Re: [simspark-devel] [Rc-ss3d-tc] Immediate help is needed for preparation!

[Patrick M. <pa...@cs...>]

Hi all,

The following is a brief summary and my recommendations on proposed changes.

Increasing ball mass increases kick distance due to the ball having more
momentum and rolling farther when it hits the ground.  Increasing the ball
size seems to have a similar effect.  Decreasing the foot height also seems
to allow for a small gain in distance and added height due to it being a
little easier to get under the ball.

I'm not in favor of increasing the ball mass or size too much as things
begin to look unnatural.  With a heavier ball the effect of friction is
reduced and the ball seems to roll farther than it probably should.  As the
ball size increases it starts to look disproportionately big compared to
the robot and also rolls farther than what looks normal.  Tests I did also
suggested that a larger ball actually produced less height although I was
not optimizing for height (just distance).  If either ball mass or size is
to be increased I would only do so by a little bit, and err on the side of
caution, as such changes could produce odd looking physics and have a
negative effect on game play.

I'm not opposed to decreasing the foot height.  The foot of the simulated
Nao is already different from the real Nao as it is a box and not a curved
and sloped downward surface like the real Nao.  Aldebaran's specs
officially list the height of the foot as 45.19mm but this is the distance
from the bottom of the foot to the ankle joint.  The actual Nao foot is
sloped such that at its shortest point (located at the front and center of
the foot ) the height is only 20mm (I just measured this).  Currently the
height of the box-shaped Nao foot in simulation is 30mm.  My experiments
showed a small improvement in kicking distance and height when decreasing
the height of the foot.  As such a change doesn't potentially adversely
affect the physics of the ball I think it might be a viable option.

One question I would ask is do we really need to make changes?  My
experiments show that with the current state of the simulator one can
already kick 12m,  which is over half the length of the field, and also
kick 2.5X the height of a Nao (although this was done in a controlled
situation of beaming the ball in front of the agent).  I would expect teams
to get better at kicking for this year's upcoming competition, which is a
focus of our team, in a similar way to which walks have progressed and
become faster over the past couple of years.

Another suggestion that has been brought up is increasing the size of the
field.  I'm not really in favor of of this as such a change will make it
harder to win the ball on defense, and then transition forward to score on
offense, which I feel is an important characteristic for the sport and
excitement of soccer; especially if we are already debating ways to
possibly increase the length of kicking relative to the size of the field.
I believe the average time it takes for the ball to travel the length of
the field in the 3D simulation league is longer than many of the other
leagues (certainly longer than real soccer), and I would rather not add to
this time.  If we do increase the size of the field I think we should also
add more players to keep a similar ratio of players to open space.

Regards,
Patrick





On Tue, Mar 20, 2012 at 5:24 AM, Hedayat Vatankhah <hed...@gm...>wrote:

>  Hi Klaus, Patrick and others!
>
> Thanks a lot for your efforts. We can increase the ball size. However,
> about the foot height, I'd personally prefer to have our robot dimensions
> follow the real Nao. If I have understood correctly, decreasing the foot
> height should have the same effect as increasing the ball size. So, isn't
> it enough to slightly increase the ball size (without having an unusually
> big ball)? And it would be great if we can send the initial suggestion to
> sserver-three-d mailing list today. What do you suggest?
>
>
>  Regards,
>
> Hedayat
>
> *Klaus Dorer <kla...@hs...> <kla...@hs...>*wrote on Fri, 09 Mar 2012 07:02:14 +0100:
>
> Hi Patrick,
>
> thanks a lot for your work!
>
> Stefan has also evaluated some settings. Below his findings.
> His conclusions are quite close to yours. Increasing ball mass seems odd,
> at least as long as friction is kept the same. He also suggests only
> marginal increase in ball size.
> The only difference is with respect to foot height. He suggests to shrink
> the foot.
> In my oppinion this would allow more flexibility, i.e. the initial
> situation does not require too much precision or put in other words,
> kicking high from various ball distances and situations can be achieved.
>
>
> Greetings
> Klaus
>
> Stefan:
>
> The magmaOffenburg agent is capable of two different forward kicks, one
> that is wider, like a normal kick, and one that is faster, called FastKick,
> that is more like rising the leg while standing. It's hard to describe, but
> the FastKick should basically lift the ball up, but less wide.
>
> Results:
> 1) foot height: 1.5cm, ball radius: 4.2cm (old)
>     - 13g (old/2):
>         - Kick: ball hardly leaves the ground
>         - FastKick: shoulder to head height (relatively reliable)
>     - 26g (old):
>         - Kick: ball hardly leaves the ground
>         - FastKick: head height (reliable)
>     - 39g (old+13):
>         - Kick: ball hardly leaves the ground
>         - FastKick: head height (reliable)
> 2) foot height: 1.5cm, ball radius: 5.2cm
>     - 13g (old/2):
>         - Kick: knee to hip height - it is easily able to push the foot
> fast enough below the ball, but then the ball doesn't get much momentum and
> simply drops down again (like a trick kicker, balancing the ball)
>         - FastKick: hip height - it isn't realy able to get below the
> ball. The FastKick accelerates the leg directly at the ball and the ball
> seems to take the acceleration easier when it is leighter.
>     - 26g (old):
>         - Kick: same behavior as with 13g
>         - FastKick: head height (reliable)
>     - 39g (old*2):
>         - Kick: ball hardly leaves the ground
>         - FastKick: head height (reliable)
>
> In general with a heighter foot, like e.g. 2cm, our kicks aren't realy
> able to get below the ball. If I rise the size of the ball radius by
> another 1cm (to 6.2cm), the ball slowly is as high as the knee and it
> happens somethimes that it touches the ball with its knee. At that size our
> kicks are also not realy capable any more to kick the ball appropriately.
> With some tweaks it may well be possible to kick it better, but at that
> size it also looks strange...
>
> Overall I realized that the weight of the ball has less impact on if it's
> high kickable. On the other hand, the size of the ball has less impact on
> the kickable distance.
> The following observations were made:
> - if the ball weight is reduced, the friction has a higher impact on the
> ball's speed, reducing possible kick distances drastically. If it is
> enlarged, the friction has less impact on the ball's speed, but it quickly
> results in an odd ball movement. It is rolling very long and stopping too
> slow.
> - If the ball gets too big, it is hard to get the foot of the agent
> stretched below the ball in such a way that it kicks proper off the ground.
> But as logical consequence, it has a bit less friction and gets farer.
> - If the ball is bigger, it is easier to kick in different directions
>
> After several hours ot testing I think that both values for the ball are
> well balanced. According to wikipedia, a soccer ball has a radius of around
> 10cm. We use 4.2cm, wich is around half the size, but I think we have to
> consider more the ratio of foot height to ball radius. We have a robot that
> is 1/3 of the size of a human, so it should get around 1/3 of the foot
> height. If I take a rough estimation of the height of the box of my
> personal foot, I would say it's on average around 3cm in the front. With
> this, to change the foot height to a value between 1 to 1.5 cm sounds
> reasonable to me and the results also show that with a thinner foot around
> 1.5cm, kicking high is not the biggest problem any more, even without
> changing the ball parameter too much.
>
> The kick distance is more problematic. Wikipedia says that the
> international ball weight is above 400g. We currently have 26g... As I sad,
> rising the ball's weight is not really the best option as long the field
> friction is kept the same. We may be able to adjust the values slightly,
> e.g. a ball radius around 5cm and a weight around 30g, to make a step
> towards more distance in kicking, but I would need to make some other
> experiments to fine tune my suggestion for real values. In the above
> experiments with 5.2cm radius and 39g weight, the ball movement was already
> quite odd and the gain in distance was from initially 4.5m to around 9m -
> not even half the field.
>
> In general the results are just my opinion and other teams may well be
> able to kick higher and/or wider with the same parameters. My suggestion
> with respect to well balanced values for foot height, ball radius and
> weight are:
> - foot height: 1 to 1.5cm
> - ball radius: 4.5 to 5cm
> - ball weight: the same
>
> The suggestion to the field size is hard. I would suggest to adapt to a
> field size of 1/3 of the internaltional commen field size for humans. This
> would be around 1.75 times the current size.
>
> The extension of the HipYawPitch to a range of -120 to 60 could be an
> option, since it fits better to the human leg range. Although in my opinion
> the above range is a bit too much and a change in the HipYawPitch itself is
> not enough. If we allow for a wider range here, we also need to adjust the
> range of the other two hip joints to allow the intended leg-freedom for
> kicking. But, over the last months I also came to the conclusion that the
> Nao is allready quite flexible and this kind of leg-feedom doesn't bring
> too much benefit. We have kicks going to the right, left, front and back if
> the agent is standing still. I could also come up with kicks in e.g. 10
> degrees steps, but as long as I have to stand still, there is no real
> benefit in more kick directions. And if I can shoot while walking, I should
> walk better ;) Humans also often do a final step before kicking.
>
> So, I hope I could help you with my suggestions and thoughts to clarify
> the values for the upcomming events. If you need any further help or have
> some questions about my tests/thoughts, you are very welcome to ask.
>
>
> Am 08.03.2012 05:24, schrieb Patrick MacAlpine:
>
> Hi all,
>
> Based on the proposed changes to the 3D simulation competition,
> particularly that of kicking, I've optimized one of our kicks for
> distance through machine learning in order to create some benchmarks
> for how changes might affect things.  The following are numbers I've
> collected on optimizing a kick for which the ball is beamed to a set
> place in front of the agent.  While these numbers show what is
> possible in the server, we have made no attempt yet to integrate the
> kick into our agent such that it is able to walk up to and position
> behind the ball before kicking it.  These numbers should be considered
> a lower bound for what is possible in the server, but with the caveat
> that it may be hard to achieve their full potential without very good
> positioning.
>
> Current server: Distance = 12m with a height of 1.4m+ (2.5X height of a
> Nao)
> Ball at half mass (.013): Distance = 8m with similar height above
> Ball at double mass (.052): Distance = 17m+ with similar height above
> Ball at +50% radius (.063): Distance = 16.5m with height greater than
> Nao although lower for longer kicks
> Foot Z half value (def $FootSizeZ  0.015): Distance = 12.5m with a
> height of 1.8m (3X height of a Nao)
>
> Based on the numbers above it seems that the goals of kicking above
> the height of a Nao (which was not something that the kicks learned
> were optimized for) as well as a distance 2/3 the length of half field
> (which is 10.5m if the field is increased by 50%) are already
> attainable.  Decreasing the mass of the ball will only decrease the
> length of kicks whereas increasing the mass of the ball would
> increase kick length.  This increase in distance with more mass occurs
> because once the ball hits the ground it rolls farther due to it
> having more momentum with greater mass.  I'm not really in favor of
> increasing the size of the ball as it starts to look a little silly
> with its larger size relative to the Nao and also a larger size will
> make it harder to get the ball by a goalie and score goals.  The
> larger ball rolls farther after hitting the ground when compared to
> the regular size ball, which is the reason for longer distance kicks,
> and makes me think that increasing the ball's radius also increases
> its mass. Decreasing the height of the foot seems to add a little
> distance as well as height to kicks but not by a considerable amount.
> If there are other changes to server parameters or combinations of
> server parameters that you wish to see benchmarked let me know and I
> will try to run them (time permitting).
>
> I'm OK with changes to the field size if it makes things better match
> the ratio of the real world, however if we do this I think we should
> also match in the number of players (11).  This way the amount of open
> space and area which individual players need to cover is also similar
> to the real world.
>
> I like the idea of adding heterogeneous players and think this would
> add a really interesting component to this year's competition.  We're
> currently exploring how modifying the dimensions of the robot model
> affects aspects of play, such as an agent's walking speed and
> stability, so as to find a range of parameter values that can be
> changed to create heterogeneous robot models without unbalancing
> gameplay.
>
> Regards,
> Patrick MacAlpine (UT Austin Villa)
>
>
>
> On Fri, Mar 2, 2012 at 1:57 PM, Hedayat Vatankhah<hed...@gm...><hed...@gm...>
> wrote:
>
> Dear all,
>
> As you can find in the attached document, TC has decided for a few
> adjustments/enhancements in the simulator for 2012 competitions. It is
> really nice if we can get these changes ready ASAP, so that we can use at
> least some of these changes in upcoming open competitions (IranOpen and
> DutchOpen). Sorry for being a little late :(.
>
> Anyway, according to the document, most of the changes are some parameter
> adjustments. We need your help to find some suitable values to suggest and
> request feedback from the teams. The primary parameters to adjust are:
> - ball size
> - ball weight
> - foot height
> - possibly foot shape if that is possible
>
> And the secondary parameters to adjust is then:
> - field size
>
>
> It's great if we can have some values in less than a week and then discuss
> those values in the list with all teams. Any help is highly appreciated!
> :)
>
> Thanks
>
> Hedayat
>
>
>
> ------------------------------------------------------------------------------
>
> Virtualization&  Cloud Management Using Capacity Planning
> Cloud computing makes use of virtualization - but cloud computing
> also focuses on allowing computing to be delivered as a service.
> http://www.accelacomm.com/jaw/sfnl/114/51521223/
> _______________________________________________
> Simspark Generic Physical MAS Simulator
> simspark-devel mailing list
> sim...@li...
> https://lists.sourceforge.net/lists/listinfo/simspark-devel
>
>  _______________________________________________
> rc-ss3d-tc mailing list
> rc-...@li...
> http://lists.robocup.org/listinfo.cgi/rc-ss3d-tc-robocup.org
>
> _______________________________________________
> rc-ss3d-tc mailing list
> rc-...@li...
> http://lists.robocup.org/listinfo.cgi/rc-ss3d-tc-robocup.org
>
>
>
> ------------------------------------------------------------------------------
> This SF email is sponsosred by:
> Try Windows Azure free for 90 days Click Here
> http://p.sf.net/sfu/sfd2d-msazure
> _______________________________________________
> Simspark Generic Physical MAS Simulator
> simspark-devel mailing list
> sim...@li...
> https://lists.sourceforge.net/lists/listinfo/simspark-devel
>
>