- Code related to "smoothing" is removed. The implemented "EKF smoother" is a half-baked approach which doesn't correspond to a well define optimal algorithm.
- Filter classes were replaced by more straightforward plain functions.
- All data is now indexed by time in seconds, not IMU stamp. This is a more intuitive and convenient approach which directly corresponds to practical algorithms. This also allows to more optimally process measurements not precisely synchronous with IMU samples, which again corresponds to real-world data.
- All data now is stored in pandas DataFrame with consistent column naming.
- Transform utilities were extended and improved.
- Everything is designed to work better together.
- Better test coverage.
- Cython was replaced by numba to avoid the need of compilation.
- HTML documentation and examples are reworked and cleaned up.