Quadcopter

The quad/ directory contains all code that programs the quadcopter. This includes any C libraries we have written, any HDL to program the Zybo on the quad, and the XSDK main project that runs on the Zybo.

Brief Intro

The main quad application is located at src/quad_app/.

To run this application on the Zybo, we need to implement the hardware drivers, and provide a main function. This is done in xsdk_workspace/real_quad/src.

We can also run this application on our own laptops for testing. We call this implementation the "virtual quad". This is done in src/virt_quad.

Building

Primarily Libraries and Executables

To build the libraries and executables:

make

You can also build each library individually inside their respective project directories:

cd src/<project> && make

NOTE: All build artifacts will be placed in quad/lib or quad/bin (depending on whether it is a library or executable, respectively)

XSDK Project (what runs on the Zybo)

To build the XSDK project, use the XSDK IDE. Refer to the Xilinx How-to for more instructions. Be sure to select the xsdk_workspace directory in the quad directory as your "workspace."

Testing

Automated Tests

Write tests! It makes automating things so much easier.

Run the unit and functional tests:

make test

And glad to hear you are interested in writing your own unit tests! Look at the README and examples in our simple testing library at src/test.

Manually testing the hardware interface

Of course, we cannot run automated tests on code that needs the Zybo. But we have manual tests that you can use to test each specific driver in the hardware interface.

Look in xsdk_workspace/real_quad/src/hw_impl_zybo_tests.c for instructions. Ideally, you would run these tests from the XSDK IDE.

Other Documents

Zybo Pinout Assignments
How to use Xilinx software tools

XSDK FYIs

Definitely first read the Xilinx How-To.

The XIlinx SDK has a few quirks that are important to watch out for:

1. From the documentation, if you abort program execution while at a breakpoint inside an interrupt handler, when re-running the program, interrupts don’t fire. You have to do a hard reset of the board (cycle power) to have interrupts work again.
2. After doing a git pull or git checkout, refresh the files by right-clicking on the project in the sidebar and clicking "Refresh"
3. The project does not detect changes in header files, so if you modify a .h file, you should do a clean before re-building, otherwise you may experience unexpected behavior. We got into the habit of always doing a clean before a build whenever creating code that will be put on the quadcopter.

Controller Network (Control algorithm)

First read the documentation for the computation graph library to understand how the graph computes functions from a directed graph.

To visualize the default control network, from the quad folder, run make diagram with graphviz installed, and an image of the control network will show up as network.png in the src/gen_diagram folder. To see the autonomous controller, you can change the call at the bottom of control_algorithm_init from connect_manual() to connect_autonomous() before running make diagram. Just be sure to change it back to connect_manual() before the final build. Below is a simplified version of the autonomous controller that shows the control network for autonomous flight using VRPN data. (Unused blocks relating to manual flight and optical flow have been removed, as well as Ts_IMU and Ts_VRPN, which are blocks that keep track of the sampling period)

One potential confusing point to take note of is the difference between "(X/Y) Vel PID" and "(X/Y) Vel" blocks. "(X/Y) Vel" is just a PID controller that has Kd=-1, which results in calculating the derivative of position to provide velocity. For clarity, it would be a good idea to create an actual differentiation block.

• Note our use of derivative on value, not error

Optical Flow

The current (end of may17-16) state of optical flow is that it can be used autonomously be uncommenting all four defines at the top of control_algorithm.c. It is relatively stable, but drifts over time. Setpoints do work, but yaw needs fixing (See issue #23 (closed)).

We are using the PX4Flow optical flow sensor. Read more about it here and here. The source code for the firmware can be found on Github.

Improvements to be made

• Compensate for rotation of sensor. The PX4Flow has the ability to enable gyroscope compensation (The firmware must be re-flashed for changes to persist), but when we tried it, the output was much worse. We then implemented our own gyroscope correction, using the complementary-filtered pitch and roll, since they will not drift over time (Using raw gyro would result in drift in position because of gyrosope angle drift). The actual data didn't look much worse, and for large, slow movements, the gyroscope compensation seemed to help prevent incorrect measurements, but it always made flight worse, probably beause of high-frequency noise. Using the complementary filter might be the cause, since it essentially high-passes the accelerometer readings, which are very noisy. We then also tried putting a low-pass filter on phi_dot and theta_dot, but it didn't help flight. Possibly the delay added by the filter caused the correction to not align with the actual movements.
• Possibly just switch to a better optical flow sensor. This one has not had any developments in the past couple years, is hard to buy, and is poorly documented. The best documentation is to actually look in the source code, because we have found multiple discrepancies in the documentation.