crazyflie-on-voice

crazyflie-on-voice provides a generic HTTP control server that allows integrating the crazyflie with custom programs, as well as with a wide range of third-party systems. In addition, crazyflie-on-voice provides a client that allows you to voice control Crazyflie 2.0 drones.

Hardware requirements

To use crazyflie-on-voice, you need a Crazyflie 2.0 and a Loco Positioning System (LPS). Follow the instructions in the LPS documentation to set up your Crazyflie to work with the LPS.

Installation and setup

crazyflie-on-voice requires Python 3.6 or higher. To get started, proceed as follows:

• First install the dependency catkin_pkg manually, by running pip install catkin_pkg.

• If you want to use the voice controller, also install the following dependencies (you only need pocketsphinx if you want to use it instead of Google's speech-to-text API):

  pocketsphinx
  pyaudio
  python-Levenshtein
  SpeechRecognition
  word2number
  

  Installing pocketsphinx and pyaudio can require some effort (use the search engine of your choice).

• To install crazyflie-on-voice, run pip install git+https://github.com/TimKam/crazyflie-on-voice.

• To start the crazyflie controller, run crazyflie-on-voice --uri=<your_crazyflie_uri>. (Replace <your_crazyflie_uri> with the URI of your Crazyflie, e.g. radio://0/80/250K.)

• Start the voice control client by running crazyflie-on-voice --voice-only.

Options

The following command line options are available:

• --uri, -u: URI of the crazyflie. Defaults to radio://0/110/2M
• --control-port, -cp: Port for the control server. Defaults to 8000.
• --planing-port, -pp: Port for the planning server. Defaults to 8001.
• --room-spec, -rs: Path to the room specification file. (see Path planning below). Defaults to ./examples/room_spec_1.yaml.
• --voice, -v: Add, if you also want to start the voice control client. (The voice control client does not start by default.)
• --voice-only, -vo: Add, if you only want to start the voice control client.
• --voice-api, -va: Speech-to-text API the voice control client users. Either google (software-as-a-service) or pocketsphinx (free software, local installation). Defaults to google.
• --control-url, -cu: URL of the control server (without port). Only relevant for the voice client (in case you want to connect to a control server on a different machine). Defaults to http://localhost.

Voice control

To control your Crazyflie by voice, use the word sequence Crazy + <direction> + <distrance>, where

• <direction>: up, down, left, right, ahead, back, start, or stop;
• <distrance>: the relative distance in centimeters, rounded to decimeters (e.g. 10 or 230).

Note that the direction is absolute, considering the x,y, and z coordinates of the LPS (and not considering the yaw angle of the crazyflie).

To stop your Crazyflie, use Crazy stop.

Pathfinding capabilities

The Crazyflie can autonomously circumvent obstacles using a custom implementation of a an A* search-based pathfinding algorithm. The bounds of the environment as well as the obstacles within the environment must be described and are assumed to be static. Currently, obstacles are described as unit cubes which are then scaled and translated so they represent objects of the correct size and position in the environment. (Rotation of objects is currently not possible but could be added easily.) For path planning, the scene is discretised as a cartesian grid with cell size 0.1m x 0.1m x 0.1m. There is no benefit in using smaller cells since the accuracy of the positioning system is limited. For each grid cell, all obstacles are sampled to determine whether they are occupied. The grid is then used to find the shortest path that avoids obstacles between two points using A*.

You can model your environment as a YAML file. For example, an obstacle of size 1.6m x 0.8m x 2.2m that has its "origin" at position 1.25m x 1.7m on the floor can be modelled like this:

- obstacle:
    - [1.60, 0.8, 2.20]
    - [1.25, 1.70, 0.00]

You should also provide the measures of the room in which you are flying (as the first entry in your YAML file). Here is an example of a full specification:

- room:
  - [4.0, 5.0, 2.5]

- table1:
  - [1.30, 0.65, 0.75]
  - [1.35, 0.68, 0.00]

- table2:
  - [1.30, 0.65, 0.75]
  - [1.35, 2.68, 0.00]

- obstacle:
  - [1.60, 0.8, 2.20]
  - [1.25, 1.70, 0.00]

The first item has to be the room measures specification. Note that for each obstacle, the first array specifies the measures [x, y, z] in meters; the second array specifies the coordinates of the corner that is closest to [0, 0, 0].

Then, run crazyflie-on-voice as follows:

crazyflie-on-voice "<your_crazyflie_uri>" --room-spec="<path-to-yaml-file>"

Generic HTTP interface

crazyflie on voice consists of two main components: a voice control client and a generic HTTP server. To only run the HTTP server, exectue crazyflie-on-voice --server --port=<port>. Replace <port> with the port on which you want your server to run.

The server accepts POST requests to its base URL. The requests have to have the following structure:

• either:

   {"command": "<command>"},
  
  , where command is either ``stop``, ``start``, or ``land``.
  

• or:

  {"distance": ["<x>, <y>, <z>]}
  

  , where <x>, <y>, <z> is the change in x, y, and z coordinates you want to achieve.

  For example:

  {"distance": [0, 0, 0.5]}}
  

Troubleshooting voice control

crazyflie-on-voice makes use of the SpeechRecognition library. In case you want to use crazyflie-on-voice with PocketSphinx and you have problems installing the package or with voice processing, read the instructions on the SpeechRecognition documentation page to and make sure SpeechRecognition works on your machine with PocketSphinx and PyAudio.

Acknowledgements

This work was partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.