This guide was written before the hardware classes were refactored. The intuition of the guide is still true, but most hardware objects are now built by overwriting metaclass methods. See the API documentation for hardware for examples :)

This documentation will be updated in v0.3.1, sorry for the delay!

Writing a Hardware Class

There are precious few requirements for Hardware objects in Autopilot.

  • Each class should have a release() method that stops any running threads and releases any system resources – especially those held by pigpio.

  • Each class should define a handful of class attributes when relevant
    • trigger (bool) - whether the device is used to trigger an event. if True, assign_cb() must be defined and the device will be given a callback function by the instantiating Task class

    • type (str) - what this device should be known as in prefs. Not enforced currently, but will be.

    • input and output (bool) - whether the device is an input or output device, if either

  • When making threaded methods, care should be taken not to spawn an excessive number of running threads, but this is a performance rather than a structural limit.

To use a hardware object in a task, its parameters (especially the pin number for pigpio-based hardware) should be stored in prefs.json.

A few basic Hardware classes are dissected in this section to illustrate basic principles of their design, but we expect Hardware objects to be extremely variable in their implementation and application.


In future versions of Autopilot, the structure of hardware prefs will be formalized similarly to the preamble of tasks to make establishing and maintaining parameterizations more transparent.

GPIO with pigpio

Autopilot uses pigpio to interface with the Raspberry Pi’s GPIO pins. All pigpio objects require that a pigpiod daemon is running as a background process. Typically this is managed by systemd or the launch script generated for Pilots.

When instantiating a piece of hardware, it must connect to pigpiod by creating a pigpio.pi object, which allows communication with the GPIO.

Input - Beambreak


the Beambreak class is now Digital_In, see its API documentation for its current implementation :)

The Digital_In class is a digital input class that registers (by default) a high-to-low logic transition and calls a callback function. When it is initialized, it connects to a GPIO pin, configures it for input, and sets the pull-up (or down) resistor.

class Beambreak(Hardware):
    # this class description has been simplified for clarity

    trigger = True
    type    = 'POKES'
    input   = True

    def __init__(self, pin, pull_ud='U', trigger_ud='D', event=None):

        # Make pigpio instance
        self.pig = pigpio.pi()

        # Convert pin from board to bcm numbering = BOARD_TO_BCM[int(pin)]

        # save which direction our trigger should be
        # TRIGGER_MAP takes string directions - eg. 'D' for 'Down'
        # or falling edge detection - and converts them to
        # pigpio constants
        self.trigger_ud = TRIGGER_MAP[trigger_ud]

        # Setup pin
        self.pig.set_mode(, pigpio.INPUT)
        self.pig.set_pull_up_down(, self.pull_ud)

        # create empty list of callbacks
        # (to handle multiple callbacks, if needed)
        self.callbacks = []

Since trigger == True, the instantiating task class will try to give it a method to call to handle the trigger. We redefine assign_cb() to make use of pigpio’s callback functionality. Since pigpio can handle multiple callback functions, one can optionally specify add=True to prevent any previous callbacks from being cleared. This has been omitted in this example for clarity, but can be inspected in the API documentation for the Digital_In class.

def assign_cb(self, callback_fn):
    cb = self.pig.callback(, self.trigger_ud, callback_fn)

To clean up the connection to the pin made by this instance of the object, we must also redefine the release method. We also redefine __del__ to attempt cleanup if the object is garbage-collected without explicitly calling release()

def release(self):

def __del__(self):

Output - LED_RGB


The LED_RGB also has been updated to use pigpio’s scripts rather than python threads for light series, but this documentation remains as an example of how similar logic could be written for other non-gpio hardware objects.

This LED_RGB class is a bit different. It’s an output device, yes, but it also manages multiple pins, uses pulse-width modulation rather than strict binary logic, and has a few extra tricks up its sleeve.

Its initialization is similar to Digital_In except we add a few threading.Event s to handle threaded lighting routines. LEDs can either be common anode or common cathode which affects the polarity of the pulse-width modulated signal, but handling different LED polarity has been omitted for brevity.

class LED_RGB(Hardware):
    # this class has also been simplified for clarity

    output = True

    def __init__(self, pins = None):

        # Dict to store color for after flash trains
        self.stored_color = {}

        # Event to wait on setting colors if we're flashing
        self.flash_block = threading.Event()

        # Event to kill the flash thread if the object is deleted
        self.end_thread = threading.Event()

        # Initialize connection to pigpio daemon
        self.pig = pigpio.pi()

        # Convert to BCM numbers
        self.pins = {k: BOARD_TO_BCM[v] for k, v in self.pins.items()}

        # set pin mode to output and make sure they're turned off
        for pin in self.pins.values():
            self.pig.set_mode(pin, pigpio.OUTPUT)
            self.pig.set_PWM_dutycycle(pin, 0)

Setting colors is straightforward - we are given a length-3 tuple of 8-bit (0-255) RGB color values and set the pulse-width modulated duty cycle accordingly. We use a recursive threading.Timer in order to manage timed light flashes – after some duration, the set_color() method is called turning the lights off.

def set_color(self, col, timed=False):

    # unpack color
    color = {'r':int(col[0]), 'g':int(col[1]), 'b':int(col[2])}

    for k, v in color.items():
        self.pig.set_PWM_dutycycle(self.pins[k], v)

    # If this is is a timed blink, start thread to turn led off
    if timed:
        # timed should be a float or int specifying the delay in ms
        offtimer = threading.Timer(float(timed)/1000,

In order to implement flash trains or other rapid sequences of lights, we make a color_series method that takes a list of rgb tuples and either a single duration (which is applied to all colors in the series) or a series of durations of equal length to the list of colors.

color_series is a convenience function that spawns a thread to handle the color series without blocking. The actual threaded_color_series method has a few extra bells and whistles to make sure series don’t overlap with one another, it is simplified here to illustrate the principle.

def color_series(self, colors, durations):
    series_thread = threading.Thread(target=self.threaded_color_series,
                                        'colors'   : colors,
                                        'durations' : durations

def threaded_color_series(self, colors, durations):
    # assume len(colors) == len(duration)
    # for this example. Iterate through both, setting colors
    for color, duration in zip(colors, durations):

With one more layer of abstraction we can create a flash method that lets us cycle through colors with a specific frequency rather than by defining color and duration series by hand

def flash(self, duration, frequency=10, colors=((255,255,255), (0,0,0))):

    # Get number of flashes in duration rounded down
    n_rep = int(duration / 1000.0 * frequency)

    # Invert frequency to duration for single flash
    # divide by 2 b/c each 'color' is half the duration
    single_dur = ((1. / frequency) * 1000) / 2.

    # make tuples of flashes and durations
    flashes = colors * n_rep
    durations = [duration] * n_rep

    self.color_series(flashes, durations)

The release function is also the same, we just make sure to turn the LEDs off before we leave them

def release(self):

USB Hardware with inputs

USB hardware logic is much more variable than GPIO-based hardware, so we’ll wait for you to help us with good examples!

For example, the Wheel class discussed previously in A Very Smart Wheel reads mouse events with the inputs package like this:

from inputs import devices

mouse = devices.mice[some_index_for_your_mouse]
event =

the event object has three attributes of interest, event.state, event.code, and event.timestamp. The code tells us if the event was a movement (ie. REL_X or REL_Y in the case of computer mice, as opposed to click events), and the state tells us how much.

The Wheel class then creates a numpy array of movements like this:

# filter events
events = [e for e in events if e.code in ('REL_X', 'REL_Y)]

# extract tuples of attributes
events = [(e.state, e.code, e.timestamp) for e in events]

# convert to numpy array
events = np.array(events,