Segway And Input Filtering


[Kerry] is building a segway type vehicle for a senior project. Though it doesn’t appear to be done yet, there is some good work finished. As with any balancing bot, it relies heavily on input from a gyroscope, and at least in this case, an accelerometer. That data can be very messy. The sensitivity of the sensor means that people often end up with vibrating or drifting in their projects. [Kerry] has put some work into finding a method of smoothing that results in fast, accurate, but smooth reaction. You can see a video of some comparisons between methods after the break.


25 thoughts on “Segway And Input Filtering

  1. Looks like there’s not really much in it. The left and right ones just look like low-pass filtered versions of the kalman output.

    Also, I hate it when people drag 30 second videos into 3 minutes so they can play their favourite music over it.

  2. Because the song is commercial, youtube has detected it, and puts a little “buy in itunes” ad on the youtube page. This doesn’t appear in embedding, so it doesn’t make a difference here, but the ad deal isn’t available in every country – I guess they have to make deals with each country’s respective royalty collection boards.

    If the video maker had simply not chosen that song, this wouldn’t be an issue…

  3. the video means absolutely nothing without the model and noise parameters used.

    also, what is the person trying to do? estimate the orientation of the board? or smoothen the orientation measurement?

    if the former, they would need ground truth to assess how well each filter does. Ground truch is hard to obtain in this case, though, so it may be better to implement the controller for the segway and simulate it with each filter to compare them.

    if the person just wants to smoothen the orientations, a simple low pass with the right parameters would do just fine — at a smaller computational cost.

  4. A completely useless video that explains and demonstrates nothing.

    I wonder when we will get a good demo of it because wild changes like that is worthless. the only VALUABLE 6 seconds of the video are in the middle where you see the input going wonky and the filters stopping it, everything else was 100% useless.

  5. Will someone with actual knowledge of nonlinear feedback system design stop all this nonsense and just publish a working mathematical model already! This isn’t the kind of thing you can get working well by hacking with it. You need the proper theoretical foundation.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.