Overview:

• Dependency & how to compile the tutorial
• How to communicate with the superquadric-model
• How to run the superquadric-model + tutorial
• Setting up before running
• How to run the overall pipeline

This example module relies on

• IOL

for the blob extraction process.

mkdir build; cd build
ccmake ..
make install

Go to the top

The user can provide the 3D point cloud (a set of points in the 3D space representing the object surface) and ask the estimated superquadric in two modes:

1. streaming
2. one-shot

Streaming: In this case, the user should send the 3D point cloud to the module, for example:

  pointPort.open("/testing-module/point:o");
  Bottle &b1=point.addList();

  for (size_t i=0; i<points.size(); i++)
  {
      Bottle &b=b1.addList();
      b.addDouble(points[i][0]); b.addDouble(points[i][1]); b.addDouble(points[i][2]);
  }

  pointPort.write();

connecting to the corresponding buffered port of the superquadric-model:

yarp connect /testing-module/point:o /superquadric-model/point:i

The estimated superquadric can be read from another buffered port

yarp connect /superquadric-model/superq:o /testing-module/superq:i

in the format:

(dimensions (x0 x1 x2)) (exponents (x3 x4)) (center (x5 x6 x7)) (orientation (x8 x9 x10 x11))

where:

• dimensions are the semi-axes lenghts of the superquadric;
• exponents are the responsible for the superquadric shape;
• center are the coordinates of center of the superquadric;
• orientation is the axis-angle orientation (derived from the Euler angles: x8, x9, x10 mentioned before).

One-shot: In this mode, the user can ask a simple query to the superquadric-model just sending a specific 2D blob and asking for the corresponding estimated superquadric. In this case, the port to use in the rpc port:

yarp connect /testing-module/superq:rpc /superquadric-model/rpc

An example of code is the following:

Bottle cmd, reply;
cmd.addString("get_superq");

Bottle &in1=cmd.addList();

for (size_t i=0; i<points.size(); i++)
{
    Bottle &in=in1.addList();
    in.addDouble(points[i][0]);
    in.addDouble(points[i][1]);
    in.addDouble(points[i][2]);
}

// Add 1 instead of 0 if you want the filtered superquadric
cmd.addInt(0);

⚠️ The get_superquadric service can also deal with an extra parameter for resetting the median filter used for improving the superquadric estimation:

//1 is for resetting the superquadric, 0 for keeping using the previous computed superquadrics.
// Usually, the filter needs to be reset when the object or its pose change.
// If nothing is specified, the filter is never reset.
cmd.addInt(0);

superqRpc.write(cmd, reply);

Note for one-shot mode: Using the thrift tool for the rpc services gives the computed superquadric (reply in the previous example) with extra parenthesis:

((dimensions (x0 x1 x2)) (exponents (x3 x4)) (center (x5 x6 x7)) (orinetation (x8 x9 x10 x11)))

If you need to handle it, use it as a Bottle in this format, rather than a Property.

Go to the top

If you want to test the superquadric-model code without writing your own code, you can use the this tutorial, following the following steps:

1. Launch the basic modules of the iCub (yarprobotinterface, iKinGazeCtrl, iKinCartesianSolver - both for right and left arm-, wholeBodyDynamics, gravityCompensator, imuFilter) and the cameras.
2. Launch the IOL modules. This xml file contains all the required modules.
3. Launch the superquadric-model and a yarpviewer. You can use this xml file.
4. Launch the tutorial code with this xml.
5. Connect everything.

You can now play with the /testing-module/rpc port:

• setting the name of the object you want to model

yarp rpc /testing-module/rpc
>>set_object_name object

• setting if you want the streaming mode or not

>>set_streaming_mode on

Enjoy! 😃

Go to the top

In order to achieve the desired performance, it is recommended to solve the following practical issues:

The calibration consists of:

• Launching SFM and two yarpviews with this xml. The viewers show the disparity map and the features matching between the cameras.
• Set the eye_vergence=5.0.
• Showing to the iCub a no-flat scene;
• Through the proper RcpPort, asking SFM to perform calibration, until ACK reply is received

yarp rpc /SFM/rpc
>> calibrate
[Ack]
>> save
[Ack]

If a error message is received instead of Ack, show a more cluttered and no-flat scene.

The only step to be performed with IOL before running the superquadric-model is to teach the robot the object name.

• Launch IOL.
• Click on the object of interest in the /iolViewer/manager/localizer viewer.
• Through the proper RpcPort:

yarp rpc /iolStateMachineHandler/human:rpc
>>this <object_name> click
[ok]

Go to the top

This module is designed in order to be executed together the superquadric-grasp module. A complete tutorial on how to execute them together and perform on the robot a complete modeling and grasping task is provided in this repository.

Go to the top