OpenMarine

Full Version: Is rudder feedback used to improve behavior?
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Is the rudder sensor that can be connected to the motor controller, only for detecting movement and end of travel of the actuator? Or is it also used in algorithms to improve the steering behavior of the autopilot? I understand that the commercial autopilots use this rudder feedback also for the improvement of their autopilot behavior (i.e. knowing where the helm is at, and therefore steering with the right amount of angle to cope for the xte).

Does pypilot do the same?

regards,
Vic
(2020-01-13, 10:48 AM)Vic Wrote: [ -> ]Is the rudder sensor that can be connected to the motor controller, only for detecting movement and end of travel of the actuator? Or is it also used in algorithms to improve the steering behavior of the autopilot? I understand that the commercial autopilots use this rudder feedback also for the improvement of their autopilot behavior (i.e. knowing where the helm is at, and therefore steering with the right amount of angle to cope for the xte).

Does pypilot do the same?

regards,
Vic
My 2 cents ... is there an AI possibility to work on prediction? Am not SW developer but Tensorflow could be used. Correct?
The basic autopilot only uses rudder angle to prevent further movement. This pilot is proven to work.

Using the rudder angle input could help eliminate integration errors, but it's really difficult because many types of rudder feedback might be used and there can be significant play in the system between the motor and rudder feedback and actual rudder angle. There are additional gains that must be tuned for how autopilot moves relative to rudder feedback.

The errors from integration of rudder position are about an order of magnitude less than errors from waves and other interactions with the wind. For this reason the autopilot would only be slightly improved at least for cruising boats at displacement speed, from knowing absolute angle and only if tuned properly, otherwise it might actually be worse or not matter using the approach, but nonetheless it is implemented in pypilot as the absolute pilot. If the rudder feedback fails it should switch back to the basic pilot.

Using AI with prediction makes rudder feedback a lot more plausible because all the additional gains (which don't fully model the system anyway) are not needed, and the network can react (or choose to ignore) the rudder angle sensor depending on how useful it actually turns out to be. This is not working just yet...