One tip..about stabilization IMU

Replies to This Discussion

Permalink Reply by Ryan Beall on November 17, 2009 at 11:46am

Yeah, if you are accounting for it then you are correct this would be fine. I have recently looked at a few GPS lag smoothing techniques etc and just found that it was easier to use airspeed and it is completely satisfactory in my flight tests. It also depends upon the algorithm being run to. These assumptions are a lot tighter in a full blown extended Kalman filter. What you guys as well as I am running is closer to a fixed gain error feedback scheme. Performance increases significantly in reguards to my assumptions in the case of a EKF. However, it is interesting that you are getting better performance with even more assumptions. IE velocity is only in the 'u' direction. I find that really interesting! I would have to look closer at the code etc to understand it more and see where each of us are differeing.

▶ Reply to This

Permalink Reply by Pbreed on November 19, 2009 at 7:10pm

I've skimmed through this whole thread, and no where did anyone talk about how its done on real airplanes. I have my instrument rating and one of the tasks one must do is to fly the aircraft and navigate while using Airspeed, Altimeter Rate of turn indicator, clock and magnetic compass. Its called partial panel Now a full size aircraft is different than an RC model, the controls are usually moved by force application rather than to fixed position like an RC plane, but I find it strange that no one has discussed the similarities and differences.

Notice that my list of instruments only included ONE gyro instrument, and that was just a rate indicator. (It also did not explicitly mention the ball level/acceleration sensor that is part of a rate of turn indicator). So I propose that with a 4 axis controls, pitch controls airspeed (One might need to have a force measurement on the elevators and some trim system to be 100% analogous) , throttle controls altitude, rudder is used to keep the horizontal acceleration centered and Alierons are used to maintain the desired rate of turn. The magnetic compass is only read while straight and level. (To turn to a desired heading one turns at a fixed rate for a measured period of time)

One probably should add some rate of turn turn to pitch or throttle compensation to replace the pilots learned behavior when entering or exiting a turn. One can probably use vertical acceleration to get close as given airspeed and rate of turn one can know the acceleration needed for level flight. Has anyone done it this way?

▶ Reply to This

Permalink Reply by Ryan Beall on November 19, 2009 at 7:49pm

The reason real aircraft are not really discussed is becuase, although it is the same physics involved, the devices that are used to measure attitude in the UAV world are completly different than in general aviation. Full scale aircraft typically use electric or vaccum driven physical gyros. They have precsion similar to the mems gyros on uav's however due to a completly different reason. However ironically enough in the T-34 turbo mentor when we do aerobatics the gyros get screwed up so you have to slave them back to reality. All this does is spin the gyro at a faster rate so that it quickly resyncs back to gravity and for that reason you can't be in accelerated flight which a human in the loop can easily determing (ie keep the wings level and power constant).

As far as your "pilot learned behavior".... I love this part about UAV's because I am learning all of these in Navy flight training and get to convert them to software. What you are referring to is typically called "feed forward gains". One of the most practical and most common is exactly what you described....roll to pitch feed forward. The roll to throttle feedforward is less common because most UAV altitude/airspeed hold algorithms are much tighter than human tollerances and can quickly catch the needed power. But you can still feed this forward as you discribed. Another good feed foward is rudder/pitch to throttle.... all of the typical trim stuff you do as a pilot can be implemented as feedfoward.

Got to love this hobby specially getting to fly the T-34 and come back and tweak my autopilot code to match what I learn.

▶ Reply to This

Permalink Reply by Pbreed on November 19, 2009 at 8:26pm

Yes the horizon and heading gyros can get "screwed" by aerobatics, but the rate of turn instrument does not. One should be able to do unusual attitude recoveries with just airspeed, altimiter and this gyro. If you are in the Navy the Military has a slightly different attitude toward redundancy you probably have multiple gyro horizons in any aircraft you will ever fly and thus are unlikely to need the simplest forms of partial panel. See if you can find an older (Say 1980) general aviation instrument flight training handbook and it will talk about this is some detail.

I'm just looking to do some fixed wing autopilot stuff mostly for fun, my last RC plane was interesting....http://www.rasdoc.com/splinter/solar2004.htm

My last flying vehicle did not look much like a fixed wing aircraft....

http://www.youtube.com/watch?v=8UJDxp2gv3o

All the software for that was developed on a Trex helicopter and I want to build a general purpose autopilot board based on that and Netburner Technology (I'm founder/cto NetBurner) and may to do some Fixed wing testing in addition to the helicopter.

Paul

▶ Reply to This

Permalink Reply by Paul Bizard on November 20, 2009 at 7:31am

This flying (thing?) is pretty cool ! Now this is what I call stabilization.
When will we get to play with our UAV on the Moon or on Mars ? We should think about that for the next contest !

Permalink Reply by William Premerlani on November 17, 2009 at 1:31pm

Hi Ryan,

It would be interesting for us to put our heads together some time and figure out the similarities and differences between our techniques.

The key feature of the DCM algorithm is its explicit accounting of the nonlinear nature of the kinematics of rotation, and the full cross coupling of the interaction between the rotation vector and the attitude matrix. That way you get very accurate estimation of attitude even before you adjust for gyro drift, especially during large pitch and roll angles, where linear techniques suffer substantial error.

I am not sure if you've seen them, but Bryan Cuervo made a couple of videos that showcase the performance our team has achieved using the DCM algorithm. Bryan has a "tail-less" delta wing that is marginally stable under manual control, and which behaves like a "polyhedral, high-wing trainer" under stabilized control. Bryan took videos both from his plane and from the ground during fully autonomous waypoint flights in a butterfly pattern.

The controls execute tight, aggressive turns, with bank angles greater than 45 degrees, and airspeeds up to 45 miles per hour. Turns are made in a horizontal plane, in spite of the fact that Bryan's plane tends to dive in a turn.

Here is a video recorded from Bryan's plane. At the end, he shows the flight track.

Here is a video from the ground.

Best regards,
Bill

▶ Reply to This

Permalink Reply by Heinrich Mueller on January 4, 2010 at 8:03am

Hello,
you can have a look to this link
http://www.voidpointer.de/rsimu/index_en.html
the autor deals also with the autopilot code and made same improvements (error correction).

▶ Reply to This

Permalink Reply by python on January 15, 2010 at 10:20pm

Accelerometer signals contains deterministic and stochastic errors,
kalman filter can estimate these variables.
Integration of Acc emerges large errors and lots of noise,
not practical.

▶ Reply to This

• ‹ Previous
• 1
• 2
• 3
• 4
• Next ›
• Page