Re: Temporal Learning
- From: Traveler <traveler@xxxxxxxxxx>
- Date: Sat, 01 Oct 2005 08:34:06 -0400
On 01 Oct 2005 07:29:51 GMT, curt@xxxxxxxx (Curt Welch) wrote:
>Traveler <traveler@xxxxxxxxxx> wrote:
>> On 30 Sep 2005 22:30:36 GMT, curt@xxxxxxxx (Curt Welch) wrote:
>
>> There is a group at CMU that has developed an impressive AIBO
>> programming framework called Tekkotsu. I am not sure if any of those
>> frameworks will work with early AIBOs.
>>
>> http://www.cs.cmu.edu/~tekkotsu/
>
>Nope, it doesn't work on mine. :( I've got the ERS-110 and it only works
>on the ERS-2xx series and ERS-7.
>
>> I'll check it out to see if it supports primitive joint motion
>> capabilities (start/stop). IMO, this is a must for a learning program.
>
>I spent some time looking it over. It looks like you can do whatever you
>need. It's all C++ real time control code sends a constant stream of
>commands to the joints as well as receive real time updates from the
>senors. You put the real time joint control code in one process which
>can't block or hang, and your sensor processing and decision code goes in a
>main process and is allowed to hang processing.
>
>I never saw a clear description of what the lowest level commands you could
>send to the joints were, but it seems you can set PID values for each
>joint. But I don't know if the PID values are controling power, speed, or
>position to each servo.
>
>It seems you get to send updated data every 8 ms so that's 125 per second.
>But it's in batches of 4 commands, so I think around 31.25 times per
>second, your code gets updated information on the position of every joint,
>and gets to send another batch of 4 motion control commands. No matter
>what is actually being sent, it should be straight forward to write a joint
>control algorithm to respond to any type of commands such as start/stop you
>want to create.
This is very interesting. The important thing is to get real time
access to the sensors and effectors. The update frequency is more than
adequate, IMO. Besides, it gives the program time to process all the
sensory data.
I understand that the wireless bandwidth allows streaming audio and
video. This is very important (especially the video part) because
intelligent visual processing should figure prominently in any
demonstration of true AI so as to leave no doubt in anyone's mind. I'd
love to see this puppy do clever things that no one would expect from
any dog, especially a mechanical one. Now, all I need is time and
money. Until then, I'll keep on dreaming. :-)
Louis Savain
Why Software Is Bad and What We Can Do to Fix It:
http://www.rebelscience.org/Cosas/Reliability.htm
.
- References:
- Re: Temporal Learning
- From: Curt Welch
- Re: Temporal Learning
- From: Traveler
- Re: Temporal Learning
- From: Curt Welch
- Re: Temporal Learning
- Prev by Date: Re: Temporal Learning
- Next by Date: Re: Temporal Learning
- Previous by thread: Re: Temporal Learning
- Next by thread: Re: Temporal Learning
- Index(es):
Relevant Pages
|