-
Notifications
You must be signed in to change notification settings - Fork 14
aldebaran/libqicore
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
QiCore README ============= This lib is a fork of the "frameManager", it aims to execute behaviors that was previous executed with the frameManager but with much more flexibility. At the moment, QiCore is only a compatibility layer, but it aims to become the new execution core of behaviors. QiCore and the frameManager are independant, meaning that you can execute behaviors with both of us. Getting Started: ---------------- Let's suppose you want to execute a behavior stored in a xar file. First of all, correctly set your PYTHONPATH to collect all python modules needed. 1/ Convert the behavior: python2 xar_converter.py path/to/behavior/behavior.xar output_folder/ This step will give you a folder with all objects needed by the behavior and a main.py, entry point of the behavior. 2/ Execute the behavior: python2 output_folder/main.py IP PORT Where IP and PORT are the ip and port of the naoqi you want to connect. The part to understand in the execution process is that the behavior is executed on the machine that lunch the behavior with the python interpreter but the result of the behavior is visible on the naoqi you are connected to. In other words, commands are send to the distant naoqi at the right time, but this distant naoqi does not know anything about the behavior. To execute behaviors the same way that frameManager does, you probably want to copy libQicore to the robot and then execute the following command on the robot: python2 output_folder/main.py 127.0.0.1 9559 Sub-Projects: ------------- QiCore can be splitted in different entities: - The converter It converts a xar file (v3) to a new format, discussed later. - The core with C++ Objects that executes a behavior Composed with Timeline, StateMachine, Box, Transition - The object factory Load python objects at run-time and connect them - The qicore_legacy Provide the support of functions provided previously by ALBehavior Status: ------- - Converter: Subject to a lot of testing, it converts successfully all tested behaviors. May be an effort can be put in printing nice error if file is incorrect. - Core: StateMachine: used to replace behaviorKeyFrames, behaviorLayers Timeline: a lot of code comes from the legacy timeline, but this object handles only actuatorCurves. Threading model has been rebuilt. Box: Represents a legacy box, a box is also a state in the stateMachine Transition: Represents a transitions between two boxes The main differences in objects representations with the frameManager is that what is called Timeline in the frameManager is splitted into : Timeline and StateMachine in QiCore. Also the Timeline in QiCore is a lot simpler, and handles much less work. - Object Factory: At the moment, only load python objects. - QiCore Legacy: Provide the same level of functionnalities than ALBehavior. Differences : qicore_legacy does not provide resource management at the moment. Objects are not ALModules for the moment. Benefits over frameManager: --------------------------- - Interactions between C++ and Python: In the frameManager, interactions between C++ and Python are not very clear, some of the work are done in python, but connections between boxes are done in C++. C++ code calls python that will call C++ again. With QiCore, the behavior is written in Python, this Python code plays with C++ objects transparently (Swig is used for the binding). Sometimes C++ has to hold a callback in the Python world, but there is an effort to minimize the number of callbacks. Callbacks are listed later. - ALBehavior, ALPythonBridge, ALMemory frameManager required all these objects to work. Signals between boxes are done by ALMemory, it is a bottleneck because ALMemory is a centralized sytem. All the Python code is executed in a special C++ Thread with an embedded Python interpreter. With QiCore, interactions between boxes are done by qi::Signal, so boxes can interacts with each others without the help of a central module. Python code can be executed directly by any stand alone python interpreter. - Timeline concept and StateMachine The timeline concept was a bit messy and unintuitive with the frameManager. A lot of people was using the timeline of the frameManager as a StateMachine. QiCore provides a Timeline that executes only actuatorCurves and a StateMachine - Code generation FrameManager generates a lot of code with each boxes, and debugging generated code is awful. QiCore aims to generate a mimum of code, and leave the original code almost untouched, code needed by the boxes are added at run-time, transparently. - Play with C++ Objects FrameManager does not provide an API to play with Timeline and others objects QiCore provides an API to interact with objects in behaviors - Maintain the code All concepts in the frameManager are intertwined, so the code is hard to maintain and understand. QiCore wants to provide clean concepts, easy to unterstand. - New format Format supported by the frameManager include embedded python code. QiCore provides a new format with python code stored as .py and meta informations stored in xml files. Objects are splitted in different files. - Stand alone FrameManager can not produce a stand alone python file that can be executed With QiCore you can simply execute a behavior with the following command: python2 behavior/main.py IP PORT That means you can execute a behavior on a desktop computer and send commands to a robot. We can also imagine that a robot send a behavior to another robot or a desktop computer can execute the same behavior on multiple robots. To do: ------ - Catch SIGINT and send a on_stop signal to all boxes to quit elegantly. - At the moment objects does not need to be in globals, may be change that. - Path to file needed by the behaviors are incorrect, for example if you want to play a sound on a robot. - All functions that handles resources just do nothing (qicore_legacy.py, waitForResources, ...) - Naoqi is still needed to create proxies on services, but it is possible to replace ALProxy in naoqi by some qimessaging binding - Boxes uses qi::signal, but are not naoqi2 objects - Timeline tests are not usable, fix it - Improve the ALMemoryWatcher and replace the polling thread by a subscribeToEvent - Fix the "from naoqi import *" in py files - Cleaner XML generation in converter/file_writer.py - Swig detects a leak with boost::shared_ptr<ALBroker>, fix it. This leaks only happends when creating a timeline - There is a mix between the new API and the old API, an effort can be put to use a minimum of the old API. - The goTo in the stateMachine can be faster if we store labels in a map, instead of going though each box. - The goTo in StateMachine and Timeline is a bit messy, it is the StateMachine that handles the goto, and then we handle the timeline. May be it will be cleaner to let the Timeline handle the goTo and contact the StateMachine if we need to change the state - To be able to control the StateMachine or the Timeline of a parent, each box have a pointer to its parent. But this pointer can be replaced by a signal. - A lot of fixes to execute behaviors exactly as frameManager To know: -------- - The StateMachine execute transitions in a thread, transitions are queued are performed in order of arrival (FIFO) - Callbacks: * When changing states the the StateMachine will call the load callback on the next state, the unload callback on the previous state, and call the onNewState callback to trigger IO between boxes. * When the timeline has reached the last frame, it calls a python callback to trigger onStopped Outputs - If you want to convert a crg file to the new format you can do the following: ar vx my_behavior.crg && tar vxf data.tar.gz to extract the .xar file and then convert it. - Functions generated by the frameManager are added at run time and generated in function_generator.py - The code that converts a old-style Timeline in a StateMachine is in the converter, in the file converter/new_format_generator.py - To be able to reproduce the goTo of the old-style Timeline, the StateMachine uses two things: labels, that allows to a state associated with the label and provides an interval that belongs the state, this is useful when the user want to jump to a frame. - Connections between boxes are done when loading the state associated and destroyed when leaving the state - The Timeline have flags to trigger each state of the StateMachine at the right time. A flag is the association of a state name and the frame you want to trigger the transition. The only exception is the first frame, there is no flag for it, cause it is the initial state and the state is entered by calling run on the stateMachine Hacks: ------ - There is a ref couting on load/unload boxes to prevent multiple loads, indeed when converting an old-style timeline to a stateMachine some objects are duplicated but you do not want to load them multiple times - The same ref couting mechanism is present with connection between boxes for the same reasons. - onLoad signal seems like a hack in the frameManager, this signal is triggerd each time we enter in a new keyframe. In the StateMachine this signal has to be triggered when entering a new state but this signal must not trigger boxes already activated. So every time the onLoad signal is triggerd, this signal is replaced by a new one with no connections. The previous onLoad is then stored because we do not want Xar format documentation: ------------------------- A Xar file is an xml file that describes a behavior Elements: - Box - Diagram - Timeline - BehaviorLayer - BehaviorKeyFrame - Script - Content - Input - Output - Parameter - Resource - ActuatorList - ActuatorCurve - Key - Link - Bitmap Box: Attributes: - name: name of the box - robot: robot targeted by the behavior - id: id of the box - tooltip: hint about how to use the box - bitmap_expanded: FIXME - plugin: FIXME - x: x coordinate of the box - y: y coordinate of the box Inner-Elements: - bitmap - script - Input - Output - Parameters - Timeline - Resources Bitmap: Content: - Path to the bitmap file Script: Attributes: - Language: - 0 -> C++ - 1 -> URBI - 2 -> Ruby - 3 -> Dynamic Librabry - 4 -> Python - 5 -> QiChat - 6 -> No Script Content: - Content Beacon Content: Content: - The script it-self. Input: Attributes: - name : Name of the input - type: - 0 -> Dynamic - 1 -> Bang - 2 -> Number - 3 -> String - 4 -> Bitmap - 5 -> Sound - type_size: size of the type - nature: - 0 -> onLoad - 1 -> unDef - 2 -> onStart - 3 -> onStop - 4 -> STMValue - stm_value_name : value associated in ALMemory - inner : - tooltip : tip for the user - id : identifier for the IO Port (local to the box) Output: Attributes: - name : Name of the input - type: - 0 -> Dynamic - 1 -> Bang - 2 -> Number - 3 -> String - 4 -> Bitmap - 5 -> Sound - type_size: size of the type - nature: - 0 -> unDef - 1 -> Stopped - 2 -> Punctual - 3 -> Recurrent - inner : - tooltip : tip for the user - id : identifier for the IO Port (local to the box) Parameter: Attributes: - name: name of the parameter - inherits_from_parent: - 0 -> parameter is not inherited - 1 -> inherited parameter - content_type: - 0 -> Bool - 1 -> Int - 2 -> Double - 3 -> String - 4 -> Resource - value: value of the parameter - default_value: default value taken by the parameter - min : min value for the parameter - max : max value for the parameter - tooltip : hint for the user - id : identifier for the parameter (local to the box) - custom_choice : FIXME Timeline: Attributes: - enable: - 0 -> disabled - 1 -> enabled - name: timeline name - fps : Frames per seconds - resource_acquisition: - 0 -> Passive - 1 -> Waiting - 2 -> Aggresive - size: FIXME - start_frame: first frame - end_frame: last frame - scale : FIXME Inner-Elements: - BehaviorLayer - ActuatorList - watches : FIXME BehaviorLayer: Attributes: - name : name of the behaviorLayer - mute : FIXME Inner-Elements: - BehaviorKeyFrame BehaviorKeyFrame: Attributes: - name : name of the keyframe - index : frame associated with the keyframe Inner-Elements: - Diagram Diagram: Attributes: - scale : FIXME Inner-Elements: - Box - Link Link: Attributes: - inputowner : id of the box local to the diagram, 0 means the parent box - outputower : same as previous - indexofinput : id of the input (local to the box) - indexofoutput : id of the output (local to the box) ActuatorList: Attirbutes: - model : FIXME Inner-Elements: - ActuatorCurve ActuatorCurve: Attributes: - name : name of the actuatorCurve - actuator : Resource concerned by the actuatorCurve - recordable : FIXME - mute : FIXME - unit : FIXME Inner-Elements: - Key Key: Attributes: - frame : frame concerned by the parent ActuatorCurve - value Resource: Attributes: - name : name of the resource - type : - 0 -> Lock - 1 -> Stop_on_demand - 2 -> Pause_on_demand - 3 -> Callback_on_demand - timeout Elements Hierarchy: <Box> <bitmap> </bitmap> <script> <content> </content> </script> <Input /> ... <Input /> <Output /> ... <Output /> <Parameter /> ... <Parameter /> <Timeline> <BehaviorLayer> <BehaviorKeyFrame> <Diagram> <Box> </Box> .... <Box> </Box> </Diagram> </BehaviorKeyFrame> <BehaviorKeyFrame> ... (same as previous) </BehaviorKeyFrame> </BehaviorLayer> <BehaviorLayer> ... (same as previous) </BehaviorLayer> <ActuatorList> <ActuatorCurve> <Key /> ... <Key /> </ActuatorCurve> <ActuatorCurve> ... (same as previous) </ActuatorCurve> </ActuatorList> </Timeline> <Resource /> ... <Resource /> </Box> New format documentation: ------------------------- This new format is not a definitive format, but is more like a test on how to create a format more human readable and easy to edit. It aims to be compatible with the previous xar format. There is two kind of files: - .xml files -> contains meta informations about the object - .py files -> contains the Python code of the object All these files are stored in the same directory Each object is made of one py file and on xml file. Let's say I have an object called foo, its reprensentation is: - l0_foo.xml - l0_foo.py Naming convention: The previous format does not provide an unique name to each object, so we have to handle it. The naming convention is pretty simple: => l + 'level of the object in previous format' + _ + 'name of the object' Let's say I have on object foo on level 42 (so in level 42 in the hierarchy of the xar format), the object will be named: - l42_foo.xml - l42_foo.py Name collisions: If two objects have the same name, the converter will add a unique number to each object. Let's say I have two objects foo, the converter will give: - l1_foo.xml - l1_foo.py - l1_foo_1.xml - l1_foo_1.py Main: There is one special file : main.py This file does not represents an object but it is the file to execute when you want to play the behavior StateMachine and Timeline: Each object can have a state_machine and a timeline associated. In this case this object take the name of the owner and appending "state_machine" or "timeline". Let's say I have a root box with a stateMachine and a timeline, this will give me: - l0_root.py - l0_root.xml - l0_root_state_machine.xml - l0_root_timeline.xml Note that StateMachine and Timeline have no .py associated, there is no code that comes with these objects. States: A state machine a different state associated with it, the naming convention is the following, let's say I have a root box with a stateMachine and three states. - l0_root.py - l0_root.xml - l0_root_state_machine.xml - l0_root_state_0.xml - l0_root_state_1.xml - l0_root_state_2.xml Note that a state has no .py associated either. The loader will load only file needed in the behavior (i.e. presents in the objects graph) and will leaves the others alone. If an object that is not connected to others in present in the directory, it will not be loaded. XML files: Elements: - Box - Timeline - Input - Output - Parameter - Resource - ActuatorList - ActuatorCurve - Key - Link - StateMachine - State Box: Attributes: - name: name of the box - robot: robot targeted by the behavior - id: id of the box - tooltip: hint about how to use the box - bitmap_expanded: FIXME - plugin: FIXME - x: x coordinate of the box - y: y coordinate of the box - bitmap Input: -> Same has the xar format Output: -> Same has the xar format Parameter: -> Same has the xar format Timeline: Attributes: - enable: - 0 -> disabled - 1 -> enabled - name: timeline name - fps : Frames per seconds - resource_acquisition: - 0 -> Passive - 1 -> Waiting - 2 -> Aggresive - size: FIXME - start_frame: first frame - end_frame: last frame - scale : FIXME Inner-Elements: - ActuatorList - watches : FIXME ActuatorList: -> Same has the xar format ActuatorCurve: -> Same has the xar format Key: -> Same has the xar format Resource: -> Same has the xar format StateMachine: Attributes: - State : state in this stateMachine - InitialState : initial state in the stateMachine - FinalState : final state in the stateMachine - Transition : a transition between two states State (in the stateMachine file): Attributes: - Name : name of the state InitialState Attributes: - Name : name of the state FinalState Attributes: - Name : name of the state Transition Attributes: - From : name of the from state - To : name of the to state State (in its own file): Attributes: - Object : object in the state - Interval : begin and end frame of the state - Label : name associated to state - Link : a connection between two objects Object: Attributes: - Name : name of the object Interval: Attributes: - begin : first frame in the state - end : last frame in the state Label: Attributes: - Name : name associated with the state Link: Attributes: - inputowner : name of the input object - outputower : name of the output object - indexofinput : name of the input - indexofoutput : name of the output Objects in their own .xml files: - Box - StateMachine - State - Timeline
About
No description, website, or topics provided.
Resources
Stars
Watchers
Forks
Packages 0
No packages published