Quote:
Originally Posted by seandepagnier
I have a basic algorithm using inertial data to predict future orientation using neural networks (tensorflow). I am developing it for implementation into pypilot (free software autopilot) to be able to anticipate hopefully eliminating the lag time from measurement to reaction which potentially reduces power consumption as well as oversteering. One of the advantage is, it gets to consider what it will do with the rudder (and potentially throttle and/or sheets eventually) commands which greatly improves, the future prediction as opposed to only predicting based on past data.
From this it actually tries lots of different future commands out to figure out what it thinks it will be doing in the near future to end up with the best result.
As mentioned people with diesel dont get to try this because it is bad for their engine. For electric drives: they can really do it, and it is a clear win for electric propulsion to have dynamic throttle not only for comfort but also for improved efficiency in waves.
Wait until I plug a camera into my prediction network with a few convolution layers.
Not sure about making matters worse (unless it is exploring rather than exploiting) but essentially, yes, the larger the boat relative to the wave as well as less power this concept makes less sense.
The other use for autopilot throttle control is to hold in place which is a bit different. It could be really useful to have this function if the anchor starts dragging for example. It is also something people fishing want for whatever reason.
I envision it also in survival situations, when a rouge wave appears, the autopilot instantly recognizes and evades. Even if it uses 3% of the battery in 30 seconds: it is better than taking damage.
Interesting feedback! I think this is just an example of that particular control system failing to perform as "going manual" is just a different control system.
|
This actually does already exist in very expensive very high end systems which give full directional control using thrusters/propellers and prop walk to maneuver. Incidentally they are
linux based as well, and have been for close on a decade. They do a lot of fancy physics calculations in the background from what I understand, the key to the whole control system was in predicting what will happen next based on what is happening now. The inventor of the system was previously a physics professor. Also for setting up from memory there were about 1200 parameters one could adjust, and layers of them all depending on one another. Big things were getting the system settings for how long until a given action produced a response, and what type of drive(s) were in use.
One of the key sensors in making them
work was having an accurate read on driveshaft
RPM using an optical
sensor, I'm not sure how that would be handled with a
sails instead of under
engine. At the time I was involved they were starting to test sensors for
docking and
collision avoidance. From what I remember it was a mix of LIDAR and ultrasonic transducers, the goal was a self park feature a la Tesla now. At the time the goal was
plug and play interface with most major systems, often piggybacking on electronic over cable control throttles. From memory a complete system including adding that electronic control package for it to piggyback on ran about 75k$ for a twin engine twin thruster
boat with two
helm stations. The IMUs at the time were not as advanced as what you now use in PYpilot, but one of the key sensors was a more accurate
GPS unit using multiple receivers to give accurate heading and position, today
Furuno makes a good one, at the time they were custom building them.
The biggest problem was always getting the right data into the system. It was lovely when it was getting just the right data and an utter nightmare when not.
I have actually been thinking of reaching out to you about some ideas for electronic engine controls a la PY pilot. I think the best design is that of Kobelt at the moment, where you run short mechanical
cables of good quality and then can shift/throttle from anywhere on board via remote/electronic levers. I like this design the best because it also allows for manual override at the box in the event of either control system failure or
electrical system loss. The actual functioning of the electronic over cable system is quite simple, I suspect it could be done with
cheap stepper motors using mostly off the shelf Teleflex
parts for the mechanical side, more easily than using linear actuators. The throttle was a continuously variable position with end stops, and the arm you connect to the cable on the best systems for setup, Glendinning, had 3 holes, you picked the hole that gave the longest
motor controlled swing for a given engine mounted lever if that makes sense. The longer the motor's swing relative to the throw of the engine end, the finer control you have. The shifter had stops for forward/reverse/neutral, and the option to set a delay in milliseconds to rest at neutral between forward and reverse when shifted quickly. I once had a boat where the owner built his own back in 1976, using straight
electrical not electronic components he bought off the shelf. It was simple,
cheap and lasted 40 years.