This library was written for MicroPython but can be used on any device capable of running Python V3.4 or above (V3.5 or above for the asynchronous version). In its default mode as described in the main README it uses specific MicroPython calls to handle precision time values.

On other platforms the application must construct a Fusion instance in a way to prevent it doing this. The application must provide a precision time value representing the instant that the vectors were acquired. It must also provide a function capable of differencing a pair of these; this is typically merely subtraction and scaling.

Main README

1. mpudata A set of data captured from a Pyboard performing calibration followed by movement around each axis in turn.
2. capture The program used to create the above dataset.
3. fusion_r_syn Synchronous test program using the dataset.
4. fusion_r_asyn Asynchronous test program using the dataset.

The test programs perform a calibration phase during which the device was fully rotated around each orthogonal axis. They then display the data as the device was partially rotated around each axis in turn.

Test programs include comments indicating how synchronous and asynchronous applications might be developed.

This caters for the case where a device acquires IMU data and returns that data to another platform performing the fusion. The remote device may run any OS (or none), run any programming language and return data in any format. It must return two or three vectors (depending on whether it is a 6DOF or 9DOF device) and a timestamp indicating when the vectors were acquired.

The platform performing the fusion may run any Python version as described in section 1.

The remote device must return a timestamp along with the IMU data, representing the time when the IMU data was acquired. This avoids inaccuracy in the fusion calculation caused by latency in the communications link. The format of the timestamp is arbitrary, but the user program must provide a function for calculating the difference between two timestamps.

User code on the device performing the fusion must convert the data received from the remote to the format acceptable to the fusion module, namely

[[ax, ay, az], [gx, gy, gz], [mx, my, mz], timestamp]

where an, gn, mn are accelerometer, gyro, and magnetometer vectors. The axes of these readings must match.

This needs to supply data at a rate dependent on the application. To track fast moving platforms the repetition rate may need to be as fast as 10ms.

Format is user-dependent. If you have a choice consider a JSON encoded list as detailed above: since this matches the call signature of the fusion code it is trivial to convert.

The library requires a function which accepts as arguments two timestamps and returns a time difference in seconds as a floating point value. In the case where the acquisition and fusion devices both run MicroPython, the acquisition device can transmit values from utime.ticks_us() and the differnencing function passed to the Fusion constructor would be

lambda start, end: utime.ticks_diff(start, end)/1000000

In other cases if timestamps roll over modulo N, the differencing function must accommodate this. In more usual case where times are floating point values without rollover it may simply be a matter of scaling:

def TimeDiff(start, end):  # Timestamps here are in μs
    return (start - end)/1000000  # Scale to seconds

If standard Python's time.time() provides the timestamps no scaling is required. The function can reduce to:

lambda start, end: start-end

If magnetometer calibration is to be used the fusion program needs some form of input to tell it that manual rotation of the target about each axis is complete.

In the case of remote applications this might be sourced by user input to the remote or to the fusion device. In the test data a button on the remote was pressed and this is stored in the file, simulating communication with the host.

The Fusion constructor should be instantiated as follows:

from fusion import Fusion
fuse = Fusion(TimeDiff)

The arg is the user supplied time differencing function.

For details of the method of updating the fusion data and retrieving the angles see the main document section 2.

This encapsulates data acquisition from the target in the generator function gdata. The resultant generator get_data reads a line of data from the target, decodes the JSON string, and yields a list of form

[[ax, ay, az], [gx, gy, gz], [mx, my, mz], timestamp]

Code is included to handle the calibration mode: in the test file completion is flagged by a special record created when a button on the device was pressed. The handling of the StopIteration exception is required because the data source is a file of finite length.

The Fusion constructor should be instantiated as follows:

from fusion_async import Fusion
fuse = Fusion(read_coro, TimeDiff)

The first argument is a coroutine. For 9DOF sensors this must be designed to return three (x, y, z) 3-tuples for accelerometer, gyro, and magnetometer data respectively followed by a timestamp. In the case of 6DOF sensors it returns two 3-tuples for accelerometer and gyro followed by a timestamp. The coroutine must include at least one await statement to conform to Python syntax rules.

The second arg is the user supplied time differencing function.

For details of the method of updating the fusion data and retrieving the angles see the main document section 3.

This encapsulates data acquisition from the target in the class GetData. In principle the constructor initiates communications with the (local or remote) IMU. The asynchronous read method acquires a record from the IMU, converts it to the correct format, and returns a list of form

[[ax, ay, az], [gx, gy, gz], [mx, my, mz], timestamp]

Code is included to handle the calibration mode: in the test file completion is flagged by a special record created when a button on the device was pressed. Further code handles the fact that the test fileis of finite length.

Main README