Initialization Plugin

The Initialization plugin is a factory for two types of initializers. This plugin will generally be leveraged by the Orchestration plugin to come up with an initial solution to initialize other plugins.

Common Functionality

All types of initializers have a few methods in common:

• request_motion_needed(): check whether this initializer needs motion, no motion, or doesn’t care.

• request_current_status(): check the status of the initialization (e.g. working on a solution, succeeded, failed, etc.)

• process_pntos_message(): consume data to calculate an initial solution

Both types also have a request_solution() method, but with different return times.

InertialInitializationStrategy

This type of initializer calculates an alignment (i.e. an initial PVA) for an inertial mechanization. Usually this will be used by the Orchestration plugin in conjunction with an Inertial Plugin.

This could be as simple as a hard-coded manual alignment, but will usually be something more complex that uses inertial and position data to calculate a PVA.

Once the initializer has received sufficient data and has successfully completed its calculation, the user can get the alignment via request_solution().

EwcInitializationStrategy

This type of initializer will generally be used by the Orchestration plugin to calculate an initial estimate and covariance to initialize its filter states with. Its unique method is request_solution(). This initializer is more open-ended and the data it needs will vary with respect to the type of states it is calculating an initial estimate and covariance for.

Cobra Initialization Plugins

Cobra currently contains three InertialInitializationStrategy implementations. Each of them provides a PVA, but they vary in other aspects:

• whether or not they provide inertial errors

• the algorithm used to produce the initial PVA

• their data and config requirements

See below for more details on each.

Cobra Tutorial-Level Manual Initialization Plugin

The TutorialInitializationPlugin provides the ManualInitialization, which serves as a reference for a very simple implementation of a InertialInitializationStrategy. Rather than calculate the initial PVA, it grabs a set of precalculated fields from the registry as its configuration. As a result, this initializer does not require any supplementary data to calculate its solution and can immediately provide an alignment.

While this initializer was designed for educational purposes, it could be used in a real system. For example, it could facilitate a transfer alignment.

Cobra Standard-Level StaticAlignInitializationPlugin

This plugin provides the StaticAlign initializer. This initializer requires:

1. A static period of no platform movement, typically around two minutes

2. A stream of inertial data from an IMU of at least navigation-grade or higher gyroscopes and tactical-grade or higher accelerometers

3. A stream of position data

The position solution is straightforward, as it is just pulled from 3.. The velocity solution is similarly straightforward, as the assumption in 1. implies it can assume zero velocity. Attitude is more complex.

First, IMU data is accumulated in order to average out any white noise. Thus, if requirement 2. is not met, time-correlated biases can introduce large errors into the calculated attitude.

The roll and pitch are solved for by using the accumulated accelerometer data to level the platform by calculating the gravity vector. Heading is calculated by using the accumulated gyroscope data to observe the rotation of the earth, also known as gyrocompassing.

Inertial biases are not calculated as part of this algorithm. Because of requirement 2., these are assumed to be negligible.

Cobra Standard-Level ManualHeadingAlignInitializationPlugin

This plugin provides the ManualHeadingAlign initializer. This alignment algorithm is similar to StaticAlign, but is suitable for lower-grade IMUs. Because of that, it requires a user-provided heading to compensate. In all, this initializer requires:

1. A static period of no platform movement, typically around several seconds

2. A stream of inertial data from tactical-grade IMU or higher (mainly because any lower grade inertial will be challenging to mechanize from in the first place)

3. A stream of position data

4. A user-provided heading

3. provides the initial position, 1. provides an assumption of zero initial velocity, and 4. provides the initial heading.

Similar to the StaticAlign algorithm, IMU data is accumulated to average out white noise. Then the accumulated accelerometer data is used to calculate the gravity vector in order to back out roll and pitch.

Differently than the StaticAlign algorithm, this algorithm also uses the accumulated IMU data to calculate inertial biases.