More
See all Show me
1. TOTO - tracking of thrown objects
2 months ago
The intention of this work is to create a new transport technique in manufacturing. In fully automated production systems, parts should be transported between workstations by robotic throwing and catching. There are two solution approaches to the transport-by-throwing task: aimed throwing (always hit the target) or active catching (tracking the catching device).

This is a successful first implementation of the active catching approach. Until now only tennis-balls are caught. A single standard gray-value camera observes the flight. An industrial gantry-robot tracks a mechanical gripper to a predicted interception position. This robot is realized at the Reinhold-Würth-University of the Heilbronn University in Künzelsau (Germany), in cooperation with the Institute of Computer Technology, TU Vienna (Austria).

Credits

43 Likes

  • Blake Whitman staff 2 months ago
    god help us.
  •  
  • Soxiam staff 2 months ago
    i am really really scared.
  •  
  • Andrew Pile staff 2 months ago
    Hell yes!
  •  
  • This comment has been deleted.


  • Dennis Barteit 2 months ago
    budweiser? football?

    no, sorry.

    but, perhaps, it will work with a real good german beer (held-braeu.de/) and a fat guy watching soccer...
  •  
  • phantomcolor plus 2 months ago
    what kind of parts would you consider "throwable"? and would you consider the application best suited for long range throwing?
  • Dennis Barteit 2 months ago
    We are currently working on aimed throwing of cylindrical objects. The accuracy is good enough, so we don't need to track the catching device. This would be the desirable appplication: no additional tracking robot. However, in long range throwing (>4-5m) this would get more complicated to achieve. We are still searching for possible applications.
  •  
  • Joe Moya plus 2 months ago
    So... has the robot been approached by the baseball sport agents? ...If not... it's only a matter of time.
  •  
  • pekka tarvainen 2 months ago
    I remember our math teacher telling us a comparison between humans and future robots catching a ball; the computations were supposedly so complicated that a machine will probaby never be able to calculate so fast :-)
    Now, the things we presently consider improbable or downright impossible...
  • Dennis Barteit 2 months ago
    It is also good to prove that your teacher was wrong :-) And all the other people saying that this won't work...

    There is interesting research about human catching skills. For example, it was studied how field players in baseball "predict" a possible catching position. It is a rather simple strategy, that was also found to be used by dogs catching frisbees (I guess, it was the same strategy). A Japanese science lab already realized a free moving robot that catches balls using this strategy. The control of human interception movement is kept as simple as possible. Hence, it doesn't seem to be a big thing to adopt it to a robot. However, there are some problems...
  •  
  • Evan Bassett 2 months ago
    I'm amused because I recognize the motors and cables used as being ones I work with every day.
  • Dennis Barteit 2 months ago
    We tried to use off-the-shelf industrial components.
  • Evan Bassett 1 month ago
    Makes sense. If you don't mind me asking, were they B&R controls? I mean obviously you have a PC in there somewhere doing the math with the camera, but I'm curious about the motion portion of the system.
  • Dennis Barteit 1 month ago
    Yes, a PC is used to detect the object in the image data, calculate the 3-D position, and predict the future flight trajectory. From this, an interception position is determined and the position information is sent to a Siemens SIMOTION motion controller (automation.siemens.com/_en/mc/mc-sol/en/2ea3f697-5b50-4d8f-b13f-80e338afad9f/index.aspx?c=r-simotion). The SIMOTION controls the movement of the robot's axes.
  •  
  • Jason Hu 1 month ago
    Good job, Dennis. As I have done a similar robot, which kicks a throwing a ball up and down in a settled area, I wonder how you calibrated your ground coordinates with one camera. I mean the camera distortion might be a problem for such an accurate positioning device. How did you modify it?
  • Dennis Barteit 1 month ago
    Thanks, Jason. Of course, calibration is a problem with a fixed camera. I have some calibration sheets at the robot (you won't see them in the clip) for the determination of extrinsic camera parameters. Thus, the operational plane of the robot defines a world coordinate system. Calibration is done with the openCV library (sourceforge.net/projects/opencvlibrary/). Also the correction of radial and tangential distortion is done with openCV. However, the intrinsic camera parameters are obtained by the Matlab camera calibration toolbox and a couple of images of a calibration sheet (vision.caltech.edu/bouguetj/calib_doc/). The distortion is not that large, 'cause I use a 8.5 mm lens.
  •  
  • bilderbuchi 1 month ago
    very nice work! *proud of TUW* :)
    i take it you know this video already: ebaumsworld.com/video/watch/80731612 ?
  • Dennis Barteit 1 month ago
    Yes, I know the robot hand of the Ishikawa Lab. They are doing really great work there. In the first publication, I know of them, in 2003, they presented a catching robot arm that could accelerate with 91G. It was equipped with a active stereo-vision system that could permanently track an object with 1000 fps. Really great work!
  •  
  • Matthew Jackson 29 days ago
    Hi Dennis, I work for a press release company and have just stumbled across your video of TOTO, its amazing. I would love to write an article on your Robot. Could you please let me know who the main suppliers of your components are. In particular I'm looking for component manufacturers such as Festo, Bosch Rexroth or Rockwell Automation. I just need to check if we have clients who have supplied you with components, so that I can write the story. Contact me directly @ matthewj@cubicice.com Kind Regards Matt
  •  
This conversation is missing your voice. Take five seconds to join Vimeo or log in.

Advertisement

2 Related collections

Statistics

  •  
    plays
    likes
    comments
  • Total
    plays 51.2K
    likes 43
    comments 20
  • Nov 21st
    plays 31
    likes 0
    comments 0
  • Nov 20th
    plays 52
    likes 0
    comments 0
  • Nov 19th
    plays 27
    likes 0
    comments 0
  • Nov 18th
    plays 250
    likes 0
    comments 0
  • Nov 17th
    plays 49
    likes 0
    comments 0
  • Nov 16th
    plays 653
    likes 0
    comments 0
  • Nov 15th
    plays 807
    likes 0
    comments 0
  • Nov 14th
    plays 36
    likes 0
    comments 0
Previous Week

Downloads

Please join Vimeo or log in to download the original file. It only takes a few seconds.