Category: Software

We have decided to use Travis for continuous integration builds of our open source repositories. Travis is automatically building the code on all branches and pull requests, which gives all developers that wants to contribute to the project, the possibility to see that their code passes the build. The current status of the latest build on the main branch, is visible through the icon in the readme in github, or on the Bitcraze page at travis.

travis-ci

The projects we have added so far to travis are written in C or python. The C projects for instance, must be compiled with special compilers for the processors used in the crazyflie which adds some extra complexity. We have created a docker image (bitcraze/builder) with the tools needed, to make life easier for developers. If you use the image when developing, there is no need to install tools locally, and the same image is used in travis builds, so you know you will get the same results as the CI-server. This also removes the problem of tools with different versions (and results) in the development- and build environment.

To use the image you can for instance type

docker run --rm -v ${PWD}:/module bitcraze/builder make

Event though it is awesome to be able to create a well known build environment through a docker container, we feel that too much typing is needed to execute a simple make.  To solve that problem we are looking at the possibility of creating a toolbelt that will handle that for you. More information on that later on, for now developers will have to find their own solutions through scripting, aliasing or other means.

Obviously you need Docker to use this image. If you have not tried it out yet, take a look at www.docker.com.

We are aiming for automated testing of our code, and even though we have a lot of work to do, we have taken the first baby step. For the moment, firmware projects are simply compiled and linked to ensure that the code is coherent. Projects that support both crazyflie 1 and 2 are built in both flavours to avoid problems for developers that might only use one of the platforms.  The python client project is only checked for PEP8 compliance, but we are looking at how to unit test. Any input from the community is welcome!

Happy hacking!

During the last week we’ve taken a big step, moving to Python 3! The reason for the move is that Python 3 is becoming broadly adopted and it has more features that we want to make use of. Also 3 > 2. This post will explain a bit of what we did, some of the problems we encounters and the current status. The numbers 2 and 3 will be thrown around a lot in the text, but to precise we’re talking about versions 2.7+ and 3.4+ (even more precise it’s been tested on 2.7.9 and 3.4.3). The next release of the client will run on Python 3, but if you want to test it now just clone the development branch on GitHub.

Status

If you have developed applications using the API and Python 2 then you might be getting a bit worried right about now. The compatibility for both Python 2 and 3 will be kept for most things, except for the client:

This will be compatible with both 2 and 3:

  • The Crazyflie Python API (everything in lib/cflib)
  • The examples for the Crazyflie Python API (everything in examples)
  • The ZMQ server using the Crazyflie Python API (bin/cfzmq)
  • The Crazyflie command-line bootloader (bin/cfloader)

But the main clients will only have Python 3 compatibility:

  • The Crazyflie Python client (bin/cfclient)
  • The Crazyflie Python headless client (bin/cfheadless)

API Examples

While doing the porting we’ve also added more examples to cover more of the Crazyflie Python API. In order to keep 2/3 compatibility for the API it’s important to be able to test it easily with the different versions. We are having unit-tests on the TODO-list, but until then we’ve been using the API examples to test. All the examples should run with both Python 2 and 3. It’s also a good thing with more examples showing how to use the API…

Porting and compatibility

The approach we used was to first run the 2to3 utility to automatically to as much as possible of the porting. After that we had to fix the rest of the errors manually and also maintain the dual 2/3 compatibility of the API.

In our previous implementation we made use of strings to store binary data that we were sending/receiving. But because of incompatibilities between Python 2 and 3 this didn’t fit very well. To make things neat for the API we found a container where we could store bytes that works with both Python 2 and 3, the bytearray. Even though we use the same type, there’s still some subtle differences in usage between the versions. After doing some testing we found ways where the syntax was the same for Python 2/3.

First of all bytearrays can be created from a string, tuple or list. When indexed by the [] operator it will give you the value of each byte.

>>> d = bytearray([i for i in range(10)])
>>> d
bytearray(b'\x00\x01\x02\x03\x04\x05\x06\x07\x08\t')
>>> d[5]
5

The main point is getting something meaningful out of the bytearray when doing the communication, here’s a few examples:

Unpacking a byte, an integer and a word from the first 7 bytes (little endian)

>>> struct.unpack("<BIH", a[:7])
(0, 67305985, 1541)

Getting a string from a subset of the data can be done by using decode and the char-set to use for decoding. We use ISO-8859-1 since the Crazyflie does not support Unicode (yet?).

>>> d = bytearray([i for i in range(97,100)])
>>> d
bytearray(b'abc')
>>> d.decode("ISO-8859-1")
'abc'

You can also easily get a tuple or a list:

>>> list(d)
[97, 98, 99]
>>> tuple(d)
(97, 98, 99)

And you can also concatenate:

>>> d + d
bytearray(b'abcabc')

And find a byte:

>>> d.find(bytearray((98, )))
1

But there’s also a few things we couldn’t get to work in a good way and have to check which version we’re running and execute different code, like the queue import that has changed name.

if sys.version_info < (3,):
    import Queue as queue
else:
    import queue

Another problem we haven’t solved is creating a bytearray from a string, so it’s also

if sys.version_info < (3,):
    self._data = bytearray(data)
else:
    self._data = bytearray(data.encode('ISO-8859-1'))

As for the client code that was ported to Python 3 without keeping the backwards compatibility there wasn’t any big issues. The biggest change was the PyQT4 API where there’s a few things that have improved when placing custom Python data in GUI objects. Before QVariant was used for this. You would create a QVariant object that wrapped the Python object. To get data out from the QVariant again you would have to explicitly say what type it had by calling the correct function (like toInt()). Now this is a lot smoother. QVariant has been skipped and you just use the Python type directly.

For more information have a look here where we found a lot of useful tips. Don’t hesitate to leave a comment if you think we could have done things differently or if you have any tips!

What’s not working

There’s still a few things we’re not sure how to fix and we have to look into it a bit more. These are:

  • There doesn’t seem to be any Python 3 bindings for the Leap Motion. According to this it’s possible to build the bindings yourself.
  • The Python 3 bindings to Marble for the GPS tab hasn’t been investigated yet

PEP-8

On a side note we’ve started using Travis CI (more on this next week) and will start creating unit-tests for the Crazyflie Python  API. As a first step we’re running PEP-8 on all the code. This will be checked automatically for all commits and pull-requests.

 

A new version of the Crazyflie PC Python client has been released, version 2015.08. It’s been a while since the last release of the client so there’s a long list of changes, including lots of fixed bugs. The main new features are:

  • Student/Teacher mode:  It’s possible to use two input devices, where one can take over control from the other. This can be used for teaching or for working with a computer auto-pilot (doc)
  • Control the LED-ring from the Flight Tab:  Now it’s possible to turn on/off the headlights directly with a click and to select the LED-ring effect from a drop-down (doc)
  • New LED-tab to set custom patterns and intensity: Enables the user to individually set the color of each LED as well as the intensity of the LED-ring
  • ZMQ access to LED-ring memory and parameters: Write patterns for the LED-ring or set/get parameter values from an external application using JSON and ZMQ (doc)
  • ZMQ input device: Simulate a joystick by sending axis values in JSON via ZMQ. This can be used to implement a computer auto-pilot using for instance the Kinect (doc)
  • Switched from PyGame to PySDL2 on Mac OSX/Windows and native input device on Linux
  • WiiMote support

The next step after the release is to shape up the code a bit, so we’ve started using Travis for building and continuous integration. The long term goal is to run Flake8 and unit-tests on the code, but we still have a bit to go. The way we’re working towards this is by slowly enabling more and more checking in Travis, fixing one type of errors at a time.

 

DWM1000 nodes
Last hacking-Friday we have had some time to put together the DWM1000 boards we ordered during the summer. The DWM1000 from Decawave is an ultra-wide-band ieee802.15.4 radio transceiver that can very precisely timestamp packets arrival and departure. More simply it means that it is a standard and it can be used to implement a real time local positioning system: this could be really handy for the Crazyflie. We soldered all the boards and we got some basic ranging working on the nodes. The next step is to implement an opensource driver to be able to implement the ranging in the Crazyflie. We will keep you updated on the progress but in the mean time here is a photo of the prototypes:

DWM1000 nodes and deck

 

Things happening on the firmware side
Recently the commit rate for the Crazyflie 1.0/2.0 firmware has increased a lot. Some of it because of pull requests, great work, and some because we are starting to move in hacks and such on feature branches into the master branch. Our new college Kristoffer has taught us that having stuff on feature branches can be a bad idea, they tend to stay there. It is then better to have them compile switched and in the master branch as it is more visible and get a better chance of getting in for real.

Github crazyflie firmware contributions

 

Here are some of the recent thing going on:

  • A situation awareness framework originating from this pull request by fredgrat. It allows the Crazyflie 1.0/2.0 to react to triggers. Currently there is a free-fall, tumbled and at-rest detection. He recently also submitted an auto-takeoff functionality. Enable the functionality with the define here.
  • The beginning of a Arduino like API for the deck port. Currently GPIO and ADC are the only functions there but more will come.
  • Possibility to fly in rate (acrobatic) mode committed here. Support in the cfclient for this is being developed so currently one have to change the parameters to activate it manually.
  • Carefree, plus and X-mode implemented in firmware. There is also support for this being added to the cfclient.
  • Automatically switch to brushless driver. Motor driver being rewritten so it can be dynamically configured. This means that if the Crazylfie 2.0 is attached to the big-quad-deck it can automatically switch over to the brushless driver during power on.

Summertime are good times, less administration and more time to develop! As soon as things has been integrated and fully tested we will do a new release of the firmware and the cfclient :).

A while ago we implemented something we called mux-mode for controllers, where the Crazyflie can be controlled from multiple controllers at once. Initially it was implemented the day before a “bring-your-kids-to-work” day at Minc (where our office is). The idea was that the kids would control roll/pitch from one controller and we would control thrust/yaw from another controller. But we would also have the possibility to take over roll/pitch by holding a button on our controller. It was a big hit and let’s just say the “take-over” functionality came in handy :-)

A couple of months later we started working with the Kinect v2 with the goal of automatically piloting the Crazyflie using it. Again the input-mux feature came in handy. Instead of having the kids controlling the roll and pitch, the autopilot was now doing it. This enabled us to work on one problem at a time, first roll/pitch then yaw and finally thrust. When we were finished the autopilot was controlling all of the axes and we just used the “take-over” functionality when things got out of control.

So far this functionality has been disabled by default, but last week we fixed it up and enabled the code. With this change we’ve renamed the feature to teacher/student and also changed the way mappings are selected for the client. Below is a screenshot of the new menu, but have a look at the wiki for more details. If you want to try it out, pull the latest version on the development branch. It’s a great feature if you know someone that want’s to try flying for the first time!

On a side-note we tried some other ways to mix the controllers, like one we called “mix-mux”. This would take the input from two devices and add them together, so if both give 25% thrust the total would be 50%. It was really fun to try, but impossible to fly (maybe we need to work on our communication skills…).

Continuing from last Monday post where the hardware wiring part was discussed we now move on to the software side. The brushed motors are controlled with a normal PWM  where the duty cycle will adjust how much power goes into the motors. A brushless motor on the other hand needs more complicated controlling and uses it’s own micro-controller to handle this. These brushless motor controllers (BLMC or Brushless ESC) comes in many flavors and sizes but what is pretty common is how they are interfaced/controlled. This is inherited from how R/C receivers are controlling servos and how the receiver gets updates from the transmitter. This is a PWM where the width of the high pulse define the duty cycle. 1ms equals min and 2ms equals max and this is repeated every 20ms, thus giving an update rate of 50Hz.ServoPwm

This way of interfacing the BLMC is currently the most common way but interfacing with I2C, CAN, etc is getting more common.

To generate the servo PWM on the Crazyflie we have just reconfigured the timer a bit using a conversion macro so that setting the motor ratio of zero will result in a 1ms high pulse and setting it to max (uint16) will result in a 2ms pulse. The period time can be set with the BLMC_PERIOD define in motors.h. The standard period time of 20ms is actually a big drawback as it adds a latency from when a new output is calculated to when it is actually set. Therefore many motor controllers allow to shrink this period down to 2.5ms (400Hz) which result in lower latency and better flight stability.

Brushless prototype board

First you need to put together the brushless prototype board from the last post. The output will be generated on the pins marked BLMC 1,2,3,4 which should be connected in the same position an rotational direction as the brushed M1, M2, M3, M4 respectively. The output signal will be 0v – 3.0v which should work fine with 5V BLMC but it might be worth keeping that in mind.

BL proto descr

Building the brushless firmware

The code which contains the brushless functionality is currently on the bigmerge branch so start by pulling the latest changes and switching to that branch.

git pull
git checkout bigmerge

Then to activate the brushless functionality enable the brushless defines

BRUSHLESS_MOTORCONTROLLER
BRUSHLESS_PROTO_DECK_MAPPING

This can be done by by either creating defines in config.h or by creating  aconfig.mk file in the same directory as the Makefile with the content:

CFLAGS += -DBRUSHLESS_MOTORCONTROLLER
CFLAGS += -DBRUSHLESS_PROTO_DECK_MAPPING

The period time BLMC_PERIOD is by default set to 2.5ms (400Hz) so change that if needed.

Then build the firmware (make sure to clean first)

make clean
make

It will build for the Crazyflie 2.0 and for wireless bootloading by default. Put the Crazyflie 2.0 in bootloader mode by holding the power button until the blue led start to blink, then flash it with the wireless bootloader (Crazyradio required).

make cload

Causion!

You are now probably dealing with powerful and dangerous stuff so make sure to take precautions. E.g. don’t have propellers mounted when you test! When you have taken all the safety precautions do your first test. Remember the Crazyflie firmware has not yet been developed for big quads, it is all at your own risk! And another thing, even though the Crazyflie 2.0 can be controlled using a mobile device we don’t recommend this, use a Crazyradio.

Tuning

With that cautions being said it is great seeing a big quad fly and we will soon put out a video showing some of our builds. Our bigger quads have flown quite well with the stock tuning (PID parameters) but tuning should be done as well. Plenty of guides can be found on ways how to do it so I will not go into details here. The values can be found in pid.h and can be updated (but not saved) live using the cfclient. To save them the pid.h file must be changed and the firmware flashed again.

Recently I got a Chromebook, mostly out of curiosity for this odd “computer that runs only a web browser”. While playing with Google dev tools I quickly saw a possibility to make a Crazyflie client as a Chrome app: the Chrome API provides USB connectivity,and HTML5 has a javascript gamepad API. A chrome app is designed to look and feel like a native application: the app does not require internet connection and is launched in its own window.

This week-end I finally got around to test it, it’s not pretty but it works :-)

crazyflie_chrome_client

The current functionality is:

  • Channel and datarate can be changed
  • Read input from a gamepad, the mapping is fixed to mode 3 and the sensitivity is fixed
  • Sends set-points to Crazyflie 33 times per seconds.

I haven’t had time to do any layout work on it (that is pretty obvious in the screenshoot :) ), but the plan is to use Angular Material to handle the GUI.

This is only a proof of concept but we are seeing a lot of potential: the Chrome app runs on Linux, Mac, Windows and Chromebook, is easy to install and is written in HTML/CSS/Javascript which seems to be a very popular platform nowadays.

I have pushed the code on Github so if anyone is interested in helping to shape up the app head to the forum to read the discussion about it.

 

As many out there think is is more fun to fly the Crazyflie rather then develop upon it (like we do :-)) we have quickly looked at ways to pilot it with a RC transmitter. This forum thread is a good starting point for a developer discussion. To summarize it a bit there are many ways of implementing it which all require more or less development and has different pros/cons

  • Attach the Crazyradio PA to the expansion module/port of an RC transmitter.
  • Implement the E-Sky transmitter protocol or other nRF24L01+ protocols in the nRF51.
  • Using the DeviationTx code and a nRF24L01+ module.
  • Attaching a RC receiver to the deck (expansion port) interface.

Today I will write a bit more about the DeviationTx as it is a great open source project and that it has support for the Crazyflie. It was over a year ago we got contacted by Victor who wanted to implement the Crazyflie protocol in the DeviationTx code base, which he did pretty quickly. We feel ashamed for taking so long to try it out. So during the weekend I freed up some time and gave it a shot. The DeviationTx project replaces the firmware in Walkera transmitters with a better one which has the possibility to support a great amount of RC models. However many of them use different transceivers modules so this must be added to the hardware. Well a bit of hardware hacking is always fun and we had a nRF24L01+PA module laying around. They have a module installation document but I found it easier follow this guide as I had the same type of module and transmitter (Devo7e).

devo7e nRF24L01 module

The module installation was done pretty quickly but what took time was to update the firmware. The instructions tells one to download the walkera update tool but I just couldn’t find it on their website, nor the original walkera devo7e firmware (which they recommend to test with first). Thankfully I could find one using google and ended up using this link. Next thing was to fire up a windows 7 virtual machine to install it in which it worked without problems. So did the flashing of the DeviationTx firmware, I flashed the nightly build 4.0.1 and copying the file system according to the instructions. What I forgot though was to edit the hardware.ini file to enable the module which I understood when the protocol selection had a star in front of the name <*CFlie> (OK, I admit, I thought the hardware wasn’t working at first…). Then I setup a new model using the <CFlie> and I used the Fixed ID to setup the address, The data rate and channel are combined in the fixed id using channel as lowest two decimal digits and the rate the first were 0, 1, and 2 for 250kbit/s, 1Mbit/s, and 2Mbit/s respectively. So channel 80 on 2Mbit/s is encoded as 280 and channel 80 on 250kbit/s as 80.

And boy I was happy when I saw the green led (radio com led) blink on the Crazyflie 2.0, but nothing happened when I pushed the trust… Then I remembered, we recently implemented a lock so that a zero thrust must be sent at least once for it to unlock (to prevent Crazyflie to fly away if the gamepad is not setup correctly and constantly sends 100% thrust :). Looking into the source code one could see the trust was at minimum 5535 which was never unlocking the thrust. Removing this lock in the Crayflie 2.0 firmware and it was flying! but in plus mode… The Deviation cflie module code seem to rotate the pitch/roll setpoints.

Next step will be to do some modifications to the Deviation cflie module for Crazyflie 2.0, adding code to unlock the thrust and disabling the axis rotation, then make a pull request to the DeviationTx project for everyone to enjoy.

A big thanks to the DeviationTx project and to Victor for the cflie module implementation! Ohh, and by the way, they are making a universialTx module that will support almost any RC model out there, including Crazyflie, can’t wait to see that.

Lately we have merged a new input subsystem to the Crazyflie python client, it allows for more flexible input configuration like connecting more than one gamepad in training mode, or having external input like the LeapMotion appear as standard input. To test the flexibility of the new input system, on Friday, we implemented a ØMQ input driverØMQ is a library that permits to easily transmit messages between program. It is high speed, low latency and extremely easy to use. It has binding and implementation for a lot of programming language, which opens Crazyflie control to a lot more people and application.

The way it currently work is: You connect the graphical Crazyflie client and select ZMQ as input device. You connect the ØMQ socket and send json to it containing pitch/roll/yaw/thrust. And voila, you are controlling Crazyflie from your own program.

We just got a Kinect 2 and started playing around with it, suddenly the new ØMQ control became very useful. We already worked with the kinect long time ago so we had control code from back then. Also we wanted to quickly do a proof of concept so we started working with the Microsoft SDK to see what is possible, we are going to use libfreenect2 on Linux later. Finally we know how hard it was to control the thrust so we wanted a way to test pitch/roll first as a proof of concept. At the end of the day we ended up with this architecture in the office:

zmq_kinekt

The C++ implementation in visual studio detects the Crazyflie and sends coordinate to the Controller python script. The controller script is a stripped down version of or previous kinect experiment and only runs PID control loops to control pitch and roll, trying to keep Crazyflie at a fixed coordinate in space. Finally the Crazyflie client is setup in training mode with ZMQ handling pitch and roll and the gamepad handling thust and yaw. The gamepad can take-over completely in case of problem.

Thanks to ZMQ resiliance we can stop any part of the system and start it again, the connections automagically reappears. So we ended up with the following workflow:

  • Fly in the Kinect detection area
  • hand-over pitch/roll to the controller loop
  • When the Crazyflie starts oscillating or going away, take over control and land
  • Stop controller, modify things, restart

This was quite painless and nice. We ended-up connecting together very different systems and they just worked. We got so interested by the experiment that we now have a full ZMQ Crazyflie server on the work that would allow other program to do what is possible with the standard Crazyflie API: scan for Crazyflie, connect, read and write log and params.

 

When the schools closed for vacation last week Minc (the company accelerator where we have our office) hosted a “bring-your-children-to-work” day. The original idea came from Fredrik who works at the Swedish Arduino office (in the same building as us). One of the goals whas to let children see cool technology and to get a chance to try it out. We quickly jumped on this since we thought it would be fun for the children to fly. But we realized there was a bit of a safety problem letting Crazyflies go wild all around the place. So we dug up an old idea we had been discussing (yes, there’s lots of those..), the input-device MUX.

The idea is to allow combinations of input-devices to work together at the same time. There’s a couple of use-cases for this that we have been discussing in the past, like combining Kinect with a controller so you can take control if you loose tracking. But the most obvious one is a “learning” mode, where you could control roll/pitch from one device and thrust/yaw from another. Or having the possibility to take over control “on-the-fly” (no pun intended…) from one device to another. For us it makes it a lot easier to give a controller to people and say “Try it out!” if we can take over control just in case things start going bad.

So we started hacking around a bit and got it working, but there’s still lots of work to be done. We revised the architecture a bit for the input-device layer in the Crazyflie Python client from what we posted before. The main change is the multiplexer, which enables the user to open multiple input-devices at the same time and combine them. The other change is to connect other devices “higher-up” in the architecture, like the Leapmotion. Instead of connecting it as an input-device with a mapping, it is connected directly to the MUX and will give well scaled values for controlling. The same goes for other devices such as the Kinect and network connections (which would allow to control Crazyflie throw the client from any other software).

input-arch-mux

 

The hardest part will not be to code this, it will be to design the UI and configuration of this functionality. We have some sketches, but any ideas are welcome! In its current state you just select a MUX, then the devices that you want to use (master first, then slave). But there’s no way of configuring what values are taken from what device.

Check out this album to see a few photos from the “bring-your-children-to-work” day!