Re: [robocup-rescue-s] Complaint

Dear Alireza,

> Firstly I guess both Babak and Arash are assuming we did reverse
> engineering to write our AmbulanceTeam agents in Impossibles.

Well! Your guess is not right! I did not say such a thing in my
last e-mail.

> We never even read the source code of MiscSimulator but every one
> else working in other teams like Caspian and S.O.S had completely read the
> source code.

Well! As far as I know, in S.O.S. we used a 3-layer MLP neural net
for estimating the deathTime.

> I completely agree with Babak in the definition of RoboCup. The solutions
> should be MultiAgent and AI solutions. Assume a team is using
> Reinforcement Learning for estimating the death time of a civilian. We
> think it needs 5,000 runs, each run at least 300 seconds. we think it will
> take a month to have a estimation with 90% precision for the death time of
> civilians.

I have no idea of how to make use of Reinforcement Learning for
obtaining a deathTime estimator. RL is intended to learn policies
i.e. learning how to do something right. RL is not a supervised
learning method. Learning process in RL is completely based on
reward/penalty functions. This means that we do not have a set
of input/output pairs. Instead, we have a measure of how good
our actions are. RL is neither a data-mining method or a method
for function approximation. Therefore, adapting a supervised
learning problem to a reinforcement learning one does not seem
to be a good idea.

So, your ideas are unclear and strange for me and I am not able to
assess in the way you assessed the time needed to estimate deathTime.
But be sure, 5000 runs is too much for any purpose. Be sure more than
80% of the pieces of the data obtained by 5000 runs are redundant.
Generally, 20 to 30 runs on carefully designed situations are more
than enough for estimation purposes.

Best Regards,
Arash

> On Tue, May 9, 2006 4:38 pm, Babak BehSaz wrote:
>> Dear Alireza,
>>
>>
>> I have two comments about your complaint.
>> First, I agree that technical committee should release
>> the final package sooner than this (sooner than release time of package
>> 0.49a). I think technical
>> committee should try to solve this problem for next years. I think it
>> will
>> promote quality of competitions dramatically. But, I think Arash's
>> argument
>> is logical, why don't you complained when package 0.49a was released?
>> This
>> change was announced four months ago and now, I think complaining is
>> not
>> acceptable. Second, let's look what is RoboCup! According to
>> explanation of Recoup official site, "RoboCup is an international joint
>> project to promote AI, robotics, and related field. It is an attempt to
>> foster AI and intelligent robotics research by providing a standard
>> problem
>> where wide range of technologies can be integrated and examined".
>> Therefore, we technically
>> expect that team's efforts promote AI. Thus, technical committee can
>> assume
>> teams designed their agent according to the concepts and methods of AI.
>> With this assumption, recent changes made by technical
>> committee doesn't have any problems (and in my opinion, they should try
>> to
>> apply such changes in entire package more and more). They just changed
>> the
>> precision of data that will be received by agents, not even model of
>> generation of these data. Hence, it is obvious that if your team uses an
>> AI method, you will
>> not confront any problems. Otherwise, it is your problem that you are
>> not
>> selected a method consistent with the definition of RoboCup. Look at
>> most
>> of current teams' methods, THESE METHODS IS DISASTROUS, not recent
>> changes. I regret to see that our league made a very few contributions
>> to
>> promotion of AI, of course if made any (our teams are only using simple
>> methods to obtain better results without even trying to use a more
>> scientific approach!).
>>
>> Best Regards,
>> Babak Behsaz
>> Previous S.O.S. Member
>>
>>
>>
>> __________________________________________________
>> Do You Yahoo!?
>> Tired of spam? Yahoo! Mail has the best spam protection around
>> http://mail.yahoo.com
>> _______________________________________________
>> robocup-rescue-s mailing list robocup-rescue-s@mailman.cc.gatech.edu
>> https://mailman.cc.gatech.edu/mailman/listinfo/robocup-rescue-s
>>
>>
>
>
> _______________________________________________
> robocup-rescue-s mailing list
> robocup-rescue-s@mailman.cc.gatech.edu
> https://mailman.cc.gatech.edu/mailman/listinfo/robocup-rescue-s
>

_______________________________________________
robocup-rescue-s mailing list
robocup-rescue-s@mailman.cc.gatech.edu
https://mailman.cc.gatech.edu/mailman/listinfo/robocup-rescue-s
Received on Tue May 09 18:54:50 2006