Open Elsklivet opened 1 month ago
The application of dead reckoning to bicycles in particular seems to be sparse. There is some academic interest in using inertial dead reckoning for estimating pedestrian location (pedestrian dead reckoning, PDR) and supporting localization for autonomous vehicles, especially underwater and in flight. Some of these could be applicable here. There is a lot of mention of the Kalman filter, so that is probably a topic that needs to come back into the spotlight here. Regardless, I think the following papers about dead reckoning could be useful:
Furthermore, here are a couple of papers to read on Kalman filtering:
And finally, some about odometry in general:
As a note, a lot of the odometry papers are talking about visual odometry, which uses cameras to determine egomotion. This is irrelevant for us. We want gyroscopic and inertial odometry.
Another note, many sources on dead reckoning use encoder signals attached to wheels as a data source. We could think about that, but that's not an option right now.
There's a good chance I'm going to need to do some research on statistical concepts and some review of linear algebra concepts before I can actually start to understand Kalman filters again, which is sort of a prerequisite to understanding anything meaningful with the PDR estimation.
Trackcycle currently isn't as intelligent with dead reckoning as I think it could be. The problem is that it's mostly just checking to see if the angle between its last known trajectory and current trajectory is big enough.
This issue is not necessarily to implement any changes, but rather to research whether dead reckoning is used for longer off duty cycles in ground based transportation without set paths (like cars). Many of these likely also use gyroscope, accelerometer, and/or magnetosphere readings (although I didn't see many tapping into magnetometer readings when I initially worked on this).