Kalman_filter_carla

Kalman filter for self driving cars using gnss and imu data. Data was collected from the CARLA open source simulator.

---

Software used

• Ubuntu 18.04
• Carla 0.9.11
• Python 3.7 API
• Python 3.6 interpreter
• VS Code for editing
• IPython 5.5.0

---

Data collection (CARLA agent)

• A class, agent is created that handles data collection from carla. Code
• Creating the class creates the carla client, blueprint library and world objects. See the carla documentation for details.
• spawn_vehicle spawns a vehicle at a location, based on the arguments. The vehicle is named vehicle
  • argument : index : The index of pre-defined location in the world at which to spawn the vehicle. If not mentioned, a random location is choosen.
• spawn_gnss spawns a gnss sensor, named gnss_sensor
  • argument : per : Period for refreshing gnss data
  • argument : std_dev : Standard deviation of gnss data (in degrees for latitude and longitude)
• reg_gnss_callback registers the passed function as a callback that will be executed when the gnss data is available.
  • argument : callback : The function that will be called with the gnss data as the argument, when data is ready.
• spawn_imu spawns an IMU sensor, named imu_sensor
  • arguemnt : per : Period for refreshing imu data
  • argument : accel_std_dev : Standard deviation for accelerometer
  • argument : gyro_std_dev : Standard deviation for gyroscope data.
• reg_imu_callback registers the passed function as a callback that will be executed when the imu data is available.
  • argument : callback : The function that will be called with the imu data as the argument, when data is ready.
• gnss_to_xyz converts gnss data passed into xyz coordinates in the current map, by comparing to the geographic coordinates(latitude, longitude) of the map centre to geographic coordinates in the passed gnss data.
  • argument : data : The gnss data to be converted
  • returns : [x, y, z] : Standard cartesian coordinates

---

Kalman filter interface

For controlling and giving inputs

• At initialization, we give the following details to the constructor:

  • initial_state
    • A 5x1 column vector, with the format [x_pos, y_pos, yaw_angle, x_vel, y_vel]
    • This is considered to be perfect data, i.e with 0 variance.
  • accel_var
    • Variance of accelerometer data
  • yaw_var
    • Variance of gyroscope yaw angular velocity estimate
  • meas_var
    • Variance of measurements (gnss data)

• For the prediction stage, to feed with inputs (x and y acceleration from acceleromter and yaw angular veloicty from gyroscope), pass the following

  • inp
    • A 3x1 column vector with the format [x_accel, y_accel, yaw_vel]
  • time
    • Current time in seconds

• For the updation stage, to feed with measurements (x, y data from GNSS sensor), pass the following

  • measurement
    • A 2x1 column vector with the format [x_pos, y_pos]
  • time
    • Current time in seconds

For getting outputs and viewing previous estimations, use the following data members

• state object stores state in same format as initial_state arguement in the constructor. Updated in both updation and measurement stages.
• states list stores tuples of state vector ,timestamp, and 0 if state was added during updation and 1 if state was added during measurement
• covar object stores the covariance matrix (5x5)
• covars list stores covariance matrices after each udation or measurement, like the states list

---

References

1. Carla documentation
2. GNSS latitude and longitude to x-y coordinate conversion