F13: Quadcopter - Revision history

Proj user6: /* Acknowledgement */

2013-12-08T06:02:00Z

‎Acknowledgement

Proj user6: /* Technical Challenges */

2013-12-08T05:59:47Z

‎Technical Challenges

Proj user6: /* Testing */

2013-12-08T05:58:43Z

‎Testing

Proj user6: /* Testing */

2013-12-08T05:58:13Z

‎Testing

Proj user6: /* PWM Out */

2013-12-08T05:56:12Z

‎PWM Out

Proj user6: /* Derivative */

2013-12-08T05:54:12Z

‎Derivative

Proj user6: /* Derivative */

2013-12-08T05:53:40Z

‎Derivative

Proj user6: /* Integral */

2013-12-08T05:53:24Z

‎Integral

Proj user6: /* Proportional */

2013-12-08T05:53:10Z

‎Proportional

Proj user6: /* Data Smoothing */

2013-12-08T05:52:41Z

‎Data Smoothing

@@ Line 385: / Line 385: @@
 == References ==
 === Acknowledgement ===
-I would like to acknowledge and thank my patient wife for putting up with all the time spent on the project.
+I would like to acknowledge and thank my patient wife Sandie for putting up with all the time spent on the project.
-I would like to acknowledge and thank Brandon for loaning me his logic analyzer.
+I would like to acknowledge and thank Brandon for loaning me his logic analyzer and being there to field my software questions.
 I would like to acknowledge and thank Rob for lending me his RC expertise, and answering all of my silly questions.
-I would like to acknowledge and thank Preet for the help with the ISR issues encountered and his teaching an interesting class.
+I would like to acknowledge and thank Preet for the help with the ISR issues encountered and for teaching a very interesting class.
 === References Used ===

@@ Line 367: / Line 367: @@
 One of the largest technical challenges found was getting the PWM input ISR to not crash the processor, and to operate correctly.  After long hours of debugging it was found that there was a low level driver that was using the same timer interrupt.  Once this was solved, the development went a lot smoother, and we were able to reliably receive the commands from the remote control, and have the flight control software act upon these commands.
-The largest technical challenge found is only partly overcome.  There seem to be some issues with the data coming out of the sensors.  In an attempt to clean this up, there was some additional data smoothing added to both the gyro output and the accel output.  This did help the stability of the flight correction, but there are still issues.
+The largest technical challenge found is still being investigated.  There seem to be some issues with the data coming out of the sensors.  In an attempt to clean this up, there was some additional data smoothing added to both the gyro output and the accel output.  This did help the stability of the flight correction, but there are still issues.
 Another technical challenge found was the process of tuning the PID gain values.  This will continue to be a work in progress until the sensor reads are clean and accurate.

@@ Line 361: / Line 361: @@
 * [http://youtu.be/6wISzeOmDDU Quadcopter 2D Testing]
-The final bit of testing attempted was flight testing.  This testing showed that we need to make sure to implement the yaw control.  Once the quadcopter lifted off, it started to yaw, which made it hard to fly correctly.  Video of flight testing can be found
+The final bit of testing attempted was flight testing.  This testing showed that we need to make sure to implement the yaw control.  Once the quadcopter lifted off, it started to yaw, which made it hard to fly correctly.  Video of flight testing can be found in the section marked Project Video.
 === Technical Challenges ===

@@ Line 357: / Line 357: @@
 The angle calculation for X, Y, and Z was tested by having the data_smoothing task print to the console 1 time out of every 100 calculations.  The test consisted of holding the quadcopter in my hand and rotating it around each axis and watch the value of angle match the correct sign, as well as test if the magnitude was in the correct ballpark.
-Once the angle calculation seemed correct, the feedback loop was closed by adding in the PID with a guess of the values.  To do this testing, we attached the quadcopter between two saw horses so that it could only rotate in a single plane.  The majority of testing was of this type.  PID gain values were tested until the X axis was quasi stable.  We then switched to the Y axis and set the PID gains to the same values.  After fixing some sign errors, and other math errors, we were able to make the Y axis mostly stable as well.
+Once the angle calculation seemed correct, the feedback loop was closed by adding in the PID with a guess of the values.  To do this testing, we attached the quadcopter between two saw horses so that it could only rotate in a single plane.  The majority of testing was of this type.  PID gain values were tested until the X axis was quasi stable.  We then switched to the Y axis and set the PID gains to the same values.  After fixing some sign errors, and other math errors, we were able to make the Y axis mostly stable as well.  A video of some 2D testing can be found at the link below.
 * [http://youtu.be/6wISzeOmDDU Quadcopter 2D Testing]
-The final bit of testing attempted was flight testing.  This testing showed that we need to make sure to implement the yaw control.  Once the quadcopter lifted off, it started to yaw, which made it hard to fly correctly.
+The final bit of testing attempted was flight testing.  This testing showed that we need to make sure to implement the yaw control.  Once the quadcopter lifted off, it started to yaw, which made it hard to fly correctly.  Video of flight testing can be found
 === Technical Challenges ===

@@ Line 341: / Line 341: @@
              motor_right_sum = throttle_in + -error_in.errorY;
-Note, what was added to one half of front/back is subtracted from the other half.  Same holds true for the left/right pair.  The design also has limitations of pulse widths min and max to make sure not to disable the motors.  If the throttle in + error is larger than the max value for pulse width, it clips at the max value.  The same is true if throttle in + error is less than min pulse width.  It holds the minimum value.
+Note, what was added to one half of front/back is subtracted from the other half.  Same holds true for the left/right pair.  This is done to keep the power coming out of props the same, and to keep the elevation control only come from the throttle.
 == Testing & Technical Challenges ==

← Older revision		Revision as of 05:54, 8 December 2013
Line 323:		Line 323:
	The derivative portion of the control loop predicts future errors based on the rate of change of past errors. It can speed up error correction or can put the brakes on as well helping stop over correction if tuned correctly.		The derivative portion of the control loop predicts future errors based on the rate of change of past errors. It can speed up error correction or can put the brakes on as well helping stop over correction if tuned correctly.

		+	===== PID Loop Implementation =====
	The entire control loop for a single axis looks like this:		The entire control loop for a single axis looks like this:

@@ Line 318: / Line 318: @@
 The derivative leg of the control loop is the change in error from the last sample to the current sample divided by the time interval.
-This code looks like:  x_derivative_error = (x_proportional_error - x_prev_error)/dt;
+This code looks like:
 The derivative portion of the control loop predicts future errors based on the rate of change of past errors.  It can speed up error correction or can put the brakes on as well helping stop over correction if tuned correctly.

@@ Line 309: / Line 309: @@
 The integral leg of the control loop is a running sum of the current proportional error multiplied by delta t, which is the time since the last sample.
-The code looks like:  x_integral_error += x_proportional_error*dt;
+The code looks like:
 The integral portion of the control loop is there to account for environmental influences over the quadcopter.  An example would be an unbalanced quadcopter such that one corner is heavier than the others.  The control loop without an I would see that the quadcopter is not level and make the changes it is programed to do, but it still stays unbalances.  The integral term of the loop will continually add error to the motors until it increases such that it levels the quadcopter.

@@ Line 300: / Line 300: @@
 The proportional leg of the control loop is simply the setpoint (from the rx input) and the current state of the quadcopter from the sensor fusion section.  The equation is given by Kp*e(t) where Kp is the gain, and e(t) is the error term at time t.
-The code is very simple: x_proportional_error = setpoint.thetaX - angle.thetaX;
+The code is very simple:
 If there were no inertia, or no friction, then the P would be all that was needed.  The exact amount of error is subtracted from the controls, and would fix the error in an ideal world.  Since the real world has both of those items, a control loop needs one or both of the other types of control.

@@ Line 272: / Line 272: @@
 Due to vibrational noise, an additional lowpass filter was added before the sensor fusion and after the I2C input.  The filter acts on the raw sensor data after it has been scaled correctly.  The code for the filter is quite simple:
-smoothed_value = smoothed_value + alpha * (input - smoothed_value);
+     smoothed_value = smoothed_value + alpha * (input - smoothed_value);
 There is very little latency thru the filter, and it really seemed to help with data noise.