/py_trees_js

Javascript library for behaviour tree visualisation, demo'd with a qt-js hybrid application

Primary LanguagePythonOtherNOASSERTION

PyTrees Js

[About] [Features] [Preview] [Exmaple - Simple Web App] [Example - PyQt App] [The JSON Specification]

About

Libraries for visualisation of runtime or replayed behaviour trees.

  • ./js - a self-contained javascript library to build apps around
  • py_trees_js - a python package that makes the js available as a pyqt resource
  • py_trees_js.viewer - a demo pyqtwebengine app

See py_trees_ros_viewer for a fully fledged pyqt integration that uses py_trees_js.

Features

  • Visualise the runtime state of a behaviour tree
  • Collapsible subtrees
  • Zoom and scale contents to fit
  • Timeline rewind & resume
  • Blackboard key-value storage view
  • Activity log view

Although designed for py_trees, the js libs (in particular, the interfaces) are not dependent on py_trees and could be used for other behaviour tree applications.

Preview

With VSCode DevContainers and on a PC with an NVIDIA GPU:

$ git clone https://github.com/splintered-reality/py_trees_js
$ code .
$ cd py_trees_js

# Reopen the project in the devcontainer
$ (use CTRL-SHIFT-P if you miss VSCode's helper dialog)

# Setup and launch
$ poetry install
$ poetry run py-trees-demo-viewer

If you do not have a PC that meets those requirements, some alternative options:

  • Install Poetry and PyQt on your system or in a venv. Clone and launch.
  • If you're just interested in seeing the demo viewer, pip install --user py_trees_js and launch the viewer
  • Create your own devcontainer with something like the desktop-lite feature. If this works, send me a PR!

Example - Simple Web App

Building a complete application that can render a behaviour tree stream is an effort that can be decomposed into two tasks:

  1. Creating the web app for rendering trees and visualising a timeline
  2. Wrapping the web app in a framework and connecting it to an external stream

The first stage is purely an exercise with html, css and javascript. The latter will depend on your use case - it could be a qt-js hybrid application (as exemplified here) for developers, an electron application for cross-platform and mobile deployment or a cloud based service.

This section will walk through how to build a web application with the provided js libraries. An example of wrapping the web app within a Qt-Js application will follow.

To get started, let's begin with a basic html page with two divs, one for the tree canvas and one for the timeline:

<!doctype html>
<html>
<head>
  <meta charset="utf-8">
  <title>PyTrees Viewer</title>
  <style>
    html {
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
    body {
      margin: 0;
      overflow:hidden;  /* no scrollbars */
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
  </style>
</head>
<body>
  <div id="canvas"></div>
  <div id="timeline"></div>
</body>
</html>

Next, bring in the javascript libraries. For exemplar purposes, it is assumed here that the libraries have been made available alongside the html page - how is an integration detail depending on the mode of deployment (see next section for an example).

Note that the py_trees-<version>.js library has only one dependency, jointjs, but that in turn has a few dependencies of it's own. The bundled libraries in the js/jointjs folder of this repository correspond to the requirements for a specific version of jointjs and have been tested to work with the accompany py_trees-<version>.js library.

You can verify that the libraries have been properly imported by calling py_trees.hello() which will print version information of the loaded javascript libraries (if found) to the javascript console.

<!doctype html>
<html>
<head>
  <meta charset="utf-8">
  <title>PyTrees Viewer</title>
  <link rel="stylesheet" href="js/py_trees-0.6.0.css"/>
  <link rel="stylesheet" type="text/css" href="js/jointjs/joint-3.0.4.min.css"/>
  <script src="js/jointjs/dagre-0.8.4.min.js"></script>
  <script src="js/jointjs/graphlib-2.1.7.min.js"></script>
  <script src="js/jointjs/jquery-3.4.1.min.js"></script>
  <script src="js/jointjs/lodash-4.17.11.min.js"></script>
  <script src="js/jointjs/backbone-1.4.0.js"></script>
  <script src="js/jointjs/joint-3.0.4.min.js"></script>
  <script src="js/py_trees-0.6.0.js"></script>
  <!-- Web app integration css here -->
  <style>
    html {
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
    body {
      margin: 0;
      overflow:hidden;  /* no scrollbars */
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
  </style>
</head>
<body>
  <script type="text/javascript">
    py_trees.hello()
  </script>
  <div id="window">
    <div id="canvas"></div>
    <div id="timeline"></div>
  </div>
</body>
</html>

Output from py_trees.hello():

********************************************************************************
                               Py Trees JS

 A javascript library for visualisation of executing behaviour trees.

 Version & Dependency Info:
  - py_trees:  0.6.0
    - jointjs :  3.1.0
       - backbone:  1.4.0
       - dagre   :  0.8.4
       - jquery  :  3.4.1
       - lodash  :  4.17.11
********************************************************************************

In the next iteration, the canvas is initialised and a callback for accepting incoming trees from an external source is prepared. To test it, pass it the demo tree provided by the library.

<!doctype html>
<html>
<head>
  <meta charset="utf-8">
  <title>PyTrees Viewer</title>
  <link rel="stylesheet" href="js/py_trees-0.6.0.css"/>
  <link rel="stylesheet" type="text/css" href="js/jointjs/joint-3.0.4.min.css"/>
  <script src="js/jointjs/dagre-0.8.4.min.js"></script>
  <script src="js/jointjs/graphlib-2.1.7.min.js"></script>
  <script src="js/jointjs/jquery-3.4.1.min.js"></script>
  <script src="js/jointjs/lodash-4.17.11.min.js"></script>
  <script src="js/jointjs/backbone-1.4.0.js"></script>
  <script src="js/jointjs/joint-3.0.4.min.js"></script>
  <script src="js/py_trees-0.6.0.js"></script>
  <!-- Web app integration css here -->
  <style>
    html {
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
    body {
      margin: 0;
      overflow:hidden;  /* no scrollbars */
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
  </style>
</head>
<body>
  <script type="text/javascript">
    py_trees.hello()
  </script>
  <div id="window">
    <div id="canvas"></div>
    <div id="timeline"></div>
  </div>
  <script type="text/javascript">
    // rendering canvas
    canvas_graph = py_trees.canvas.create_graph()
    canvas_paper = py_trees.canvas.create_paper({graph: canvas_graph})

    render_tree = function({tree}) {
      py_trees.canvas.update_graph({graph: canvas_graph, tree: tree})
      py_trees.canvas.layout_graph({graph: canvas_graph})
      if ( canvas_graph.get('scale_content_to_fit') ) {
        py_trees.canvas.scale_content_to_fit(canvas_paper)
      }
      return "rendered"
    }
    render_tree({tree: py_trees.experimental.create_demo_tree_definition()})
  </script>
</body>
</html>

At this point, your web app should be visualising a single tree and zoom/collapse/scale to fit interactions functional. I'm happy, you should be too!

Adding a timeline to the application is optional, but the code does not change significantly and is a very useful feature to have. The built-in demo app's index.html does exactly this. The code is reproduced below for convenience.

<!doctype html>
<html>
<head>
  <meta charset="utf-8">
  <title>PyTrees Viewer</title>
  <link rel="stylesheet" href="js/py_trees-0.6.0.css"/>
  <link rel="stylesheet" type="text/css" href="js/jointjs/joint-3.0.4.min.css"/>
  <script src="js/jointjs/dagre-0.8.4.min.js"></script>
  <script src="js/jointjs/graphlib-2.1.7.min.js"></script>
  <script src="js/jointjs/jquery-3.4.1.min.js"></script>
  <script src="js/jointjs/lodash-4.17.11.min.js"></script>
  <script src="js/jointjs/backbone-1.4.0.js"></script>
  <script src="js/jointjs/joint-3.0.4.min.js"></script>
  <script src="js/py_trees-0.6.0.js"></script>
  <!-- Web app integration css here -->
  <style>
    html {
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
    body {
      margin: 0;
      overflow:hidden;  /* no scrollbars */
      height: 100%  /* canvas is intended to fill the screen, cascading heights achieves this */
    }
  </style>
</head>
<body>
  <script type="text/javascript">
    py_trees.hello()
  </script>
  <div id="window">
    <div id="canvas"></div>
    <div id="timeline"></div>
  </div>
  <script type="text/javascript">
    // rendering canvas
    canvas_graph = py_trees.canvas.create_graph()
    canvas_paper = py_trees.canvas.create_paper({graph: canvas_graph})

    // event timeline
    timeline_graph = py_trees.timeline.create_graph({event_cache_limit: 100});
    timeline_paper = py_trees.timeline.create_paper({
        timeline_graph: timeline_graph,
        canvas_graph: canvas_graph,
        canvas_paper: canvas_paper,
    })

    // react to window resizing events
    $(window).resize(function() {
      py_trees.canvas.on_window_resize(canvas_paper)
      py_trees.timeline.on_window_resize(timeline_paper)
    })

    render_tree = function({tree}) {
      py_trees.timeline.add_tree_to_cache({
          timeline_graph: timeline_graph,
          canvas_graph: canvas_graph,
          canvas_paper: canvas_paper,
          tree: tree
      })
      return "rendered"
    }
  </script>
</body>
</html>

Example - PyQt App

The py-trees-demo-viewer app is a qt-js hybrid application using qtwebengine. Every time a qt button is pressed, an internally generated tree snapshot is sent to render_tree() in the embedded web application. From here, it is not too hard to imagine connecting the qt application to an actual external source. The qt layer then acts as a shim or relay transferring messages to the internal web app.

How does it work?

  • The js libs are made available as a .qrc resource [1]
  • A simple web app is made available as another .qrc resource
  • Both resources are consumed by the QWebEngine View to serve the app

[1] This can be made available separately and as a dependency to the actual pyqt application. For instance, the py_trees_js package is a dependency of py_trees_ros_viewer.

In more detail...

The JS Libraries

  1. Bundle the javascript resources into a .qrc file
  2. Generate the resources as a c++ library / python module
  3. Deploy the c++ library/python module in your development environment

In this case, the py_trees and jointjs javascript libraries have been listed in py_trees_js/resources.qrc, generated using py_trees_js/gen.bash, resulting in the importable module py_trees_js/resources.py. From this point, any pythonic Qt application wishing to visualise behaviour trees need only import this module from the py_trees_js package.

The Web App

  1. Bundle the .html/.css pages into a .qrc file
  2. Import into directly into designer when building your Qt application

In this case, our web app (py_trees_js/viewer/html/index.html) has been rolled into py_trees_js/viewer/web_app.qrc which is directly loaded into py_trees_js/viewer/web_view.ui where the URL property of the QWebEngineView widget has been configured with the resources index.html.

You could alternatively, generate a module from the .qrc and import that into the relevant python code as was done for the javascript resources.

The Qt Application

The Qt application can be designed in whatever way you're most comfortable with - via Designer, pure C++ or python. In this case, Qt's Designer is used to produce the .ui files which are generated into python modules and finally customised and brought together as a PyQt5 application. Refer to py_trees_js/viewer for more details or as a reference example from which to start your own Qt-Js hybrid application.

Key elements:

  1. Build your Qt application around a QWebEngineView widget
  2. Link/import the javascript module in the web engine view class
  3. Load the html page into the QWebEngineView view

Do not use the QWebView widget - this is deprecating in favour of the QWebEngineView widget. The most notable difference is that QWebView uses Qt's old webkit, while QWebEngineView makes use of Chromium's webkit.

Note that the second step automagically makes available the javascript resources to the application when it's loaded. It's not terribly fussy about where it gets loaded, see py_trees_js/viewer/web_view.py for an example:

# This is the module generated by running pyrcc5 on the js libraries .qrc
# It could have been equivalently deployed in a completely different python package
import py_trees_js.resources

Loading the web page can be accomplished in designer. Simply point it at your qresource file and set the dynamic URL property on the QWebEngineView widget. Alternatively you can import the resource module and load it via QWebEngineView's load api.

Qt-Js Interactions

Qt and JS can interact directly over snippets of javascript code (via runJavaScript() or over QWebChannel (a mechanism similar to sigslots) where more complexity is needed. The example application here calls on the render_tree() method we created earlier in the web application to send trees to the app. Example code from py_trees_js/viewer/viewer.py which handles button clicks to cycle through a list of demonstration trees:

def send_tree_response(reply):
    console.logdebug("reply: '{}' [viewer]".format(reply))


@qt_core.pyqtSlot()
def send_tree(web_view_page, demo_trees, unused_checked):
    demo_trees[send_tree.index]['timestamp'] = time.time()
    console.logdebug("send: tree '{}' [{}][viewer]".format(
        send_tree.index, demo_trees[send_tree.index]['timestamp'])
    )
    javascript_command = "render_tree({{tree: {}}})".format(demo_trees[send_tree.index])
    web_view_page.runJavaScript(javascript_command, send_tree_response)
    send_tree.index = 0 if send_tree.index == 2 else send_tree.index + 1

send_tree.index = 0

The JSON Specification

TODO: A JSon schema

Roughly, the specification expects json objects of the form:

  • timestamp: int
  • behaviours: dict[str, dict]
  • (optional) visited_path: list[str]
  • (optional) blackboard: {
  • behaviours: dict[str, dict[str, str]],
  • data: dict[str, str]
  • }
  • (optional) activity: list[str]

where each behaviour in the dict has specification:

  • id: str
  • status: Union[INVALID,FAILURE, RUNNING, SUCCESS]
  • name: str
  • colour:
  • (optional) children: List[str]
  • (optional) data:

Identification strings (id's) must be unique and are used as both keys for the behaviours dictionary, children and visited_path variables.

An example (extracted from py_trees.experimental.create_demo_tree_definition()):

{
    timestamp: 1563938995,
    visited_path: ['1', '2', '3', '4', '5', '7', '8'],
    behaviours: {
        '1': {
            id: '1',
            status: 'RUNNING',
            name: 'Selector',
            colour: '#00FFFF',
            children: ['2', '3', '4', '6'],
            data: {
                Type: 'py_trees.composites.Selector',
                Feedback: "Decision maker",
            },
        },
        '2': {
            id: '2',
            status: 'FAILURE',
            name: 'Worker',
            colour: '#FFA500',
            children: ['7', '8', '9'],
            data: {
                Type: 'py_trees.composites.Sequence',
                Feedback: "Worker"
            },
        },
    }
    'blackboard': {
        'behaviours': {  # key metadata per behaviour
            '2': {
                '/parameters/initial_value': 'r',
                '/state/worker': 'w'
            },
        },
        'data': {
            '/parameters/initial_value': 'foo',
            '/state/worker': 'bar',
        },
    'activity': [
        "<text style='color: blue;'>Worker initialised with 'foo'</text>'",
        "<text style='color: red;'>Worker wrote 'bar'</text>'",
    ]
}