[robocup-athome] Rulebook discussion?

Sun Feb 28 13:22:07 PST 2010

Hello Thomas,

thank you for your useful and constructive comments on the rule book.
I'll give you my (personal) views on them.

Thomas Breuer schrieb:
> Hi together,
>
> we (team b-it-bots) have a few questions and suggestions:
>
> ----------------------------------------------------------------
> Robot Inspection and Poster Session:
> ----------------------------------------------------------------
> - Is the environment known in this test?
I'd say this would make sense.
> - What exactly (!) should the robot do to register itself?
There are some examples in the text on what to say. I don't think this
is really important, what matters is that your robot is operative, able
to go to the desk, say some sentences and preferably hand over the
registration form.
> - Where is the poster presentation? Who will judge the posters? The 
> time for doing so is too few (7 minutes for self registration AND 
> poster session)
We'll have many teams and not much time this year, that's why the
schedule for this test is so narrow. You'll have two minutes for the
registration. After that, the team leader has 5 minutes for a poster
presentation and answering questions from the other team leaders.
Parallel to that, someone from the TC will inspect the robot and the
robot has to leave autonomously (So e.g. 4 min. inspection and 1 minute
leaving).

Probably this is going to be hectic somehow, however the time
limitations are just there. The only solution I see to this would be to
reduce the number of teams that are accepted to participate in the
competition.

>
> ----------------------------------------------------------------
> Follow Me:
> ----------------------------------------------------------------
> - Too few time for all the tasks. Either tasks have to be reduced or 
> time limit has to be extended. Suppose, that a robot needs 30-60 
> seconds to reach the next checkpoint, then there will be nearly no 
> time anymore to fullfil the given tasks.
You have five minutes for this test. Suppose the robot needs 1 minute to
calibrate on the operator. It needs 30 seconds to get from one
checkpoint to the next (4x). Then you have 2 minutes left for the tasks.
If you manage to do the the first and second task in 30 seconds and the
third one in one minute, you'll cross the finish line right in time.

The tests are going to be more difficult than in the last year in
general, however. This is intended. Maybe no team is going to be able to
achieve the full score in this test and several others.
> - Checkpoint 3:  The owner, who has to be recognized is the 
> professional walker (whose face is learned at the beginning), is this 
> correct? This should be made clear.
Agreed.
>
> ----------------------------------------------------------------
> Go Get It!:
> ----------------------------------------------------------------
> - 2 objects are lying on the floor. What happens, if the robot 
> collides with one of them or drives over them? We suggest that the 
> robot is allowed to collide with the objects on the floor (at least if 
> they are not visible for the laser scanner).
Good point, this should be made clear in the rules. I'm not sure about
allowing the robot to drive over anything lying on the floor.

Anyway, if your robot drives over one the objects and falls over,
according to the current version of the rules I'd say then your team can
be disqualified for dangerous behaviour.

The 'laser scanner' rule however makes no sense in this form, as there
is no requirement that robots have any laser scanners on them.
> - Why does the robot just have one try to solve a subtask? What 
> happens, if the robot fails to grasp the object. Is it not allowed to 
> try to grasp the object again? What happens, if a robot identifies an 
> object, which is not graspable for it? Can it then identify another 
> one to grasp it? How do the referees know that a robot tried to solve 
> a sub-task? Or is meant, that a robot will just get 1 time points for 
> identifying and grasping etc? We suggest to delete this rule / sentence.
The idea behind this is that it should not be possible to solve the task
by just getting as many objects as possible in the time, without
explicitly identifying them.
But your objections are right. I would suggest making this less
restrictive by giving a penalty for unsuccessfully grasping an object,
or grasping the wrong one, as (in a similar form) in Who is Who.
>
> ----------------------------------------------------------------
> Who Is Who:
> ----------------------------------------------------------------
> - We suggest, that the provided names are common / easy names.
This is how it is planned to be. We'll have to work on a clearer
definition of that point. Maybe a general rule would make sense, as for
the predefined objects.
> - We suggest, that if the robot stands next to a person (in range of 1 
> meter), that it is allowed to advise the person to look into the 
> camera. It should be to allowed that the robot always (without being 
> next to a person) says, that the person should look into the camera.
It is intentional that the persons are not necessarily looking into the
camera to increase complexity. However, I would be okay with saying that
the robot may tell a person to look at it while trying to recognize it.
However, if the robot asks something to look into its camera which isn't
a human, this would result in a -150 points penalty, as it is now.
>
> ----------------------------------------------------------------
> Enhanced Who Is Who:
> ----------------------------------------------------------------
> - We suggest, that if the robot stands next to a person (in range of 1 
> meter), that it is allowed to advise the person to look into the 
> camera. It should be to allowed that the robot always (without being 
> next to a person) says, that the person should look into the camera.
see above
> - Who defines the graspable drinks? Can the team select own drinks? We 
> suggest, that each team can select their own drinks.
I'd be okay with that.
> - In this test also there is too few time to solve all tasks. There 
> should be delivered just one or two drinks.
I think you're right, however if you look at the other stage 2 games,
it's going to be pretty hard to achieve the full score there as well, so
you could say it's balanced in some kind of way.
> - Is it allowed to directly grasp all 3 drinks, or does one drink have 
> to be grasped, then delivered and then the next drink grasp etc. This 
> will be unfair for teams which have just one manipulator and / or are 
> not able to carry more than one drink at a time. We suggest, that 
> either all teams have to grasp one drink at a time, or it should be 
> allowed to grasp all drinks in a row and put them back on the table to 
> get the points for grasping.
I wouldn't support that view. The task is to deliver three drinks. If
the robot has the ability to transport more than one drink, then it is
better suited for the task and at a justified advance. I don't see why
this should be unfair.

I'd say just grasping drinks and putting them back is currently possible
following the rules, since it's not explicitly forbidden, however this
will cost you extra time if you're planning to actually deliver them.

> - What does deliver mean? Does it mean to hand over the drink, or is 
> it ok to have the drink at the robot and the robot says something like 
> "Here is youe Ice-Tea, please take it from my platform"?
That's not specified yet. We could split the points so handing over
gives more points than letting the user grab it from some kind of tray.
> In sense of fair play, we suggest, that every team is just allowed to 
> grasp and deliver one drink at a time.
I don't agree. If a robot is able to deliver multiple objects at the
same time, why not reward that?
> - Since the highest score last year was only 1050 points, we suggest, 
> that this test is not furhter complicated. We suggest either to take 
> the test of the last year, or to reduce the complexity of the test. 
> Our suggestion is the following: The order for the drinks is given, 
> then it is searched for persons (before grasping), if the first person 
> is found, it is identified and the correct drink is delivered. Then it 
> is searched for the next person and the according drink is delivered. 
> Otherwise it is luck, if you grasp a drink and find the corresponding 
> person (e.g. it could be, that you find all persons, but not the 
> person whose drink the robot grasped at the beginning).
The persons move by max. 2 meters and don't change their pose
(standing/sitting), so you can search for persons first and then decide
to which person you deliver the drink. The problem of re-encoutering the
same person is less difficult than finding him/her in the first place.
> - We further suggest, that the team places the drinks, such that the 
> robot could grasp them. This esures, that the drinks stand not too 
> close to each other or too far away (out of the range for grasping)
This would mean the robot does not have to recognize the drinks, as you
can put them in a predefined position. So I don't agree on that.

We should define general constraints on the placement of objects in this
test, however. Maybe there should be a general rule for all tests (e.g.
max. 15cm from border, min. 15cm space between objects).
>
> ----------------------------------------------------------------
> General Purpose Service Robot
> ----------------------------------------------------------------
> - Task is too abstract. It has to be more conrete.
We're working on that in the TC. If you have concrete suggestions, don't
hesitate to tell us.
> - Are the speech commands given by a trained user? We suggest, that 
> the speech commands are given by a team member.
Sounds plausible to me.
>
> ----------------------------------------------------------------
> Shopping Mall
> ----------------------------------------------------------------
> - It is not defined, whether the robot  can grasp and deliver all 
> three objects at a time. Due to fair play (robots which cannot carry 
> more than one object or just have one manipulator), we suggest, that 
> just one object can be delivered at a time.
Same as for Enhanced Who's Who.
> - Once it is said, that the robot is guided by a team member, and then 
> it is said that an unknown user shows the robot the three location. We 
> suggest, that a team member guides the robot and gives the speech 
> commands.
That's contradictory, of course. I think that a team member would be
okay, regarding the overall complexity of this test.
> - How many objects are in each shelf? Are there many of the same 
> objects standing very next to each other? To be able to grasp the 
> objects, we suggest, that from each object just one is standing in the 
> shelf with enough room around it to be able to grasp it.
This needs to be specified more clearly. However, I think it would be
interesting to have many objects of the same type standing next to each
other.
> - Can good lighting conditions be assumed in this test?
This is probably going to be a real supermarket. There won't be extra
lighting. As far as my supermarket experience goes, this probably means
that there's going to be light from fluorescent lamps attached to the
ceiling.

Regards,
David (team homer)
>
>
> Cheers,
> Thomas (team b-it-bots)
> _______________________________________________
> robocup-athome mailing list
> robocup-athome at lists.robocup.org
> http://lists.robocup.org/listinfo.cgi/robocup-athome-robocup.org
>