Extended Kalman Filter (EKF)¶
Copter and Plane can use an Extended Kalman Filter (EKF) algorithm to estimate vehicle position, velocity and angular orientation based on rate gyroscopes, accelerometer, compass, GPS, airspeed and barometric pressure measurements.
The advantage of the EKF over the simpler complementary filter algorithms (i.e. “Inertial Nav”), is that by fusing all available measurements it is better able to reject measurements with significant errors. This makes the vehicle less susceptible to faults that affect a single sensor. EKF also enables measurements from optional sensors such as optical flow and laser range finders to be used to assist navigation.
Current stable version of ArduPilot use the EKF2 as their primary attitude and position estimation source with DCM running quietly in the background. If the autopilot has two (or more) IMUs available, two EKF “cores” (i.e. two instances of the EKF) will run in parallel, each using a different IMU. At any one time, only the output from a single EKF core is ever used, that core being the one that reports the best health which is determined by the consistency of its sensor data.
Most user should not need to modify any EKF parameters but the information below provides some information on those parameters that are most commonly changed. More detailed information can be found on the developer EKF wiki page.
Should the EKF2 or EKF3 be used?¶
In general we recommend users stick with the EKF2 but there are some cases where the EKF3 should be used. Below is a list of advantages of each:
- EKF2 is used by default for most users, has had the most testing and is considered the most stable
- EKF2 can accept external position estimates from Vicon systems or ROS SLAM (HectorSLAM, Cartographer, etc). EKF3 will get this feature once this PR is merged
- EKF3 should be used on tailsitters or any other vehicle that spends a significant amount of time pointing directly up or down. The reason is that the EKF2 only estimates accelerometer Z-axis offsets while EKF3 estimates for all 3 axis
- EKF3 accepts some newer sensor sources including Beacons, Wheel Encoders and Visual Odometry
- EKF2 estimates gyro scale factors but the EKF3 does not. In general this is not important because gyro scale factors are nearly always very close to 1.0. This may be important for vehicles that spin very rapidly
Choosing the EKF and number of cores¶
AHRS_EKF_USE: set to “1” to use the EKF, “0” to use DCM for attitude control and inertial nav (Copter-3.2.1) or ahrs dead reckoning (Plane) for position control. In Copter-3.3 (and higher) this parameter is forced to “1” and cannot be changed.
AHRS_EKF_TYPE: set to “2” to use EKF2 for attitude and position estimation, “3” for EKF3.
- 1: starts a single EKF core using the first IMU
- 2: starts a single EKF core using only the second IMU
- 3: starts two separate EKF cores using the first and second IMUs respectively
Plane and Rover will fall back from EKF2 or EKF3 to DCM if the EKF becomes unhealthy or the EKF is not fusing GPS data despite the GPS having 3D Lock. There is no fallback from EKF3 to EKF2 (or EKF2 to EKF1)
Using the parameters above it is possible to run up to 5 AHRSs in parallel at the same time (DCMx1, EKF2x2, EKF3x2) but this can result in performance problems so if running EKF2 and EKF3 in parallel, set the IMU_MASK to reduce the total number of cores.
Commonly modified parameters¶
EK2_ALT_SOURCE which sensor to use as the primary altitude source
- 0 : use barometer (default)
- 1 : use range finder. Do not use this option unless the vehicle is being flown indoors where the ground is flat. For terrain following please see copter and plane specific terrain following instructions) which do not require changing this parameter
- 2 : use GPS. Useful when GPS quality is very good and barometer drift could be a problem. For example if the vehicle will perform long distance missions with altitude changes of >100m.
EK2_ALT_M_NSE: Default is “1.0”. Lower number reduces reliance on accelerometers, increases reliance on barometer.
EK2_GPS_TYPE: Controls how GPS is used.
- 0 : use 3D velocity & 2D position from GPS
- 1 : use 2D velocity & 2D position (GPS velocity does not contribute to altitude estimate)
- 2: use 2D position
- 3 : no GPS (will use optical flow only if available)
EK2_YAW_M_NSE: Controls the weighting between GPS and Compass when calculating the heading. Default is “0.5”, lower values will cause the compass to be trusted more (i.e. higher weighting to the compass)
As mentioned above, a more detailed overview of EKF theory and tuning parameters is available on the developer wiki’s Extended Kalman Filter Navigation Overview and Tuning.