pyRobots: a toolkit for robot executive control¶
As you may well know if you ever tried to use them to implement under-specified tasks, state machines are not the best tool to code robot controllers.
pyRobots provides a set of Python decorators to easily turn standard
functions into background tasks which can be cancelled at anytime and to
make your controller resource aware (no, a robot can not turn left AND
right at the same time).
It also provides a event-based mechanism to monitor specific conditions and asynchronously trigger actions.
It finally provides a library of convenient tools to manage poses in a uniform way (quaternions, Euler angles and 4D matrices, I look at you) and to interface with existing middlewares (ROS, naoqi, aseba...).
Main features¶
Turns any Python function into a background action with the decorator
@action.Robot actions are non-blocking by default: they are instanciated as futures (lightweight threads),
Actions can be cancelled at any time via signals (the
ActionCancelledsignal is raised):@action def safe_walk(robot): try: robot.walk() except ActionCancelled: robot.go_back_to_rest_pose() action = robot.safe_walk() time.sleep(1) action.cancel()
Lock specific resources with a simple
@lock(...)in front of the actions. When starting, actions will wait for resources to be available if needed:L_ARM = Resource() R_ARM = Resource() ARMS = CompoundResource(L_ARM, R_ARM) @action @lock(ARMS) def lift_box(robot): #... @action @lock(L_ARM) def wave_hand(robot): #... @action @lock(L_ARM, wait=False) def scratch_head(robot): #... robot.lift_box() robot.wave_hand() # waits until lift_box is over robot.scratch_head() # skipped if lift_box or # wave_hand are still running
Supports compound resources (like
WHEELS==LEFTWHEEL+RIGHTWHEEL)Create event with
robot.whenever(<condition>).do(<action>)Poses are managed explicitely and can easily be transformed from one reference frame to another one (integrates with ROS TF when available).
Extensive logging support to debug and replay experiments.
Support for a particular robot only require to subclass GenericRobot
for this robot (and, obviously, to code the actions you want your robot to
perform).
Code Documentation¶
The documentation is currently sparse. Please fill bug reports everytime you can not figure out a specific bit.
Main entry points¶
Full package documentation¶
robots package¶
Subpackages¶
robots.concurrency package¶
-
robots.concurrency.action.action(fn)[source]¶ When applied to a function, this decorator turns it into a asynchronous task, starts it in a different thread, and returns a ‘future’ object that can be used to query the result/cancel it/etc.
The main methods available on these ‘future’ object include
RobotAction.wait()to wait until the action completes, andRobotAction.cancel()to request the action to stop (ie, it raises anActionCancelledsignal within the action thread). SeeRobotActionfor the full list of available methods.Action implementation may want to handle the
ActionCancelledsignal to properly process cancellation requests.Usage example:
@action def safe_walk(robot): try: robot.walk() except ActionCancelled: robot.go_back_to_rest_pose() action = robot.safe_walk() time.sleep(1) action.cancel()
In this example, after one second, the
safe_walkaction is cancelled. This sends the signalActionCancelledto the action, that can appropriately terminate.
Concurrency support for pyRobot.
This module provides:
- an implementation of
SignalingThread(threads that explicitely handle signals like cancelation) - heavily modified Python futures to support robot action management.
- A future executor that simply spawn one thread per future (action) instead of a thread pool.
These objects should not be directly used. Users should instead rely on the
action() decorator.
Helpful debugging commands:
>>> sys._current_frames()
>>> inspect.getouterframes(sys._current_frames()[<id>])[0][0].f_locals
-
class
robots.concurrency.concurrency.RobotAction(actionname)[source]¶ Bases:
concurrent.futures._base.Future
-
class
robots.concurrency.concurrency.RobotActionExecutor[source]¶
robots.events package¶
pyRobots’ events implementation
-
class
robots.events.events.EventMonitor(robot, var, value=None, becomes=None, above=None, below=None, increase=None, decrease=None, oneshot=False, max_firing_freq=10, blocking=True)[source]¶ -
ABOVE= '>'¶
-
BECOMES= 'becomes'¶
-
BELOW= '<'¶
-
DECREASE= '-='¶
-
INCREASE= '+='¶
-
VALUE= '='¶
-
-
class
robots.events.events.Events(robot)[source]¶ Exposes high-level primitives to create and cancel event monitors.
robots.robot.GenericRobotcreates and holds an instance ofEvents()that you can use: you should not need to instanciate yourself this class.-
every(var, max_firing_freq=10, blocking=True, **kwargs)[source]¶ Alias for
whenever().
-
on(var, **kwargs)[source]¶ Creates a new
EventMonitorto watch a given event model (one shot).On the first time the event is fired, the monitor is removed.
Returns: a new instance of EventMonitorfor this event.
-
stop_all_monitoring()[source]¶ Stops all event monitoring, but do not interrupt event callbacks, if any is running.
You may want to use
stop_all_monitoring()instead ofcancel_all()when you need to prevent new events of being raised from an event callback (cancel_all()would interrupt this callback as well).
-
whenever(var, max_firing_freq=10, blocking=True, **kwargs)[source]¶ Creates a new
EventMonitorto continuously watch a given event.varcan either be a predicate or the name of an entry in the robot’s state container (robot.state). In the later case, a supplementary keyword argument amongstvalue=,become=,above=,below=,increase=,decrease=must be provided to define the behaviour of the monitor.Example:
# using the robot state: robot.whenever("touch_sensor", value = True).do(on_touched) robot.whenever("sonar", below = 0.4).do(on_obstacle_near) robot.whenever("bumper", becomes = True).do(on_hit_obstacle) # using a predicate: def is_tired(robot): # do any computation you want... now = datetime.datetime.now() evening = now.replace(hour=20, minute=0, second=0, microsecond=0) return robot.state["speed"] > 1.0 and now > evening robot.whenever(is_tired).do(go_to_sleep)
Parameters: - var – either a predicate (callable) or one of the key of
robot.state. - max_firing_freq – set how many times pe second this event may be triggered (default to 10Hz. 0 means as many as possible).
- blocking – if
True, the event callback is blocking, preventing new event to be triggered until the callback has completed (defaults toTrue). - kwargs – the monitor behaviour (cf above)
Returns: a new instance of
EventMonitorfor this event.- var – either a predicate (callable) or one of the key of
-
robots.poses package¶
-
class
robots.poses.position.PoseManager(robot)[source]¶ Bases:
objectA pose is for us a dict
{'x':x, 'y':y, 'z':z, 'qx':qx, 'qy':qy, 'qz':qz, 'qw':qw, 'frame':frame}, ie a (x, y, z) cartesian pose in meter interpreted in a specific reference frame, and a quaternion describing the orientation of the object in radians.This class helps with:
- converting from other convention to our convention,
- converting back to other conventions.
-
angular_distance(angle1, angle2)[source]¶ Returns the (minimal, oriented) angular distance between two angles after normalization on the unit circle.
Angles are assumed to be radians.
The result is oriented (from angle1 to angle2) and guaranteed to be in range ]-pi, pi].
-
distance(pose1, pose2='base_link')[source]¶ Returns the euclidian distance between two pyRobots poses.
If the second pose is omitted, “base_link” is assumed (ie, distance between a pose and the robot itself).
-
get(raw)[source]¶ takes a loosly defined ‘pose’ as input and returns a properly formatted and normalized pose.
- Input may be:
- a frame
- an incomplete pose dictionary
- a list or tuple (x,y,z), (x,y,z,frame) or (z,y,z,rx,ry,rz) or (x,y,z,qx,qy,qz,qw)
-
inframe(pose, frame)[source]¶ Transform a pose from one frame to another one.
Uses transformation matrices. Could be refactored to use directly quaternions.
-
static
isin(point, polygon)[source]¶ Determines if a 2D point is inside a given 2D polygon or not.
Parameters: - point – a (x,y) pair
- polygon – a list of (x,y) pairs.
Copied from: http://www.ariel.com.au/a/python-point-int-poly.html
-
class
robots.poses.ros_positions.ROSFrames[source]¶
robots.resources package¶
-
robots.resources.lock.lock(res, wait=True)[source]¶ Used to define which resources are acquired (and locked) by the action.
This decorator may be used as many times as required on the same function to lock several resources.
Usage example:
L_ARM = Resource() R_ARM = Resource() ARMS = CompoundResource(L_ARM, R_ARM) @action @lock(ARMS) def lift_box(robot): #... @action @lock(L_ARM) def wave_hand(robot): #... @action @lock(L_ARM, wait=False) def scratch_head(robot): #... robot.lift_box() robot.wave_hand() # waits until lift_box is over robot.scratch_head() # skipped if lift_box or # wave_hand are still running
Parameters: - res – an instance of Resource or CompoundResource
- wait – (default: true) if
true, the action will wait until the resource is available, iffalse, the action is skipped if the resource is not available.
robots.mw package¶
robots.helpers package¶
An ANSI-based colored console log handler, based on https://gist.github.com/758430, and with a few special features to make sure it works well in pyRobots’ concurrent environment.
-
class
robots.helpers.ansistrm.ConcurrentColorizingStreamHandler(scheme=None)[source]¶ Bases:
logging.StreamHandlerA log handler that:
- (tries to) guarantee strong thread-safety: the threads generating log
message can be interrupted at any time without causing dead-locks (which
is not the case with a regular
StreamHandler: the calling thread may be interrupted while it owns a lock on stdout) - propagate pyRobots signals (ActionCancelled, ActionPaused)
- colors the output (nice!)
-
bright_scheme= {40: (None, 'red', False, False), 10: (None, 'blue', False, False), 20: (None, 'white', False, False), 50: ('red', 'white', True, False), 30: (None, 'yellow', False, False)}¶
-
color_map= {'blue': 4, 'black': 0, 'yellow': 3, 'cyan': 6, 'green': 2, 'magenta': 5, 'white': 7, 'red': 1}¶
-
csi= '\x1b['¶
-
dark_scheme= {40: (None, 'red', False, False), 10: (None, 'blue', False, False), 20: (None, 'black', False, False), 50: ('red', 'black', True, False), 30: (None, 'yellow', False, False)}¶
-
handle(record)[source]¶ Override the default handle method to remove locking, because Python logging, while thread-safe according to the doc, does not play well with us raising signals (ie exception) at anytime (including while the logging system is locking the output stream).
-
is_tty¶
-
mono_scheme= {40: (None, None, False, False), 10: (None, None, False, False), 20: (None, None, False, False), 50: (None, None, False, False), 30: (None, None, False, False)}¶
-
reset= '\x1b[0m'¶
-
xmas_scheme= {40: ('red', 'yellow', False, True), 10: ('red', 'yellow', False, True), 20: ('red', 'white', False, True), 50: ('red', 'white', False, True), 30: ('red', 'yellow', False, True)}¶
- (tries to) guarantee strong thread-safety: the threads generating log
message can be interrupted at any time without causing dead-locks (which
is not the case with a regular
robots.robot module¶
-
class
robots.robot.GenericRobot(actions=None, supports=0, dummy=False, immediate=False, configure_logging=True)[source]¶ Bases:
objectThis class manages functionalities that are shared across every robot ‘backends’ (ROS, Aseba,...)
You are expected to derive your own robot implementation from this class, and it is advised to use instances of
GenericRobotwithin a context manager (iewith MyRobot as robot: ...construct).Its role comprises of:
- automatic addition of proxy methods for the robot actions
- actions execution (spawning threads for actions via
self.executor - pose management through the
robot.posesinstance variable - event monitoring through the
robot.on(...).do(...)interface
GenericRobotdefines several important instance variables, documented below.Variables: - state – the state vector of the robot. By default, a simple dictionary. You can overwrite it with a custom object, but it is expected to provide a dictionary-like interface.
- poses – an instance of
PoseManager. - executor – instance of
RobotActionExecutorresponsible for spawning and starting threads for the robot actions. You should not need to access it directly.
Example of a custom robot:
from robots import GenericRobot class MyRobot(GenericRobot): def __init__(self): super(MyRobot, self).__init__() # create (and set) one element in the robot's state. Here a bumper. # (by default, self.state is a dictionary. You can safely # overwrite it with any dict-like object. self.state["my_bumper"] = False # do whatever other initialization you need for your robot # Implement here all the accessors you need to talk to the robot # low-level, like: def send_goal(self, pose): # move your robot using your favorite middleware print("Starting to move towards %s" % pose) def stop(self): # stop your robot using your favorite middleware print("Motion stopped") def whatever_other_lowlevel_method_you_need(self): #... pass # create actions @action def move_forward(robot): #... pass with MyRobot() as robot: # Turn on DEBUG logging. # Shortcut for logging.getLogger("robots").setLevel(logging.DEBUG) robot.debug() # subscribe to events... robot.whenever("my_bumper", value = True).do(move_forward) try: while True: time.sleep(0.5) except KeyboardInterrupt: pass
Note
A note on debugging
Several methods are there to help with debugging:
loglevel(): default toINFO.logging.DEBUGcan be useful.running(): prints the list of running tasks (with their IDs)actioninfo(): give details on a given action, including the exact line being currently executed
Parameters: - actions (list) – a list of packages that contains modules with
actions (ie, modules with functions decorated with
@action). Proxies to these actions are appended to the instance of GenericRobot upon construction. - supports – (default: 0) a mask of middlewares the robot
supports. Supported middlewares are listed in
robots.mw.__init__.py. For examplesupports = ROS|POCOLIBSmeans that both ROS and Pocolibs are supported. This requires the corresponding Python bindings to be available. - dummy (boolean) – if
True(defults toFalse), the robot is in ‘dummy’ mode: no actual actions are performed. The exact meaning of ‘dummy’ is left to the subclasses of GenericRobot. - immediate (boolean) – if
True(defaults toFalse), actions are executed in the main thread instead of their own separate threads. Useful for some specific debugging scenarios. - configure_logging (boolean) – if
True(default), configures a default colorized console logging handler.
-
cancel_all()[source]¶ Sends a ‘cancel’ signal (ie, the
ActionCancelledexception is raised) to all running actions.Note that, if called within a running action, this action is cancelled as well. If this is not what you want, use
cancel_all_others()instead.Actions that are not yet started (eg, actions waiting on a resource availability) are simply removed for the run queue.
-
cancel_all_others()[source]¶ Sends a ‘cancel’ signal (ie, the
ActionCancelledexception is raised) to all running actions, except for the action that call :meth:`cancel_all_others` (note that its currently running subactions will be cancelled).Actions that are not yet started (eg, actions waiting on a resource availability) are simply removed for the run queue.
-
filtered(name, val)[source]¶ Helper to easily filter values (uses an accumulator to average a given ‘name’ quantity)
-
static
sleep(duration)[source]¶ Active sleep. Must used by actions to make sure they can be quickly cancelled.
-
wait(var, **kwargs)[source]¶ Alias to wait on a given condition. Cf
robots.events.Eventsfor details on the acceptable conditions.
-
wait_for_state_update(timeout=None)[source]¶ Blocks until the robot state has been updated.
This is highly dependent on the low-level mechanisms of your robot, and should almost certainly be overriden in your implementation of a GenericRobot subclass.
The default implementation simply waits
ACTIVE_SLEEP_RESOLUTIONseconds.
robots.introspection module¶
Minimum Working Example¶
...that includes the creation of a specific robot
import time
from robots import GenericRobot
from robots.decorators import action, lock
from robots.resources import Resource
from robots.signals import ActionCancelled
# create a 'lockable' resource for our robot
WHEELS = Resource("wheels")
class MyRobot(GenericRobot):
def __init__(self):
super(MyRobot, self).__init__()
# create (and set) one element in the robot's state. Here a bumper.
self.state.my_bumper = False
# do whatever other initialization you need :-)
def send_goal(self, pose):
# move your robot using your favorite middleware
print("Starting to move towards %s" % pose)
def stop(self):
# stop your robot using your favorite middleware
print("Motion stopped")
def whatever_lowlevel_method_you_need(self):
pass
@lock(WHEELS)
@action
def move_forward(robot):
""" We write action in a simple imperative, blocking way.
"""
# the target pose: simply x += 1.0m in the robot's frame. pyRobots
# will handle the frames transformations as needed.
target = [1.0, 0., 0., "base_link"]
try:
robot.send_goal(target)
while(robot.pose.distance(robot.pose.myself(), target) > 0.1):
# robot.sleep is exactly like time.sleep, except it lets the pyrobots
# signals pass through.
robot.sleep(0.5)
print("Motion succeeded")
except ActionCancelled:
# if the action is cancelled, clean up your state
robot.stop()
with MyRobot() as robot:
# Turn on DEBUG logging.
# Shortcut for logging.getLogger("robots").setLevel(logging.DEBUG)
robot.debug()
robot.whenever("my_bumper", value = True).do(move_forward)
try:
while True:
time.sleep(0.5)
except KeyboardInterrupt:
pass