Rock

the Robot Construction Kit

... at runtime

This page is going to describe the runtime configuration interface of the components that use the transformer oroGen plugin. It will then describe runtime support for the transformer in the context of Ruby scripting, and data display. The integration in the system management layer will come in the next pages.

Component Interface

Components that use the transformer oroGen plugin get the following interface elements:

  • normal stream aligner properties. Each aligned port has a port_name_period property that allows to set its period. The transformer has a transformer_max_latency property that allows to override the default set in the transformer declaration block
  • a static_transforms property that allows to set the transformations that are known and do not change at startup, and
  • a dynamic_transforms port that allows to send changing transformations to the transformer. The data must be sent using the base/samples/RigidBodyState type. Since the source and target frames are stored in the data structure, all streams that are
  • finally, for each declared frame, a _frame_name_frame property that allows to set the actual frame name for the component’s internal frame.

If, for any reasons, one wants to read dynamic transformations from a port separate from dynamic_transforms, it is made possible by manually feeding the data to the transformer C++ object. This must be done before the call to TaskBase::updateHook

base::samples::RigidBodyState rbs;
while(_my_transform_input.read(rbs, false) == RTT::NewData)
    _transformer.pushDynamicTransform(rbs);

Transformation Specifications

The transformation specification file is a Ruby file that lists the known static and dynamic transformations in a system. It is usually stored in a config/transforms.rb file.

Static transformations are declared with

static_transform translation, rotation,
  "source_frame_name" => "target_frame_name"

Translation and rotations are resp. specified with Eigen::Vector3d and Eigen::Quaternion. The very useful Eigen::Quaternion.from_axis(angle, axis) call is often used in these files. The translation or rotation can be omitted if they are identities.

Dynamic transformations are declared with

dynamic_transform "producer_task_name.port_name",
  "source_frame_name" => "target_frame_name"

Where producer_task_name and port_name refer to the data stream that defines the required transformation. If the task has only one RigidBodyState output port, the port name can be left empty.

For our laser filter example, a possible configuration file would look like:

# The frame names in this file are arbitrary !
static_transform Eigen::Vector3.new(0.3, 0, 0),
  "Body" => "ServoFrom"
static_transform Eigen::Quaternion.from_angle_axis(1, Eigen::Vector3.UnitX),
  "ServoTo" => "Lidar"
dynamic_transform "servo.transform_samples",
  "ServoFrom" => "ServoTo"

Ruby scripting

When using the components in Ruby scripts, some automatic configuration can be achieved.

Three bits are needed to configure the system properly:

  1. the transformation declaration file (see above)
  2. proper configuration of the _frame_name_name properties on the components. These properties are required for every frame that is required during computation, and for the components that generate frame transformations (as e.g. the servo driver compontn). In our example, we therefore need to setup the laser filter and the servo.
  3. declarations that tell in which frame the data produced by the tasks is expressed in. This is not strictly required, but allows proper configuration of the 3D display (video below).

Finally:

# You need to initialize before you can use the transformer
Orocos.initialize
Orocos.transformer.load_conf('config/transforms.rb')
Orocos.run ... do
  lidar = Orocos.name_service.get 'lidar'
  filter = Orocos.name_service.get 'lidar_filter'
  servo = Orocos.name_service.get 'servo'

  # Tell in which frames is the data expressed. These are not
  # arbitrary anymore: they MUST match the frame names listed
  # in the transformer's
  lidar.lidar_samples.frame = 'Laser'
  filter.filtered_samples.frame = 'Laser'

  # Properly setup frame names. These are not arbitrary anymore:
  # they MUST match the frame names listed in the transformer's
  # configuration files
  filter.body_frame = "Body"
  filter.input_frame = "Laser"
  servo.source_frame = "ServoFrom"
  servo.target_frame = "ServoTo"

  lidar.lidar_samples.connect_to(filter.lidar_samples)

  # Finalize transformer configuration (see below for explanations)
  # For static transformations the task should not yet be configured
  Orocos.transformer.setup(lidar, filter, servo)

  lidar.configure
  lidar.start
  filter.configure
  filter.start
  servo.configure
  servo.start

  # Wait for ENTER on input
  STDIN.readline
end

The Orocos.transformer.setup call provides the following additional features:

  • it verifies that all required xxx_frame properties are set with frames that exist in the transformer configuration files,
  • it connects the necessary frame producers to the dynamic_transformations ports of the tasks that require it. In this example, it means that the servo’s transform_samples output is connected to the filter’s dynamic_transformations input,
  • it provides the necessary static transformations in the corresponding task properties,
  • finally, it saves and broadcasts the system transformation configuration so that it can be picked up by e.g. the 3D display

TODO: video