MARCY 1 ATTITUDE HOLD 4 U
Attitude hold on Marcy 1 seems to be a disaster. In the wind, She seemed to actually stabilize with the attitude hold on. When She got near the trees, that must have confused the thermopiles & caused her to bank away from the trees, leading to a seemingly stable position in the middle of the fairway.
In 1 short break between storms, the attitude hold was a disaster.
Cyclic seems to be responding properly but is lagging by 1/2 a complete phase.
Lacking thermopile calibration, currently hard coding the gain & sign. Relying on manual takeoff, taking neutral from whatever the attitude is when the autopilot is engaged.
There is a way to sense when the magnetometer flips over due to pitch. Detect when its gain gets below a minimum. Record thermopile pitch at that point. When gain rises above minimum, if thermopile pitch is reversed, phase is reversed. Theoretically it couldn't reverse pitch & get mag above minimum at the same time. What if it really reverses pitch in 1 revolution where mag is below minimum?
The magnetometer can sense pitch with the thermopile for roll only. The thermopile is a much better pitch estimator than the mag.
Marcy 1 uses throttle modulation for attitude control & maximum flight time, but it reduces thrust by 1/2 & is worthless in wind. 1 option is to overdrive the motor with a 3S or a longer propeller. Also still searching for an actuator which can survive thousands of cycles & is light enough. Not happy about giving Canadia $20 for a crummy coil.
Throttle modulation may also reduce flight time if it causes the motor to run outside its optimum RPM & causes inertia changes in the propeller's airflow.
MARCY 2 CONTINUES AS A TRI ROTOR
With the weather closed back in, desired a more weather proof Marcy vehicle.
Rebuilt her IMU using the straight LISY300AL's from last year's buying spree & a 3/4" balsa cube. The ancient IDG300's just have too much trouble initializing & are too unstable to bother with. Can't see them ever being used for anything now.
Went back to radio foam instead of Marcy Foam #2009 because of the lack of space.
Unfortunately, a 1st flight test revealed the need for Marcy Foam. A bit too much vibration. Despite using the same motors & battery, the smaller Marcy 2 configuration has a lot more power than Vika 2. Maybe the extra cable, CF tubing, & plywood is significant.
RETURN OF STILL PHOTO TIMELAPSES
These are made in 1920x1440. If computing power wasn't so limited, they could go up to 3000x2300.
Now some notes on still photo timelapses.
Create a JPEG list of all the photos: ls `pwd`/*.jpg > list.jpg
Add the file format to the beginning of list.jpg. The format is JPEGLIST, framerate, width, height:
JPEGLIST
4
3648
2736
Create a Cinelerra project with framerate 4, canvas size 2048x2048, YUV color.
Load list.jpg into Cinelerra
Shrink camera so the images fit on the canvas.
Disable track playback.
Rewind to frame 0.
Attach motion.
Set motion thus:
Set the following:
Track translation & track rotation enabled.
Calculation: Save coords to /tmp
Action: Do nothing
Previous frame same block enabled
Draw vectors enabled
Position settling speed: 10
Maximum position offset: 50
Translation search steps: 256
Translation direction: Both
Rotation search radius: 10
Rotation search steps: 4
Rotation center: 0
Max angle offset: 50
Rotation settling speed: 10
Enable track playback.
Set translation block size so the block covers 1/4 of the scaled image & not the black area.
Adjust Block X & Block Y so the block is over an area which always has detail.
Set translation search radius to cover entire area of scaled image but not black area.
Remove all /tmp/motion0* files on every render node.
Render using renderfarm to compute the motion vectors. Use Quicktime MPEG-4 quantization 5.
Disable track playback.
Copy all the /tmp/motion0* files from every render node to the master node.
Set the following in Motion:
Calculation: Load coords from /tmp
Action: Stabilize Pixel
Draw vectors disabled.
Attach Time Average below Motion.
In Time Average set the following:
Frame count: 1
Restart for every frame: off
Don't buffer frames: on
Replace: on
Threshold: 1
Border: 4
In Settings->Set Format set Color model to YUVA-8 bit
Rewind & enable track playback.
Render to Quicktime JPEG maximum quality WITHOUT the renderfarm. That's the rough stabilization print. Motion must be rendered contiguously to accumulate the motion history. Next comes the smooth stabilization.
Disable track playback.
In Settings->Set Format set Color model to YUV-8 bit, Frame rate to 24.
Import the Quicktime movie in a new track & play through it to determine the new range of motion.
Disable playback in the new track.
Rewind to frame 0.
Attach Motion.
Set the following:
Track translation enabled.
Track rotation disabled.
Calculation: Save coords to /tmp
Action: Do nothing
Previous frame same block enabled
Draw vectors enabled
Position settling speed: 3
Maximum position offset: 50
Translation search steps: 256
Translation direction: Both
Enable playback of the new track only.
Reduce translation search radius to cover only the new range of motion.
Remove all /tmp/motion0* files on every render node.
Render using renderfarm to compute the motion vectors. Use Quicktime MPEG-4 quantization 5.
Disable track playback.
Copy all the /tmp/motion0* files from every render node to the master node.
Set the following in Motion:
Calculation: Load coords from /tmp
Action: Stabilize Subpixel
Draw vectors disabled.
Attach Time Average below Motion.
Use the same Time Average settings as the old track.
In Settings->Set Format set Width: 1920 Height: 1440
Rewind & enable track playback.
Render to the final output format WITHOUT the renderfarm. This is the smooth stabilization print.
This is our final timeline.
Comments
I suppose this is obvious to most people, but I hadn't really thought about it until then.