TODO: update self.position by integrating measurements contained in navdata
     R = np.array([[cos(navdata.rotZ),-1*sin(navdata.rotZ),self.position[0]],
                   [sin(navdata.rotZ),cos(navdata.rotZ),self.position[1]],
                   [0,0,1]]);
     
     x = navdata.vx * dt;
     y = navdata.vy * dt;
     P_global = np.dot(R, np.array([[x],[y],[1]]));
     self.position = np.array([[P_global[0]],[P_global[1]]]);
     plot_trajectory("odometry", self.position)
Пример #28
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        # TODO: update self.position by integrating measurements contained in navdata
        R = np.array(
            [[cos(navdata.rotZ), -1 * sin(navdata.rotZ), self.position[0]],
             [sin(navdata.rotZ),
              cos(navdata.rotZ), self.position[1]], [0, 0, 1]])

        x = navdata.vx * dt
        y = navdata.vy * dt
        P_global = np.dot(R, np.array([[x], [y], [1]]))
        self.position = np.array([[P_global[0]], [P_global[1]]])
        plot_trajectory("odometry", self.position)
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor

        :The navdata contains the following values:
        :rotX - roll angle in radians (rotation around x axis)
        :rotY - pitch angle in radians (rotation around y axis)
        :rotZ - yaw angle in radians (rotation around z axis)
        :vx - velocity along the x axis
        :vy - velocity along the y axis
        '''

        # rotZ gives us the absolute angle around the z axis
        # positive is in the counter clockwise direction
        self.position[0] += navdata.vx * dt * cos(navdata.rotZ)
        self.position[1] += navdata.vx * dt * sin(navdata.rotZ)
        plot_trajectory("odometry", self.position)
Пример #30
0
    def state_callback(self, dt, linear_velocity, yaw_velocity):
        from math import sin, cos

        self.x[3:5] = linear_velocity
        self.x[5] = yaw_velocity
        #print yaw_velocity
        self.A[0:2, 3:5] = np.array([[cos(self.x[2]), -sin(self.x[2])],
                                     [sin(self.x[2]),
                                      cos(self.x[2])]]) * dt
        self.A[3:5, 3:5] = np.array([[cos(self.x[2]), -sin(self.x[2])],
                                     [sin(self.x[2]),
                                      cos(self.x[2])]])

        self.x = self.predictState(self.A, self.x)
        self.sigma = self.predictCovariance(self.A, self.sigma, self.Q)

        # visualize position state
        plot_trajectory("kalman", self.x[0:2])
        plot_covariance_2d("kalman", self.sigma[0:2, 0:2])
Пример #31
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor

        :The navdata contains the following values:
        :rotX - roll angle in radians (rotation around x axis)
        :rotY - pitch angle in radians (rotation around y axis)
        :rotZ - yaw angle in radians (rotation around z axis)
        :vx - velocity along the x axis
        :vy - velocity along the y axis
        '''
        
        # rotZ gives us the absolute angle around the z axis
        # positive is in the counter clockwise direction
        self.position[0] += navdata.vx*dt*cos(navdata.rotZ)
        self.position[1] += navdata.vx*dt*sin(navdata.rotZ)
        plot_trajectory("odometry", self.position)
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        # compute transformation matrix
        psi = navdata.rotZ
        R = np.array([[cos(psi), -sin(psi)], [sin(psi), cos(psi)]])

        vx = navdata.vx
        vy = navdata.vy
        ds_local = np.array([[vx * dt],[vy * dt]])

        ds_global = np.dot(R, ds_local)

        self.position = self.position + ds_global

        # TODO: update self.position by integrating measurements contained in navdata
        plot_trajectory("odometry", self.position)
Пример #33
0
    def measurement_callback(self, t, dt, navdata):
        from plot import plot, plot_trajectory
        #plot("x_true", self.simulator.drone.x[0])
        #plot("y_true", self.simulator.drone.x[1])
        #plot("z_true", self.simulator.drone.x[2])
    
        rot = self.rotation_to_world(navdata.rotZ)
        
        updated_state = self.copy_state(self.state[-1]);
        updated_state.velocity = np.dot(rot, np.array([[navdata.vx], [navdata.vy], [navdata.vz]])) + self.noise()
        updated_state.position += dt * updated_state.velocity
        self.state.append(updated_state);
        
        current = self.state.pop(0)

        self.update_setpoint(t)
        
        plot_trajectory("setpoint", self.state_desired.position)
        plot_trajectory("integrated", current.position)
        lin_vel = self.user.compute_control_command(t, dt, current, self.copy_state(self.state_desired))
        
        self.simulator.set_input_world(lin_vel, 0)
Пример #34
0
    def measurement_callback(self, t, dt, navdata):
        '''
        :param t: time since simulation start
        :param dt: time since last call to measurement_callback
        :param navdata: measurements of the quadrotor
        '''

        #Rotation matrix using homogeneous coordinates
        R = np.array([[cos(navdata.rotZ), sin(navdata.rotZ)*-1, self.position[0]], [sin(navdata.rotZ), cos(navdata.rotZ), self.position[1]], [0,0,1]]);

        #Position vector in local coordinates (homogeneous)
        localPos = np.array([[navdata.vx*dt], [navdata.vy*dt], [1]]);

        #Multiply to transform into global coordinates
        globalPos = np.dot(R, localPos);
        
        self.position[0] = globalPos[0];
        self.position[1] = globalPos[1];
      


        
        
        plot_trajectory("odometry", self.position)
Пример #35
0
    def measurement_callback(self, t, dt, navdata):        
        import plot
                
        rot = self.rotation_to_world(navdata.rotZ)
        self.state.lin_velocity = np.dot(rot, np.array([[navdata.vx], [navdata.vy], [navdata.vz]]))

        self.state.position += dt * self.state.lin_velocity
        self.state.orientation = np.array([[navdata.rotX], [navdata.rotY], [navdata.rotZ]])

        self.update_setpoint(t)

        yaw_vel = (navdata.rotZ - self.last_yaw) / dt
        self.last_yaw = navdata.rotZ
        
        noise_lin_vel = np.random.normal(0.0, 0.5, (2, 1))
        noise_yaw_vel = np.random.normal(0.05, 0.5, (1, 1))
        
        self.user.state_callback(t, dt, np.array([[navdata.vx, navdata.vy]]).T + noise_lin_vel, yaw_vel + noise_yaw_vel[0])

        vis_setpoint = self.state_desired.position;
        vis_setpoint[2] += 0.01
        plot.plot_trajectory("setpoint", vis_setpoint);

        if(self.counter % 7 == 0): # approx. 30Hz
            
            for i in range(len(self.markers)):
                visible, rel_pos, rel_yaw = self.is_marker_visible(self.markers[i], self.state)
                
                if visible:
                    plot.plot_trajectory("test", self.state.position);
                    plot.plot_trajectory("test", np.dot(self.rotation_to_world(self.state.orientation[2]), rel_pos) + self.state.position);
                    
                    noise_pos = np.random.normal(0, 0.005, (2, 1));
                    noise_yaw = np.random.normal(0, 0.03, (1, 1));
                    
                    marker_pos_world = np.array([[self.markers[i][0], self.markers[i][1]]]).T
                    marker_pos_quadrotor = rel_pos[0:2] + noise_pos
                    
                    marker_yaw_world = self.markers[i][2]
                    marker_yaw_quadrotor = rel_yaw + noise_yaw[0]
                    
                    self.user.measurement_callback(marker_pos_world, marker_yaw_world, marker_pos_quadrotor, marker_yaw_quadrotor)
                    break;
                        
    
        self.counter +=1

        self.kalman_state = State();
        self.kalman_state.position[0:2] = self.user.x[0:2]; #xy
        self.kalman_state.position[2] = self.state.position[2]; #z
        self.kalman_state.orientation[2] = self.user.x[2]; #psi

        #self.kalman_state.lin_velocity[0:2] = self.user.x[3:5]; #xdot,ydot
        #self.kalman_state.ang_velocity[2] = self.user.x[5]; #psidot
                    
        #print "est yaw: ", self.kalman_state.orientation[2]
        
        tmp = navdata.rotZ
    
        while(tmp > math.pi):
            tmp -= 2 * math.pi
        while(tmp < -math.pi):
           tmp += 2 * math.pi
        #print "act yaw: ", tmp
        
        lin_vel, ang_vel = self.controller.compute_control_command(t, dt, self.kalman_state, self.state_desired) + self.noise();
        self.simulator.set_input_world(lin_vel, ang_vel)
Пример #36
0
def mce_irl(grid_size, p_slip, avoid_states):

    cwd = os.getcwd()
    fig_dir = "figs"
    if not os.path.exists(fig_dir):
        os.makedirs(fig_dir)
   
    # common style arguments for plotting
    style = {
        'border': {'color': 'red', 'linewidth': 0.5},
        'cmap': "Blues",
    }

    # set-up mdp
    world, reward, terminal = setup_mdp(grid_size, p_slip, avoid_states)

    # show our original reward
    ax = plt.figure(num='Original Reward').add_subplot(111)
    tt = P.plot_state_values(ax, world, reward, **style)
    plt.colorbar(tt)
    plt.draw()

    plt.savefig(os.path.join(fig_dir, "gt_reward.png"))


    print("\nGenerating expert trajectories ...\n")
    # generate "expert" trajectories
    trajectories, expert_policy = generate_trajectories(world, reward, terminal)

    # show our expert policies
    ax = plt.figure(num='Expert Trajectories and Policy').add_subplot(111)
    P.plot_stochastic_policy(ax, world, expert_policy, **style)

    for t in trajectories:
        P.plot_trajectory(ax, world, t, lw=5, color='white', alpha=0.025)

    plt.draw()
    plt.savefig(os.path.join(fig_dir, "demonstrations.png"))


    '''
    print("ME-IRL ...")
    # maximum entropy reinforcement learning (non-causal)
    reward_maxent = maxent(world, terminal, trajectories)

    # show the computed reward
    ax = plt.figure(num='MaxEnt Reward').add_subplot(111)
    P.plot_state_values(ax, world, reward_maxent, **style)
    plt.draw()
    '''

    print("\nPerforming MCE-IRL ...\n")
    # maximum casal entropy reinforcement learning (non-causal)
    reward_maxcausal = maxent_causal(world, terminal, trajectories)

    # print(reward_maxcausal)

    # show the computed reward
    ax = plt.figure(num='MaxEnt Reward (Causal)').add_subplot(111)
    tt = P.plot_state_values(ax, world, reward_maxcausal, **style)
    plt.colorbar(tt)
    plt.draw()
    plt.savefig(os.path.join(fig_dir, "maxent_causal_reward.png"))

    print("\nDone! Rewards learned\n")

    # plt.show()
    return (reward_maxcausal, world, terminal)
 def visualizeState(self):
     # visualize position state
     plot_trajectory("kalman", self.x[0:2])
     plot_covariance_2d("kalman", self.sigma[0:2,0:2])
Пример #38
0
 def visualizeState(self):
     # visualize position state
     plot_trajectory("kalman", self.x[0:2])
     plot_covariance_2d("kalman", self.sigma[0:2, 0:2])
Пример #39
0
    def simulate_step(self, t, dt):
        self.step_count += 1



        #if(int(t / 8.0) % 2 == 0):
        #    self.xdot_desired[0] = 0.75;
        #    self.xdot_desired[1] = 0.1;
        #else:
        #    self.xdot_desired[0] = -0.75;
        #    self.xdot_desired[1] = -0.1;
        #self.xdot_desired = np.dot(self.drone.yaw_rotation(), self.xdot_desired)


        #inputCurrents = self.controller.calculate_control_command(dt, self.theta_desired, self.thetadot_desired,self.x_desired,self.xdot_desired)
        inputCurrents, acc = self.controller.calculate_control_command3(dt, self.xdot_desired, self.thetadot_desired.item(2), self.theta_desired, self.x_desired)
        omega = self.drone.omega;#thetadot_in_body()  # calculate current angular velocity in body frame

        #torques_thrust      = self.drone.torques_thrust(np.array([inputCurrents]).transpose())
        torques_thrust       = self.drone.torques_thrust(inputCurrents)
        #print acc.transpose();
        # print omega.transpose()
        linear_acceleration  = self.linear_acceleration(torques_thrust[3], self.drone.theta, self.drone.xdot)  # calculate the resulting linear acceleration
        omegadot             = self.angular_acceleration(torques_thrust[0:3,0], omega)  # calculate resulting angular acceleration
        #print "angular acc", omegadot.transpose()
        #linear_acceleration = self.linear_acceleration2(inputCurrents, self.drone.theta, self.drone.xdot)  # calculate the resulting linear acceleration
        #omegadot            = self.angular_acceleration2(inputCurrents, omega)  # calculate resulting angular acceleration

        omega = omega + dt * omegadot  # integrate up new angular velocity in the body frame

        #print "inputs:", inputCurrents
        #print "omega:", omega.transpose(), "omegadot:", omegadot.transpose()
        self.drone.omega    = omega;
        self.drone.thetadot = np.dot(self.drone.angle_rotation_to_world().transpose(), omega)  # calculate roll, pitch, yaw velocities
        self.drone.theta    = self.drone.theta + dt * self.drone.thetadot  # calculate new roll, pitch, yaw angles

        #print self.drone.xdot.transpose(), self.drone.theta.transpose(), omega.transpose(), omegadot.transpose()
        #print "d", inputCurrents
        #print "a", torques_thrust[0:3,0], "b", omega, "c", omegadot

        #print "thetadot:",self.drone.thetadot
        #print("New theta",self.drone.theta)
        self.drone.xdoubledot = linear_acceleration
        self.drone.xdot       = self.drone.xdot + dt * linear_acceleration  # calculate new linear drone speed
        self.drone.x          = self.drone.x + dt * self.drone.xdot  # calculate new drone position
        #print "acc", linear_acceleration
        #print "theta", self.drone.theta
        #print("Position",self.drone.x.transpose())

        if(sys.platform == "skulpt"):
            import plot
            plot.plot_pose("ardrone", self.drone.x, self.drone.theta)
            plot.plot_trajectory("ardrone", self.drone.x)
            plot.plot_motor_command("ardrone", inputCurrents)


        elif(self.step_count % 50 == 0): #save trajectory for plotting
            vel  = np.dot(self.rotation(self.drone.theta).transpose(), self.drone.xdot)
            vel  = self.drone.xdot
            #vel = self.drone.x
            #vel = acc
            #vel = self.drone.thetadot_in_body();
            #vel = self.drone.xdot;

            self.x.append(vel.item(0))
            self.y.append(vel.item(1))
            self.z.append(vel.item(2))

            ang = self.drone.theta

            self.roll.append( ang.item(0))
            self.pitch.append(ang.item(1))
            self.yaw.append(  ang.item(2))
            #print self.theta_desired.item(2)
            self.cmd1.append(inputCurrents[0] - inputCurrents[2])
            self.cmd2.append(inputCurrents[1] - inputCurrents[3])
            self.cmd3.append(inputCurrents[0] - inputCurrents[1])
            self.cmd4.append(inputCurrents[3])
            self.roll_des.append(self.theta_desired[0])
            self.pitch_des.append(self.theta_desired[1])
            self.yaw_des.append(self.theta_desired[2])
Пример #40
0
    def measurement_callback(self, t, dt, navdata):
        import plot

        rot = self.rotation_to_world(navdata.rotZ)
        self.state.lin_velocity = np.dot(
            rot, np.array([[navdata.vx], [navdata.vy], [navdata.vz]]))

        self.state.position += dt * self.state.lin_velocity
        self.state.orientation = np.array([[navdata.rotX], [navdata.rotY],
                                           [navdata.rotZ]])

        self.update_setpoint(t)

        yaw_vel = (navdata.rotZ - self.last_yaw) / dt
        self.last_yaw = navdata.rotZ

        noise_lin_vel = np.random.normal(0.0, 0.5, (2, 1))
        noise_yaw_vel = np.random.normal(0.05, 0.5, (1, 1))

        self.user.state_callback(
            t, dt,
            np.array([[navdata.vx, navdata.vy]]).T + noise_lin_vel,
            yaw_vel + noise_yaw_vel[0])

        vis_setpoint = self.state_desired.position
        vis_setpoint[2] += 0.01
        plot.plot_trajectory("setpoint", vis_setpoint)

        if (self.counter % 7 == 0):  # approx. 30Hz

            for i in range(len(self.markers)):
                visible, rel_pos, rel_yaw = self.is_marker_visible(
                    self.markers[i], self.state)

                if visible:
                    plot.plot_trajectory("test", self.state.position)
                    plot.plot_trajectory(
                        "test",
                        np.dot(
                            self.rotation_to_world(self.state.orientation[2]),
                            rel_pos) + self.state.position)

                    noise_pos = np.random.normal(0, 0.005, (2, 1))
                    noise_yaw = np.random.normal(0, 0.03, (1, 1))

                    marker_pos_world = np.array(
                        [[self.markers[i][0], self.markers[i][1]]]).T
                    marker_pos_quadrotor = rel_pos[0:2] + noise_pos

                    marker_yaw_world = self.markers[i][2]
                    marker_yaw_quadrotor = rel_yaw + noise_yaw[0]

                    self.user.measurement_callback(marker_pos_world,
                                                   marker_yaw_world,
                                                   marker_pos_quadrotor,
                                                   marker_yaw_quadrotor)
                    break

        self.counter += 1

        self.kalman_state = State()
        self.kalman_state.position[0:2] = self.user.x[0:2]
        #xy
        self.kalman_state.position[2] = self.state.position[2]
        #z
        self.kalman_state.orientation[2] = self.user.x[2]
        #psi

        #self.kalman_state.lin_velocity[0:2] = self.user.x[3:5]; #xdot,ydot
        #self.kalman_state.ang_velocity[2] = self.user.x[5]; #psidot

        #print "est yaw: ", self.kalman_state.orientation[2]

        tmp = navdata.rotZ

        while (tmp > math.pi):
            tmp -= 2 * math.pi
        while (tmp < -math.pi):
            tmp += 2 * math.pi
        #print "act yaw: ", tmp

        lin_vel, ang_vel = self.controller.compute_control_command(
            t, dt, self.kalman_state, self.state_desired) + self.noise()
        self.simulator.set_input_world(lin_vel, ang_vel)
Пример #41
0
    def measurement_callback(self, t, dt, navdata):
        import plot
        from math import hypot

        yaw_vel = (navdata.rotZ - self.last_yaw) / dt
        self.last_yaw = navdata.rotZ

        noise_lin_vel = np.random.normal(0.0, 0.1, (2, 1))
        noise_yaw_vel = np.random.normal(0.0, 0.1, (1, 1))

        u = self.user.state_callback(
            t, dt,
            np.array([[navdata.vx, navdata.vy]]).T + noise_lin_vel,
            yaw_vel + noise_yaw_vel[0])

        #for beacon in self.user.beacons: #should the beacons be in user or internal? the grading safely happens in javascript right?
        #    if(hypot(beacon.position[0] - self.simulator.drone.x[0], beacon.position[1] - self.simulator.drone.x[1]) <= 0.5):
        #        self.user.beacon_callback(beacon)

        if (self.counter % 7 == 0):  # approx. 30Hz

            for i in range(len(self.markers)):
                visible, rel_pos, rel_yaw = self.is_marker_visible(
                    self.markers[i])

                if visible:
                    plot.plot_trajectory("test", self.simulator.drone.x)
                    plot.plot_trajectory(
                        "test",
                        np.dot(
                            self.rotation_to_world(
                                self.simulator.drone.theta[2]), rel_pos) +
                        self.simulator.drone.x)

                    noise_pos = np.random.normal(0, 0.005, (2, 1))
                    noise_yaw = np.random.normal(0, 0.03, (1, 1))

                    marker_pos_world = np.array(
                        [[self.markers[i][0], self.markers[i][1]]]).T
                    marker_pos_quadrotor = rel_pos[0:2] + noise_pos

                    marker_yaw_world = self.markers[i][2]
                    marker_yaw_quadrotor = rel_yaw + noise_yaw[0]

                    self.user.measurement_callback(marker_pos_world,
                                                   marker_yaw_world,
                                                   marker_pos_quadrotor,
                                                   marker_yaw_quadrotor)
                    break

        self.counter += 1

        lin_vel = np.zeros((2, 1))
        yaw_vel = 0

        if type(u) is tuple and len(u) == 2:
            noise_lin_vel = np.random.normal(0.0, 0.15, (2, 1))
            noise_yaw_vel = np.random.normal(0.1, 0.1, (1, 1))

            lin_vel = u[0] + noise_lin_vel
            yaw_vel = u[1] + noise_yaw_vel[0]
        else:
            print "state_callback(...) should return a tuple of linear velocity control command (2x1 numpy.array) and yaw velocity command (float)!"

        u_internal = [lin_vel[0], lin_vel[1], yaw_vel, 0]

        self.simulator.set_input(u_internal)
Пример #42
0
###############################################################################
## Plotting

method_dirs = ['curvature', 'compromise', 'laptime', 'sectors', 'estimated']
plot_dir = os.path.join(
  os.path.dirname(__file__), '..', 'data', 'plots', track.name,
  method_dirs[args.method]
)
if not os.path.exists(plot_dir): os.makedirs(plot_dir)

if args.plot_corners or args.plot_all:
  plot_corners(
    os.path.join(plot_dir, "corners." + args.ext),
    track.left, track.right, track.mid.position(trajectory.s),
    track.corners(trajectory.s, K_MIN, PROXIMITY, LENGTH)[1]
  )

if args.plot_path or args.plot_all:
  plot_path(
    os.path.join(plot_dir, "path." + args.ext),
    track.left, track.right, trajectory.path.position(trajectory.s),
    trajectory.path.controls
  )

if args.plot_trajectory or args.plot_all:
  plot_trajectory(
    os.path.join(plot_dir, "trajectory." + args.ext),
    track.left, track.right, trajectory.path.position(trajectory.s),
    trajectory.velocity.v
  )
Пример #43
0
    def simulate_step(self, t, dt):
        self.step_count += 1

        #if(int(t / 8.0) % 2 == 0):
        #    self.xdot_desired[0] = 0.75;
        #    self.xdot_desired[1] = 0.1;
        #else:
        #    self.xdot_desired[0] = -0.75;
        #    self.xdot_desired[1] = -0.1;
        #self.xdot_desired = np.dot(self.drone.yaw_rotation(), self.xdot_desired)

        #inputCurrents = self.controller.calculate_control_command(dt, self.theta_desired, self.thetadot_desired,self.x_desired,self.xdot_desired)
        inputCurrents, acc = self.controller.calculate_control_command3(
            dt, self.xdot_desired, self.thetadot_desired.item(2))
        omega = self.drone.omega
        #thetadot_in_body()  # calculate current angular velocity in body frame

        #torques_thrust      = self.drone.torques_thrust(np.array([inputCurrents]).transpose())
        torques_thrust = self.drone.torques_thrust(inputCurrents)
        #print acc.transpose();
        # print omega.transpose()
        linear_acceleration = self.linear_acceleration(
            torques_thrust[3], self.drone.theta,
            self.drone.xdot)  # calculate the resulting linear acceleration
        omegadot = self.angular_acceleration(
            torques_thrust[0:3, 0],
            omega)  # calculate resulting angular acceleration
        #print "angular acc", omegadot.transpose()
        #linear_acceleration = self.linear_acceleration2(inputCurrents, self.drone.theta, self.drone.xdot)  # calculate the resulting linear acceleration
        #omegadot            = self.angular_acceleration2(inputCurrents, omega)  # calculate resulting angular acceleration

        omega = omega + dt * omegadot  # integrate up new angular velocity in the body frame

        #print "inputs:", inputCurrents
        #print "omega:", omega.transpose(), "omegadot:", omegadot.transpose()
        self.drone.omega = omega
        self.drone.thetadot = np.dot(
            self.drone.angle_rotation_to_world().transpose(),
            omega)  # calculate roll, pitch, yaw velocities
        self.drone.theta = self.drone.theta + dt * self.drone.thetadot  # calculate new roll, pitch, yaw angles

        #print self.drone.xdot.transpose(), self.drone.theta.transpose(), omega.transpose(), omegadot.transpose()
        #print "d", inputCurrents
        #print "a", torques_thrust[0:3,0], "b", omega, "c", omegadot

        #print "thetadot:",self.drone.thetadot
        #print("New theta",self.drone.theta)
        self.drone.xdoubledot = linear_acceleration
        self.drone.xdot = self.drone.xdot + dt * linear_acceleration  # calculate new linear drone speed
        self.drone.x = self.drone.x + dt * self.drone.xdot  # calculate new drone position
        #print "acc", linear_acceleration
        #print "theta", self.drone.theta
        #print("Position",self.drone.x.transpose())

        if (sys.platform == "skulpt"):
            import plot
            plot.plot_pose("ardrone", self.drone.x, self.drone.theta)
            plot.plot_trajectory("ardrone", self.drone.x)
            plot.plot_motor_command("ardrone", inputCurrents)

        elif (self.step_count % 50 == 0):  #save trajectory for plotting
            vel = np.dot(
                self.rotation(self.drone.theta).transpose(), self.drone.xdot)
            vel = self.drone.xdot
            #vel = self.drone.x
            #vel = acc
            #vel = self.drone.thetadot_in_body();
            #vel = self.drone.xdot;

            self.x.append(vel.item(0))
            self.y.append(vel.item(1))
            self.z.append(vel.item(2))

            ang = self.drone.theta

            self.roll.append(ang.item(0))
            self.pitch.append(ang.item(1))
            self.yaw.append(ang.item(2))
            #print self.theta_desired.item(2)
            self.cmd1.append(inputCurrents[0] - inputCurrents[2])
            self.cmd2.append(inputCurrents[1] - inputCurrents[3])
            self.cmd3.append(inputCurrents[0] - inputCurrents[1])
            self.cmd4.append(inputCurrents[3])
            self.roll_des.append(self.theta_desired[0])
            self.pitch_des.append(self.theta_desired[1])
            self.yaw_des.append(self.theta_desired[2])