[ros-users] PR2 fingertip sensor based grasping

Joe Romano jrom at seas.upenn.edu
Wed Jun 15 21:53:48 UTC 2011


Hi Ibrahim,

There is a package that basically does this:

http://www.ros.org/wiki/pr2_gripper_sensor_action<http://www.ros.org/wiki/pr2_gripper_sensor_action#Advanced>

You would want to use the "event_detector" action (
http://www.ros.org/wiki/pr2_gripper_sensor_action#Advanced).
http://www.ros.org/doc/api/pr2_gripper_sensor_msgs/html/msg/PR2GripperEventDetectorCommand.html

Here is an tutorial example of it being used to detect an object being
placed on the table:
http://www.ros.org/wiki/pr2_gripper_sensor_action/Tutorials/Grab%20and%20Release%20an%20Object%20Using%20pr2_gripper_sensor_action


Currently, it is capable of detecting 3 things:
- Contact accelerations via the accelerometer inside the PR2 hand. This
works best for hard contacts (so probably not great for clothes, but maybe
if you control the conditions just right).
- Side impacts with the pr2 grippers fingertip pressure sensors
- Slip signals along the fingerpad of the pr2 pressure sensors.

It is tuned with default parameters to have generally good performance (good
signal to noise ratio for generic robot motions), however, it is definitely
possible to tune it to do a much better job of detecting things like hitting
cloth if you have known constraints on your problem. For example: If you
know that you are going to move with some specific small motion, it is
possible to lower the threshold values significantly. If you'd like to
contact me we can discuss some more off the mailing list and I can provide
you with more detail based on your specific application.

Here is a slight warning/disclaimer:
As you said, the touch from brushing against clothing can be very light and
faint. I worry, that for light contacts, the change in pressure will be
barely discernible (your signal to noise ratio on the sensors will be very
low), and that you may not always get a good detection. It is a fairly easy
thing to test however.


-Joe

On Wed, Jun 15, 2011 at 5:18 PM, Ibrahim Awwal
<ibrahim.awwal at berkeley.edu>wrote:

> On Wed, 15 Jun 2011 14:17:27 -0700, Ibrahim Awwal <
> ibrahim.awwal at berkeley.edu> wrote:
>
> Hi all,
>
> In our work with clothing at Berkeley, we often need to grasp clothing
> articles. Currently we basically approach a point that we think the clothing
> article is at and alternate  closing the gripper and moving closer until the
> gripper does not close all the way, which is not terribly reliable (false
> negatives occur sometimes with thin articles) and also slower than it could
> be. We would like to be able to simply slide the gripper under the article
> and then close when an article is detected by the fingertip sensors. I have
> looked at some data and it seems like it should be possible to detect a
> change in pressure resulting from the gripper moving under a clothing
> article, although the response is somewhat small compared to say, me
> touching the gripper sensor with my finger.
>
> I just wanted to check before I go ahead and do this, is there anything out
> there that already does this? I.e. something where the gripper moves toward
> a point and closes once it senses a pressure increase. It seems like the
> current fingertip pressure controllers are intended for situations where you
> first position the gripper in the correct place and then the sensors are
> used to apply just enough force to keep the grasp instead of closing all the
> way every time.
>
> One thing I was wondering about is how do people generally deal with the
> fingertip pressure readings? It seems like there is noticeable drift in the
> values over time, so do you just re-zero the data periodically, or is there
> a better way to compensate? Also, I would appreciate any tips on how to go
> about creating such a controller, because I've never done anything with low
> level control on the PR2. Thanks!
>
> --
> -Ibrahim Awwal
>
>  Oops, I meant this for ros-users, >.<
>
> --
> -Ibrahim Awwal
>
>
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ros.org/pipermail/ros-users/attachments/20110615/39c52af7/attachment-0002.html>


More information about the ros-users mailing list