simspark-devel

[simspark-devel] About 3D Devel Plan

[Feng X. <hen...@ma...>]

Hi all,

First, happy new year and wish you have a good time in the year 2009!

And here is my situation, we(WrightEagle) decided not to participate simlation3D this year because of being short of hands. And I am now a little busy with my paper and SPL league. But I will try my best to finish my assignment.

Considering the previous discussion about the OGRE stuff, I think I'd better probably give up this task for 2009 and here is my opinion: our simulator has these 3 key problems which are kind of restricting or shrinking our research area.

First, no noise added on the controlling joint speed. This ideal thing will never happen on the real robot. And, because of the elasticity of the material of the real robot, the real angle is not that reported by the sensor. However, ODE does not support elastic joint or soft things. But we can simulate this by adding noise to it.

Second, the sync model which is already in the todo list. According to games of ChinaOpen2008, making a good sync model will let the simulator run more stable agents. Then the strategy becomes much more important. The dream of integrating 2D and 3D will probably come true. I think the strategy is a great challenge because it is hard to scale the speed of opponent agent which is relatively easier in 2D.

Third, the information that agent received from the server is too few. I'd suggest we provide the restricted vision perceptor with the information of hands and legs. And, we'd better provide the camera perceptor which is firstly making our robot model like the real robot and secondly extend the research area. However, this makes the sensor informaion quite large. But if the sync model is good, this problem will probably be solved.




Best Regards,
Feng Xue
2009-01-07

Re: [simspark-devel] About 3D Devel Plan

[Yuan X. <xuy...@gm...>]

Hello & Happy new year!

2009/1/7 Feng Xue <hen...@ma...>:
> Hi all,
>
> First, happy new year and wish you have a good time in the year 2009!
>
> And here is my situation, we(WrightEagle) decided not to participate
> simlation3D this year because of being short of hands. And I am now a
> little busy with my paper and SPL league. But I will try my best to finish
> my assignment.

What a pity!
I am even afraid there will be not enough teams in RoboCup2009!

>
> Considering the previous discussion about the OGRE stuff, I think I'd better
> probably give up this task for 2009 and here is my opinion: our simulator
> has these 3 key problems which are kind of restricting or shrinking our
> research area.
>
> First, no noise added on the controlling joint speed. This ideal thing will
> never happen on the real robot. And, because of the elasticity of the
> material of the real robot, the real angle is not that reported by the
> sensor. However, ODE does not support elastic joint or soft things. But we
> can simulate this by adding noise to it.

Adding noise is easy for MC, but hard for teams ;-)

> Second, the sync model which is already in the todo list. According to games
> of ChinaOpen2008, making a good sync model will let the simulator run more
> stable agents. Then the strategy becomes much more important. The dream of
> integrating 2D and 3D will probably come true. I think the strategy is a
> great challenge because it is hard to scale the speed of opponent agent
> which is relatively easier in 2D.
>

But I think the sync mode is suitable for research, but not for competition.
Because the computation time of agent is an important factor, the
agent should react in real (or given) time.


> Third, the information that agent received from the server is too few. I'd
> suggest we provide the restricted vision perceptor with the information of
> hands and legs. And, we'd better provide the camera perceptor which is
> firstly making our robot model like the real robot and secondly extend the
> research area. However, this makes the sensor informaion quite large. But if
> the sync model is good, this problem will probably be solved.

I have a draft version of image camera, see this:
http://www2.informatik.hu-berlin.de/~xu/image_sensor.png
But I hacked the communication between agent and server...

-- 
Sincerely,

Xu Yuan
Institute of Informatics, Humboldt University Berlin
--------------------------------------------------

Re: [simspark-devel] About 3D Devel Plan

[Yuan X. <xuy...@gm...>]

Hi Joschka and all,


> Actually, I know of at least 2 teams from Humanoid League who are
> thinking about joining the 3D Sim-League this year, so I hope it'll be
> alright ;-)
>

Good news!


> It's sad, but understandable. You probably are very busy with the SPL
> stuff. One question though: the Aldebaran people were really
> interested in getting in touch with us to help and improve our Nao
> model. Do you think you would have some time to talk to them? I can
> get you in touch.

I also interest in improving the Nao model.
Is there anything that I can do?

> Since it's a technical issue, I would say it's a TC decision; but
> probably nobody is going to complain as long as there _is_ a
> decision ;-)
>
> True. Does anybody have a suggestion for suitable noise models?
>
> Again TC, I think. But someone in the committee should bring up the
> issues, take the initiative, and push for a deadline; otherwise
> chances are that there will be no decision (from my experience).
>

Maybe we need to CC this mail to TC ;-)


> Very cool, great work, Yuan :-) And Player integration sounds very
> nice, too!
>

Thanks.
About the Player integration, do teams need to change their code a lot for it?


-- 
Sincerely,

Xu Yuan
Institute of Informatics, Humboldt University Berlin
--------------------------------------------------

Re: [simspark-devel] About 3D Devel Plan

[Hedayat V. <hed...@ai...>]

Hi!


/*"Yuan Xu" <xuy...@gm...>*/ wrote on 01/09/2009 01:46:10 PM:
> Hi Joschka and all,
>   
>   
>> Since it's a technical issue, I would say it's a TC decision; but
>> probably nobody is going to complain as long as there _is_ a
>> decision ;-)
>>
>> True. Does anybody have a suggestion for suitable noise models?
>>
>> Again TC, I think. But someone in the committee should bring up the
>> issues, take the initiative, and push for a deadline; otherwise
>> chances are that there will be no decision (from my experience).
>>
>>     
>
> Maybe we need to CC this mail to TC ;-)
>   
Not a bad suggestion! ;)
Anyway, I think the discussion will be started soon. :)

>> Very cool, great work, Yuan :-) And Player integration sounds very
>> nice, too!
>>
>>     
>
> Thanks.
> About the Player integration, do teams need to change their code a lot for it?
>   
The communication layer should be written from scratch. There will be no 
S-expression parser/generator or any low level communication code. 
Player provides client libraries which include classes for different 
sensors and actuators, and teams should use them instead of 
communicating through network. And in fact there might be no network 
communications at all (e.g. for integrated agent). So I think it won't 
need much time and if we can finish it soon, it'll be possible to use it 
for this year's competitions.

Good luck,
Hedayat

>
>