darwin2k-users — A list for discussing issues relating to the use of Darwin2K

[Darwin2k-users] Darwin2K 0.88 released

[Chris L. <cl...@te...>]

Hi,

I've just released version 0.88.  The main updates are:

- Better installation instructions, including common platform-specific
options for Irix, Linux, and Solaris

- Sample wrapper scripts for both autoconf and configure, for Irix,
Linux, and Solaris.  These make it easier to tell Darwin2K where
libraries and headers for XForms and RTC are, and include other
platform-specific options (e.g. using /usr/bin/CC instead of g++ under
Irix).

- Preliminary support for modules whose geometry is defined by a text
file read at run-time.  This allows users to define some types of new
modules without having to actually write code.

I've also updated the RTC tarballs to make it easier to configure
Darwin2K to use RTC.

Chris

-- 
"The worst habit you can possibly acquire is to become uncritical 
towards your own concepts and at the same time skeptical towards 
those of others. Once you arrive at that state you are in the grip 
of senility, regardless of your age."       -Karl von Terzaghi

Re: [Darwin2k-users] Reproducing Sims' work in Darwin2k

[Martin C. M. <ma...@me...>]

Cool.  Thx.

Re: [Darwin2k-users] Reproducing Sims' work in Darwin2k

[Chris L. <cl...@te...>]

The "task parameters", which belong to either the high-level simulation
code or self-contained components (e.g. controllers or trajectories) are
encoded in the genome.  The main limitation is that only a fixed number
of parameters are encoded for a given problem; there's (currently) no
facility that allows for variation in length (or topology) of the task
parameters.  Adding some other representation for controller/brain code
wouldn't be that big a deal; you'd just have to change the portion of
the genetic operators that works on the task parameters (currently
uniform crossover).

Chris
-- 
"The worst habit you can possibly acquire is to become uncritical 
towards your own concepts and at the same time skeptical towards 
those of others. Once you arrive at that state you are in the grip 
of senility, regardless of your age."       -Karl von Terzaghi

Re: [Darwin2k-users] Reproducing Sims' work in Darwin2k

[Christopher L. <ch...@MI...>]

On Wed, 2001-10-03 at 17:16, Martin C. Martin wrote:
> Are any elements of the controller (or "brain") encoded in the genome? 
> From my reading of Chris' thesis, I can't see any.

I think the mission parameters are evolved, but general purpose
controllers are used, designed to work for all the evolved robots.

-the 'other' Chris

-- 
Christopher Lee <ch...@mi...>
(617) 253-5095 <http://web.mit.edu/chlee/www>
Postdoctoral Associate of Mechanical Engineering
Field and Space Robotics Laboratory
Massachusetts Institute of Technology
77 Massachusetts Ave, Room 3-472m
Cambridge, MA 02139

[Darwin2k-users] Reproducing Sims' work in Darwin2k

[Martin C. M. <ma...@me...>]

Hey guys,

Are any elements of the controller (or "brain") encoded in the genome? 
From my reading of Chris' thesis, I can't see any.

Thanks,
Martin

Re: [Darwin2k-users] Physics simulator: collisions?

[Chris L. <cl...@te...>]

Hi guys,

I was in Australia for a week or so (no mountain climbing, but hiked
some smaller hills); I'll get back to you within a few days.  (Things
are pretty insane at JPL right now given recent budget insanity.)

Chris



-- 
"The worst habit you can possibly acquire is to become uncritical 
towards your own concepts and at the same time skeptical towards 
those of others. Once you arrive at that state you are in the grip 
of senility, regardless of your age."       -Karl von Terzaghi

Re: [Darwin2k-users] Coevolving with Darwin2k

[Christopher L. <ch...@MI...>]

On Wed, 2001-09-26 at 11:10, Martin C. Martin wrote:
> Can the Darkwin 2k simulator handle the simulation, control, collision
> detection, etc. between multiple creatres?  E.g. so that two robots could
> compete to touch and cover a block like Sims did?

I wonder where Chris Leger is these days -- mountain climbing somewhere?
Well, I'll jump in until higher authority can answer better.

The lower-level dynamics libraries for D2K can handle multiple
mechanisms, I think this includes collection detection *between* the
mechanisms.  If not, it will be simple to fix.

However, the bookkeeping is not there for multiple *robots* in the
higher-level robot-simulation code (built on top of the dynamics
libraries).  This is something I plan to try to help with very soon
since I need to start simulating multiple robots for my research.  I'm
hoping it won't be too hard, but lots of classes are connected in
different ways for paring the robots descriptions and starting the
simulation, and I'm not sure exactly how deep the assumptions underlying
a single robot run.  Chris Leger can probably answer much more easily,
since he wrote the code.

-Chris

-- 
Christopher Lee <ch...@mi...>
(617) 253-5095 <http://web.mit.edu/chlee/www>
Postdoctoral Associate of Mechanical Engineering
Field and Space Robotics Laboratory
Massachusetts Institute of Technology
77 Massachusetts Ave, Room 3-472m
Cambridge, MA 02139

[Darwin2k-users] Coevolving with Darwin2k

[Martin C. M. <ma...@me...>]

Another message from the bodies and brains list:

Can the Darkwin 2k simulator handle the simulation, control, collision
detection, etc. between multiple creatres?  E.g. so that two robots could
compete to touch and cover a block like Sims did?

I'm assuming I'd have to change the evolution part, I'm just wondering about
the evaluation/simulation part.

Thanks,
Martin

Re: [Darwin2k-users] Physics simulator: collisions?

[Christopher L. <ch...@MI...>]

On Thu, 2001-09-20 at 15:37, Martin C. Martin wrote:
> I'll need support for more complex terrain (probably including a BSP)
> and the ability to import said terrain from a file that lists polys.  I
> suspect I'll have to write that myself, which isn't too scary. We can
> chat about it at lunch tomorrow.

I wrote some code to import terrains generated by terraform 

	http://terraform.sourceforge.net/

and successfully imported a 200x200 grid terrain (I haven't checked this
code back into D2K yet).  Unfortunately, the data structures used to
represent the surface in D2K are pretty huge, and a 400x400 terrain did
bad things to my comptuer (even with 256Megs of RAM).  An example with
lousy camera work is at
	http://web.mit.edu/chlee/www/turbo-rover.m1v
We can probably help to make things more efficient with some hacking.  A
video-game guru like yourself should be able to kick some butt in that
department.

 -Chris


-- 
Christopher Lee <ch...@mi...>
(617) 253-5095 <http://web.mit.edu/chlee/www>
Postdoctoral Associate of Mechanical Engineering
Field and Space Robotics Laboratory
Massachusetts Institute of Technology
77 Massachusetts Ave, Room 3-472m
Cambridge, MA 02139

Re: [Darwin2k-users] Physics simulator: collisions?

[Martin C. M. <ma...@me...>]

On 20 Sep 2001, Christopher Lee wrote:

> Ok -- I found it now:
> 
> 	http://www.genarts.com/karl/evolved-virtual-creatures.html

Yeah, and be sure to check out the MPEG at:

http://alife.ccp14.ac.uk/ftp-mirror/alife/zooland/pub/research/ci/Alife/karl-sims/creatures-demo.mpg

I'll need support for more complex terrain (probably including a BSP)
and the ability to import said terrain from a file that lists polys.  I
suspect I'll have to write that myself, which isn't too scary. We can
chat about it at lunch tomorrow.

Cheers,
Martin

Re: [Darwin2k-users] Physics simulator: collisions?

[Christopher L. <ch...@MI...>]

On Thu, 2001-09-20 at 14:57, Martin C. Martin wrote:
> Could you use it
> to do a robotically realistic version of Karl Simms stuff, where robots
> learn to locomote across an infinate plane?

Ok -- I found it now:

	http://www.genarts.com/karl/evolved-virtual-creatures.html


-- 
Christopher Lee <ch...@mi...>
(617) 253-5095 <http://web.mit.edu/chlee/www>
Postdoctoral Associate of Mechanical Engineering
Field and Space Robotics Laboratory
Massachusetts Institute of Technology
77 Massachusetts Ave, Room 3-472m
Cambridge, MA 02139

Re: [Darwin2k-users] Physics simulator: collisions?

[Christopher L. <ch...@MI...>]

On Thu, 2001-09-20 at 14:57, Martin C. Martin wrote:
> So for my postdoc at MIT I'll be co-evolving bodies and brains, think
> Karl Simms.  I'm looking at what physics engine to use, and noticed that
> Darwin2k rolled it's own.  Does it handle collisiions?  Could you use it
> to do a robotically realistic version of Karl Simms stuff, where robots
> learn to locomote across an infinate plane?

Hey Martin --

Your project sounds cool.  I think Chris Leger's dynamics engine is very
good, and it certainly handles collisions.  I haven't seen Karl Simms
stuff, but I think D2K would do fine for what you are talking about.
I'm building a number of simulations using this code right now.  The
learning curve is somewhat steep, but the documentation is very good.  I
can help you get set-up and show you what I've been working on here.

-Chris

-- 
Christopher Lee <ch...@mi...>
(617) 253-5095 <http://web.mit.edu/chlee/www>
Postdoctoral Associate of Mechanical Engineering
Field and Space Robotics Laboratory
Massachusetts Institute of Technology
77 Massachusetts Ave, Room 3-472m
Cambridge, MA 02139

[Darwin2k-users] Physics simulator: collisions?

[Martin C. M. <ma...@me...>]

Hey all,

So for my postdoc at MIT I'll be co-evolving bodies and brains, think
Karl Simms.  I'm looking at what physics engine to use, and noticed that
Darwin2k rolled it's own.  Does it handle collisiions?  Could you use it
to do a robotically realistic version of Karl Simms stuff, where robots
learn to locomote across an infinate plane?

Enquiring minds want to know,
Martin

[Darwin2k-users] Darwin2k 0.84 released

[Chris L. <cl...@te...>]

After much hacking, hair-pulling, documenting, and a several liters of
coffee, I'm finally releasing Darwin2k 0.84.  This is a pretty major
release, as the dynamic simulation code has been rewritten and
encapsulated into a separate package called dyno (though it's still
distributed w/ Darwin2k).  There's also some documentation (finally!),
generated by doc++ - you can download the darwin2k-docs.tar.gz tarball,
or get it at http://darwin2k.sourceforge.net/darwin2k-docs/index.html. 
I'll be updating the docs at that URL fairly regularly, as they're not
done yet.  This release has some bugs in it (i.e. don't be too surprised
to find a core file lying around), and some of the demos are broken;
however, since there are a number of people just getting started with
Darwin2k, I felt it was important to get the re-written version released
so that they wouldn't have to learn code that would be immediately
outdated.

You can download the tarball for 0.84 at:

  http://download.sourceforge.net/darwin2k/darwin2k-0.84.tar.gz   
(code)
  http://download.sourceforge.net/darwin2k/darwin2k-docs.tar.gz    (html
documentation)

Thanks to Chris Lee for help with automake/autoconf stuff and GUI
restructuring (and spreading the word at MIT...)

Chris

-- 
[ Chris Leger               cl...@ro... ]
[ Phone: (818) 393-4462     Fax: (818) 393-3254          ]

[Darwin2k-users] Please email me if you are on this list

[Chris L. <cl...@te...>]

Please send me mail at cl...@ro... if you're on this
list; I need to know if anyone's subscribed.

Thanks,
Chris

-- 
[ Chris Leger               cl...@ro... ]
[ Phone: (818) 393-4462     Fax: (818) 393-3254          ]