firefoxmetzger / scikit-bot Goto Github PK

This is likely a more general bug in the generation of XSD from meta-SDF. It looks as if the required tag is not overwritten correctly when an element is //included in the meta-SDF.

Ignition uses ign-transport to communicate between different parts of the library, and we can introspect this communication using ropy.Subscriber.

Under the hood, this uses a mix of zmq and the ignition CLI (via subprocess) to make itself known to ignition and to start receiving messages. The main reason why this isn't implemented cleaner is that I didn't find documentation on how topic discovery works in ignition. However, there appears to be some documentation (living here) that describes how ignition passes messages around to discover topics.

A great addition would be to implement this discovery protocol here, so that we can cleanly interface with Ignition.

Possible related issue: gazebosim/gz-transport#225

when running tests, the file ropy/ignition/tests/test_subscriber.py produces a lingering process every time ignition gazebo is started to test subscription.

It will be better if the test could clean up after itself and remove any child and grandchild processes.

If a pose element is omitted, it is set to the default value "0 0 0 0 0 0"; however it should be set to a pose object with pose.value = "0 0 0 0 0 0" instead.

Reproducing example:

import skbot.ignition as ign

sdf_root = ign.sdformat.loads(
"""<?xml version="1.0" ?>
<sdf version="1.8">
  <model name="some_name"> 
    <link name="link1">
      <pose relative_to="frame1">1 0 0 0 0 0</pose>
    </link>
    <frame name="frame1" />
  </model>
</sdf>
"""
)

print(sdf_root.model.frame[0].pose)
# prints: "0 0 0 0 0 0"

Testing the ignition interface opens the simulator to interface with it. Currently, this is done via subprocess.Popen and without any checks that the simulator is ready. Instead, we just time.sleep(3) and hope it will have initialized after this time. This works most of the time, but might randomly break. It also wastes time as the test may wait much longer than it needs to.

What is needed is a mechanism to detect if ignition is running and has finished initializing.

See for example:

https://scikit-bot.org/en/latest/_autosummary/sdformat_v1.8/skbot.ignition.sdformat.bindings.v18.sensor.Sensor.html#skbot.ignition.sdformat.bindings.v18.sensor.Sensor

The type of the attribute is not extracted and a new-line is missing at the beginning of the description. This is likely an xsData bug.

Currently ropy.transform has functions for rigid body transformations (i.e. coordinate transformation, translation + rotation), and N-dimensional rotations. A welcome addition would be to add functions that allow easy access to other basic transformations.

The following functions are still needed:

• tf.translation N-dimensional translation
• tf.scaling N-dimensional scaling (scale each dimension differently)
• tf.uniform_scaling scale each dimension by the same factor (i.e. update scale parameter)
• tf.reflection N-dimensional reflection
• tf.shear N-dimensional shearing

The version agnostic SDFormat bindings are an effort to make SDFormat backward compatible at the python level.

The idea is that you can load any version of SDFormat into a version-agnostic object tree, and then consume this tree as if it were in any (other) SDFormat version. In other words, if you have written an algorithm that assumes SDFormat v1.6 then this algorithm should also work when you input SDFormat v1.4 or SDFormat v1.8 .

The current implementation supports reading SDFormat into an object tree, and conversion of that tree into a pose graph (A tutorial will hopefully come soon(TM).)

Currently, the important elements (model, joint, link, frame) are implemented; however, less frequently used elements still need implementation. Below is a TODO list of all the elements (up to v1.8) that need implementation. Implementation is needed for all elements that are not checked.

If you need/want a specific element, feel free to comment on this issue. If you are interested in contributing a missing element, please also comment here to avoid duplicate effort.

Element Status:

• Actor
• Air Pressure
• Altimeter
• Atmosphere
• Audio Sink
• Audio Source
• Battery
• Box Shape
• Camera
• Capsule
• Collision Engine
• Collision
• Contact
• Cylinder Shape
• Ellipsoid Shape
• Force Torque
• Frame
• Geometry
• GPS
• Gripper
• Gui
• Heightmap Shape
• Image Shape
• Imu
• Inertial
• Joint
• Lidar
• Light State
• Light
• Link State
• Link
• Logical Camera
• Magnetometer
• Material
• Mesh Shape
• Model State
• Model
• Navsat
• Noise
• Particle Emitter
• Physics
• Plane Shape
• Plugin
• Polyline Shape
• Population
• Projector
• Ray
• Rfid
• Rfid Tag
• Scene
• Sdf (root)
• Sensor
• Sonar
• Sphere Shape
• State
• Surface
• Transceiver
• URDF
• Visual
• World

In addition to a perspective projection via perspective_frustum there are a few more functions that are viable to include:

• perspective_focal(focal_length, image_shape) which is the same as perspective_frustum but takes the focal_length directly
• parallel(direction, image_shape) which gives us access to the family of parallel projections
• stereographic(radius, image_shape) which gives us access to stereographic projections
• perspective_frustum in N dimensions

Bonus points if we can write them in a general, N-dimensional way. This would give us some really nice visualization power for higher dimensional spaces (e.g. joint-space). In this case, the proposed signature may have to change.