[robocup-athome] Rulebook discussion?

[Sven Wachsmuth]

Dear Dirk,
Thank you for all your effort towards the new rule book.

Here are a few suggestions:

- Enhanced Who Is Who
    Although the task is not directed toward mobile object manipulation, 
it seems that
    in subtask 'delivering the drink' grasping seems to be critical for 
continuing the task.
    I would suggest that it is allowed to hand the drink over to the 
robot, but that you
    will loose the points for grasping.

- General purpose service robot
    This is the most critical task in the rulebook, because the skills 
of the robot needed
    are not completely clear. They very much depend on the assumptions 
behind the
    actions which are not specified.
    1. Understanding Language: Understand a long and complex spoken 
sentence is much
          more critical than asking for further missing information 
(like in the next category).
          The understanding will be mostly partially correct, so it is 
difficult to judge. The robot
          needs to know the vocabuary and language constructs used by 
the speaker.
          I would suggest that the robot is allowed to ask for repeating 
the sentence, and to
          ask for specific information that could not be extracted from 
the speech input.
    2. Understanding itself: It is not clear how the robot knows what is 
sufficient to perform
         the task. Further, it is difficult to judge a wrong question? I 
would suggest to make the
        distinction between between categories (1) and (2) that in (1) a 
person who knows the
        capabilities of the robot is giving the instruction, while in 
(2) a naive user should try to do it,
        so that the robot needs to communicate what kind of details it 
needs to know. Judging
        complete success instead of wrong questions.
    3. understand the world:
        There needs to be a list of possible scene information which 
might be questioned. Otherwise
        the task is much two open in terms of speech understanding and 
strategies how to extract the knowledge
        from the scene.
        "tricked"  questions involve much more knowledge about the world 
than in stage 1 is needed. Either
        the possible tricks are known beforehand or a complete model 
qualitative physics would need to be implemented.
        It seems to me that "tricked" questions would need to be 
implemented by a "trick" that solved the specific
        question but would not be a general capability of the robot.
        I would suggest to leave out "tricked" questions. Instead one 
could request something from an episodic memory.

- shopping mall
    it is not clear if objects are taken from the pre-defined set, if it 
is a full regular shelf, if there are other
    people in the shop. Thus, how much reality do we want? If we 
transform the real shop by too much
    constraints into another lab, the message of going into the wild 
will not be impressive.

Best regards
Sven
(Team ToBI)


Dirk Holz schrieb:
> Hi all,
>
> due to the upcoming local RoboCup events, we want to finalize the 
> rulebook as soon as possible (by the 1st of March).
>
> If you want to contribute any ideas to the rulebook or have 
> suggestions or comments, please retrieve the rulebook from
> http://www.robocupathome.org/
> read it carefully and start an according discussion here on the list.
>
> Any comments given after finalizing the rulebook will be ignored and 
> cannot make it into the final rulebook.
>
> Best,
> Dirk
> _______________________________________________
> robocup-athome mailing list
> robocup-athome at lists.robocup.org
> http://lists.robocup.org/listinfo.cgi/robocup-athome-robocup.org