Plot data from logged experiments available in Matlab with RViz

[traversaro]

Data from old experiments is available in Matlab.

After a bit of discussion on how to proceed, and given that at the moment we do not have Licenses for the Matlab Robotics Toolbox that has native ROS support, we plan to dump the data to visualize in .mat file, and to read it in C++, using one of the solution available for this.

cc @diegoferigo @MartaLorenzini

[traversaro]

cc @S-Dafarra

[traversaro]

I just realized that for the offline visualizer perhaps the best idea is to move all the related code in https://github.com/claudia-lat/MAPest .

[MartaLorenzini]

Here it is a video of the experiment simulation on RViz using an available dataset. We have found that the transformation matrices are erroneous and we are investigating upon that. We plan to do some tests with the Xsens suit to check its accuracy and as a consequence the accuracy of the data we are using. HRIrviz_simulation.ogv.tar.gz

[iron76]

Cool video. I like the visualisation (the idea of representing torques with a circular arrow is nice even though it might be an RViz idea). I have a few questions.

1. Why are the arrows so gigantic? I am assuming that at the beginning of the video there is no contact between the iCub and the human. Therefore I assume that the arrows visible at the beginning of the video are due to noise.
2. Is the distance between the iCub and the human due to the erroneous transformation matrices?
3. The joint torques visualisation is not yet there. Any plan to have those included soon?

[MartaLorenzini]

1. As regards the dimensions of the arrows, it's possible to set both force and torque scales (the force is represented by the red arrow while the torque is represented by the yellow one). Setting a proper scale neither force or torque is visible at the beginning, like we can expect. forcesDisplay.ogv.tar.gz

2. The distance between iCub and the human is actually due to wrong transformation matrices. We are investigating th reasons and we are planning to do some tests with the Xsens suit to check its accuracy.

3. The joint torques visualisation is now available. Here it is a video displaying the torques at the right and at the left hip along the y axis since they are the most significant torques developed during the bowing task, the other torques are set to zero. However it could be possible to display all the torques developed at all the joints. hipTorquesDisplay.ogv.tar.gz

[iron76]

Cool! Things are getting closer and closer to image that I had in mind. A few more questions/considerations.

1. There seems to be an out-of-sync between movements an forces. In the video forcesDisplay.ogv.tar.gz it is quite evident that forces increase before the actual contact and decrease before the contact breaking.

2. The hip position seems to be quite noisy. This was not the case in the previous video HRIrviz_simulation.ogv.tar.gz

3. Visualising the forces and torques seems quite ineffective if we keep the current choices (i.e. arrows and circles are becoming too long or excessively small). In the following some ideas to improve the visualisation:

• Linear scale. We are currently keeping a linear scale which might not be ideal for visualisation purposes. What about some other scale (e.g. logarithmic)?

• Form Factor. The form factor of the arrows (i.e. the ratio between the length and the diameter) is continuously changing (constant diameter and changing length). We might consider different solutions with the arrow becoming chubbier when exceeding a certain length.

• Arrow length/circle radius. Instead of chancing the arrow length or the circle radius, we might consider changing its color, from green (small force, small torque) to red (high force, high torque) or a combination of the two.

[MartaLorenzini]

Here it is a video of the pHRI simulation in RViz including the force plates and the forces and moments that the human exchanges with them, iCub and the forces and moments the the human exchanges with them and the torques developed by the human during the movement enhancing the torques at the hips. The simultaneous visualization of all them together could be a bit confusing. As regards the out-of-sync between movements and forces could be again due to errors in the data (as the transformation matrices). The noise issue on hip position has been solved. pHRIsimulation.ogv.tar.gz