• Some general info about the Vicon system
• The software I implemented for it
  • what it does
  • how to use it
  • where to find the documentation

• Cameras should be arranged such that markers are seen from different angles.
• Markers on an objects should be far apart.
• Cameras need sufficient overlap for successful calibration.
• Regular re-calibration is needed (Vicon recommends every day).

Goal:

Get object poses relative to a known, fixed world origin, persistent over re-calibrations.

Idea:

Use a fixed Vicon subject as world origin → origin subject

Problem:

Vicon does not support this...

Solution:

Use arbitrary origin in Vicon Tracker and transform all objects to origin subject frame.

→ ViconTransformer

• World origin is defined by the "Pamy_left_1" object (former "rll_ping_base")
• For more robustness, combined with other ceiling markers → "Pamy_ceiling"

• vicon_transformer: Vicon-related but project-agnostic
• pam_vicon: PAM-specific (uses vicon_transformer)
• spatial_transformation: generic transformation-related code
  • Transformation class (uses quaternions for rotational part)
  • Compute transformation between point clouds

Important Classes:

• ViconReceiver: Get raw data from Vicon Tracker (via network)
• PlaybackReceiver: Get raw data from recorded file
• ViconTransformer: Wrapper around Receiver that transforms all subjects to reference frame of the origin subject.
• o80Driver/o80Standalone: Template classes for o80 integration

Executables:

• vicon_print_data: Print raw Vicon data to terminal
• vicon_record: Record Vicon data for later playback

Important Classes:

• o80Driver/o80Standalone: Implementation for PAM objects
• PamVicon: Wrapper around o80 front end with convenience methods (get_robot_pose(), get_table_pose())

Executables:

• vicon_o80_standalone: Run o80 back end
• vicon_o80_print_data: Print Vicon data received through o80
• record_tennicam_vicon_trajectory: Record data for aligning tennicam
• compute_tennicam_to_vicon_transform: Compute tennicam transform

from vicon_transformer import ViconReceiverConfig, ViconReceiver, ViconTransformer

def main():
    config = ViconReceiverConfig()

    with ViconReceiver("vicon_pc_hostname", config) as receiver:
        vt = ViconTransformer(receiver, "my_origin_subject")

        # get new frame from receiver
        vt.update()

        if vt.is_visible("robot_arm_marker"):
            robot_tf = vt.get_transform("robot_arm_marker")
            ...

from pam_vicon import PamVicon


def main() -> None:
    # requires vicon_o80_standalone to be running
    SEGMENT_ID = "vicon"
    pv = PamVicon(SEGMENT_ID)

    # update with latest Vicon data provided through o80
    pv.update()

    robot_pose = pv.get_robot_pose()
    table_pose = pv.get_table_pose(yaw_only=True)

    print(f"Robot Base Pose: {robot_pose}")
    print(f"Table Pose: {table_pose}")

Note: Table pose is computed based on corner positions

• Run tennicam_client_display with --vicon to set table pose from Vicon.
• Test by moving ball along edges of table.
• Does currently not include robot.

Added two options in HysrOneBallConfig:

• use_vicon: If true set initial robot and table pose via Vicon.
• vicon_segment_id: Shared memory ID used by o80 (default "vicon").

List of objects for o80 is hard-coded in pam_vicon/o80.hpp:

constexpr std::size_t NUM_SUBJECTS = 11;

enum Subjects  // <-- changes here also need to be updated in Python bindings (srcpy/o80.cpp)
{
    PING_BASE = 0,
    BALL_LAUNCHER_FRONT,
    BALL_LAUNCHER_HEAD,
    TABLE_CORNER_1,
    // ...
};

const std::map<std::string, size_t> _subject_name_to_index = {
    {"Pamy_ceiling", Subjects::PING_BASE},
    {"Ballmaschine Frontmarker", Subjects::BALL_LAUNCHER_FRONT},
    {"Abschusskopf Marker", Subjects::BALL_LAUNCHER_HEAD},
    {"Eckteil 1", Subjects::TABLE_CORNER_1},
    // ...
};

Goal:

Set object origins based on robot CAD data (where marker positions are exactly known).

Problem:

Vicon does not support this...

Solution (so far):

Create object, then manually edit XML file to fix marker positions.

New solution:

Use vskeditor.py to edit marker positions in the file.

• Script: https://gitlab.is.tue.mpg.de/-/snippets/40
• Demo Video: https://asciinema.org/a/THBnqgVbQ6IzQhfWV3wSgIoy4

• Re-calibrate Tennicam-Vicon-transformation.
• Objects of new robot need to be tested and potentially modified to be more robust.
• Visualise individual robot links (base, upper arm, lower arm) based on Vicon.
• Estimate joint angles based on Vicon.

• Packages (links to package documentation in READMEs)

  • vicon_transformer: https://github.com/intelligent-soft-robots/vicon_transformer
  • pam_vicon: https://github.com/intelligent-soft-robots/pam_vicon
  • spatial_transformation: https://github.com/MPI-IS/spatial_transformation

• Additional documentation in pam_documentation:

  • https://intelligent-soft-robots.github.io/pam_documentation/C9_vicon.html
  • https://intelligent-soft-robots.github.io/pam_documentation/C9.1_tennicam_vicon_transform.html

• Vicon objects files: https://atlas.is.localnet/confluence/display/AGBS/Vicon+Objects