Collision detection patent
Pierre Terdiman
I recently heard rumors about a new patent regarding collision
detection/collision response. As far as I know, this is a patent from John
Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
I'd really like to have more information about that one. You know how rumors
can be. Crazy things are said. For example one reported it was covering most
collision detection methods, which have been used in games for years. Which
is, let me find the appropriate word.... oh yeah: grotesque. And shameful as
well.
I think I'm wrong in the way I understood that patent, am I not?
I'd love to hear what's the exact issue here, what it is supposed to change
for game developers, and so on.
Regards,
--
/*
Pierre Terdiman * Home: p.ter...@wanadoo.fr
Software engineer * Zappy's Lair: www.codercorner.com
*/
John Nagle
> I recently heard rumors about a new patent regarding collision
> detection/collision response. As far as I know, this is a patent from John
> Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
>
> I'd really like to have more information about that one. You know how rumors
> can be. Crazy things are said. For example one reported it was covering most
> collision detection methods, which have been used in games for years. Which
> is, let me find the appropriate word.... oh yeah: grotesque. And shameful as
> well.
Sure. See
http://www.animats.com/topics/patents.html
Read claim 1 of the second patent listed for what's covered.
The general idea is that older "penalty method" systems
break down on the hard cases, but this one doesn't. Most other
demos of physically-based animation have simple blocks or balls
banging around. Even then, they usually don't get motion
that makes things look like they have weight, one of the
hardest things to do in animation. We have human figures falling
downstairs, and they looks right. And it's not vaporware; we
have demos, downloads, videos, and a shrink-wrapped product,
and we've had them for years.
We took a technology that sucked and made it work. That's
a patentable improvement.
I'd be glad to discuss this with industry people who have a
problem with this. I'm also willing to consider licensing
games or academic work that are 100% open source and covered
under the GPL at no charge.
John Nagle
Animats
www.animats.com
Menlo Park, CA
Bill
> Pierre Terdiman wrote:
> > I recently heard rumors about a new patent regarding collision
> > detection/collision response. As far as I know, this is a patent from John
> > Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
> >
> > I'd really like to have more information about that one. You know how rumors
> > can be. Crazy things are said. For example one reported it was covering most
> > collision detection methods, which have been used in games for years. Which
> > is, let me find the appropriate word.... oh yeah: grotesque. And shameful as
> > well.
>
> Sure. See
>
> http://www.animats.com/topics/patents.html
>
> Read claim 1 of the second patent listed for what's covered.
>
> The general idea is that older "penalty method" systems
> break down on the hard cases, but this one doesn't. Most other
> demos of physically-based animation have simple blocks or balls
> banging around. Even then, they usually don't get motion
> that makes things look like they have weight, one of the
> hardest things to do in animation. We have human figures falling
> downstairs, and they looks right. And it's not vaporware; we
> have demos, downloads, videos, and a shrink-wrapped product,
> and we've had them for years.
>
> We took a technology that sucked and made it work. That's
> a patentable improvement.
It is so long as you patent the improvement and not the technology.
So what exactly would someone have to do to be in violation of your
patent? It sounds like all they would have to do is use the penalty method
for collision response and have it look right. If so, I too think this is
bogus and I don't see how it could ever hold up under scrutiny. Also,
how would you expect to ever enforce such a thing? And one final thing,
why, John, why? You must have started hanging around with lawyers.
B.
d.b...@virgin.net
> I'd be glad to discuss this with industry people who have a
> problem with this. I'm also willing to consider licensing
> games or academic work that are 100% open source and covered
> under the GPL at no charge.
Maybe I understand this wrong, is it only patented for people
to use your source code or does this mean if someone else creates
something 'similar' based on same 'kind' of techiques that this breaches
the patent?
No offence I`m sure your product is very good, but I`ve resently been
reading up on physics and trying to do collision response in a game, yes
what I`m doing is 'far' too basic compared to your stuff, but the info I`m
reading has been writen in lots of books and most go over my head atm, so
since physics required for these effects has been studied for a long time
thousands of years by many people and been writen about extensively I don`t
see how someone can patent it now.
Yes if it was something very specific that wasn`t known about but
I read the patent info on your site and it seems very general and would cover
a lot of things, ie "Slip Control" basically covers anything that needs friction
to keep an object upright, no?
cyas
Dave
Thatcher Ulrich
> Pierre Terdiman wrote:
> > I recently heard rumors about a new patent regarding collision
> > detection/collision response. As far as I know, this is a patent from John
> > Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
> >
> > I'd really like to have more information about that one. You know how rumors
> > can be. Crazy things are said. For example one reported it was covering most
> > collision detection methods, which have been used in games for years. Which
> > is, let me find the appropriate word.... oh yeah: grotesque. And shameful as
> > well.
>
> Sure. See
>
> http://www.animats.com/topics/patents.html
>
> Read claim 1 of the second patent listed for what's covered.
>
> The general idea is that older "penalty method" systems
> break down on the hard cases, but this one doesn't. Most other
> demos of physically-based animation have simple blocks or balls
> banging around. Even then, they usually don't get motion
> that makes things look like they have weight, one of the
> hardest things to do in animation. We have human figures falling
> downstairs, and they looks right. And it's not vaporware; we
> have demos, downloads, videos, and a shrink-wrapped product,
> and we've had them for years.
>
> We took a technology that sucked and made it work. That's
> a patentable improvement.
>
> problem with this. I'm also willing to consider licensing
> games or academic work that are 100% open source and covered
> under the GPL at no charge.
I definitely have a problem with it. Not because I think your algorithm
is obvious or the patent is invalid (although I would argue that as
well), but because I'm against software patents in general. I think
software patents pose an insidious threat to the software industry, and
particularly the small independent shop, which I gather Animats is one
of.
Here are a few reasons I'm against software patents:
* Too much non-patent prior art. The history of software innovation is
written in source code, which patent lawyers don't understand, and most
of which is proprietary anyway, so searches for prior art are necessarily
flawed. Even the stuff that's open-sourced is impossible to search in a
comprehensive manner because there's just so damn much of it, and we lack
the technology to automatically search source code for an implementation
of a specific algorithm. "Aha" the lawyers may say, "this is where the
patent system can add value -- by putting software innovation in a big
comprehensive public searchable database." Which is of course absurd if
you consider the amount and complexity of material that would have to go
in, and the unsolved problem of searchability for algorithms.
Furthermore, the existing software patent system is very much *not*
providing this service... I have yet to see a software patent which
includes source code, or even has good-quality exposition like you'd
usually find in a journal article.
* The USPTO is doing a terrible job reviewing software patents. Because
of budget and resources issues, mismanagement, and more importantly
because of the intractability of the prior-art searching problem, the
patent office is issuing a large number of low-quality software patents.
Which is disasterous, in light of the next problem:
* Patents are extremely expensive to file, to get, to enforce, and to
defend against, compared with the amount of coverage they provide. The
price tag in legal fees to file a software patent is say $3K-$6K, plus
potential extra expenses (patents I've been involved with in the past
have required the lawyers to fly to Washington and have lunch with the
examiner, so as to help push the patent through... think about that for a
second). Once it's issued, that's just the tip of the iceberg if you
ever have to fight over it. On the other side, if I'm a developer and
someone decides that I'm infringing their patent, *just receiving the
letter* will set me back $100 in order to run it past a lawyer. Even if
I'm completely innocent, I would have to defend myself, which could
quickly turn into thousands of dollars, not to mention the lost
productive time.
* As a practical matter, software patents favor companies who own large
patent portfolios (i.e. large corporations), and patent-mills who don't
produce actual products. Why? Because in the real world, BigCorp Inc
deals with the expense of patents by building up a large defensive
portfolio which covers some useful stuff, and then when threatened with
patent infringement by some other product-producing company, they do a
cross-licensing agreement. In effect, BigCorp says, "look -- we have
1000 patents in this area of technology, so you either cross-license your
patents with ours, or we'll sue your pants off using our 1000 patents."
It's almost invariably cheaper to cave in, even if you're not infringing
any of those patents. Net effect: employment for a bunch of patent
lawyers, but no license fees paid, and no value created. A small
software company will never be able to collect license fees from BigCorp,
because of this defensive portfolio and the threat of crippling
litigation. This situation actually isn't so bad, in a industry ecology
where you assume that everybody has a patent on *something* vaguely
useful, and that everybody derives most of their revenue from shipping
actual products. Mutually Assured Destruction, or at least heavy losses,
await both combatants in any struggle where both sides can afford some
lawyer-time, so actual cases tend to be few. The only real drawback is
the wasted overhead of getting and maintaining the patents in the first
place. It gets really freaking scary, though, when you consider patent-
mills, which exist only to collect license fees and don't actually ship
product. A defensive, cross licensing strategy is worthless against a
patent-mill, because they can never infringe anything; they don't ship
products.
The one group in all these scenarios which really gets the short end of
the stick is the small software company that *does* ship product. If
SmallShop LLP adopts a strategy of trying to earn revenue through
licensing, they can never collect from the BigCorps of the world, since
BigCorp will just say, "cross-license or we'll go after your product with
our portfolio". The only alternative targets are the other little guys.
It's hard to imagine that strategy paying off, a) because the other
little guys by definition don't have much money, and b) how many of them
could possibly be infringing? Unless SmallShop is tired of trying to
sell actual product and prefers to be a patent mill.
So speaking about your case in particular, who do you expect to collect
fees from?
--
Thatcher Ulrich
http://tulrich.com
Tim Tyler
:> > I recently heard rumors about a new patent regarding collision
:> > detection/collision response. As far as I know, this is a patent from John
:> > Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
:> >
:> > I'd really like to have more information about that one.
:> > [...] Crazy things are said. For example one reported it was
:> > covering most collision detection methods, which have been used in
:> > games for years. Which is, let me find the appropriate word....
:> > oh yeah: grotesque. And shameful as well.
:>
:> Sure. See http://www.animats.com/topics/patents.html
:>
:> Read claim 1 of the second patent listed for what's covered.
:>
:> The general idea is that older "penalty method" systems
:> break down on the hard cases, but this one doesn't. Most other
:> demos of physically-based animation have simple blocks or balls
:> banging around. [...]
: I definitely have a problem with it. Not because I think your algorithm
: is obvious or the patent is invalid (although I would argue that as
: well), but because I'm against software patents in general. [...]
From the claims, it looks *very* much like common-or-garden collision
detection, as described in the textbooks.
I have /plenty/ of collision detection in my programs - for example, see
*all* the applets at http://atoms.org.uk/
The molecular modelling folk appear to have been doing this exact same
thing for donkey's years - see, everything beaneath http://www.msi.com/
for example.
FWIW, I'm also totally opposed to software patents - and indeed patents
in general.
--
__________ http://alife.co.uk/ http://mandala.co.uk/
|im |yler t...@cryogen.com http://hex.org.uk/ http://atoms.org.uk/
Thatcher Ulrich
> Thatcher Ulrich <t...@tulrich.com> wrote:
> : John Nagle <na...@animats.com> wrote:
> :> Pierre Terdiman wrote:
>
> :> > I recently heard rumors about a new patent regarding collision
> :> > detection/collision response. As far as I know, this is a patent from John
> :> > Nagle, related to Animats's Softimage plug-in called "Falling Bodies".
> :> >
> :> > I'd really like to have more information about that one.
> :> > covering most collision detection methods, which have been used in
> :> > games for years. Which is, let me find the appropriate word....
> :> > oh yeah: grotesque. And shameful as well.
> :>
> :> Sure. See http://www.animats.com/topics/patents.html
> :>
> :> Read claim 1 of the second patent listed for what's covered.
> :>
> :> The general idea is that older "penalty method" systems
> :> break down on the hard cases, but this one doesn't. Most other
> :> demos of physically-based animation have simple blocks or balls
>
> : I definitely have a problem with it. Not because I think your algorithm
>
> From the claims, it looks *very* much like common-or-garden collision
> detection, as described in the textbooks.
The "clever twist" which is claimed is using non-linear springs in the
collision response.
> I have /plenty/ of collision detection in my programs - for example, see
> *all* the applets at http://atoms.org.uk/
Fun :)
> The molecular modelling folk appear to have been doing this exact same
> thing for donkey's years - see, everything beaneath http://www.msi.com/
> for example.
I can't tell at first glance if these examples use non-linear springs for
collison response. But for molecular simulation, they'd pretty much
*have* to be, since the physical forces are non-linear.
John Nagle
> The "clever twist" which is claimed is using non-linear springs in the
> collision response.
Yes. With linear springs, either the system needs incredibly
stiff springs that lead to incredibly stiff numerical systems, or
a hard collision results in too much interpenetration. With
nonlinear springs, you still get numerical stiffness, but it's
manageable. Read the patent to see how I did it. (One advantage
of patenting something like this is that I can disclose how I do it.)
The history in this area is that a number of people tried
penalty methods and decided that they sucked. After that, there
was much work in the impulse/constraint direction (Mirtich, Baraff,
etc.) hoping to avoid having to deal with numerical stiffness.
That sort of works if your rigid bodies are very rigid, and there's
not much friction or damping. Everything looks like it's made out
of steel, though. If you have impulse-based collisions, all
collisions happen in one frame time. This looks really stupid
for big objects. If you try to fix that, you run into numerical
stiffness again, and lose the advantage claimed for the
impulse/constraint approach. (Visit the Mathengine
newsgroup at news.mathengine.com, and you can see their
users running into this problem.) So that ran into a dead end.
You can't evade numerical stiffness; you have to deal with it.
Then there's the soft-body world, where everything looks like
it's made out of Jell-O. Remember Hypermatter? Try to make
the material more rigid, and the simulations come unglued.
So that's why previous work didn't solve the problem.
I wanted to do character work with real physics, so I had
to actually solve the problem. The results are on
You can play the videos, and if you have Softimage, can download
a plug-in and try this stuff yourself. Go look at the demos.
Characters falling downstairs, being kicked into a limo by
a giant foot, and other tough situations. They look right,
too. Then go look at the competition. Mostly balls and
blocks banging around. Any questions?
> > The molecular modelling folk appear to have been doing this exact same
> > thing for donkey's years - see, everything beaneath http://www.msi.com/
> > for example.
>
> I can't tell at first glance if these examples use non-linear springs for
> collison response. But for molecular simulation, they'd pretty much
> *have* to be, since the physical forces are non-linear.
Actually, no. See the National Institute of Health's introduction
to molecular modelling. They use Hooke's Law linear springs.
http://cmm.info.nih.gov/modeling/guide_documents/molecular_mechanics_document.html
Besides, molecular modelling doesn't have general collision
detection. All the connections between the atoms are predefined.
And they don't have the huge mass and velocity ranges that cause
trouble in animation and game work. Molecular modelling is
like soft-body simulation in that regard.
This should clear up some of the confusion.
John Nagle
Animats
www.animats.com
Tim Tyler
: Thatcher Ulrich wrote:
:> Tim Tyler wrote:
:> > The molecular modelling folk appear to have been doing this exact same
:> > thing for donkey's years - see, everything beaneath http://www.msi.com/
:> > for example.
:>
:> I can't tell at first glance if these examples use non-linear springs for
:> collison response. But for molecular simulation, they'd pretty much
:> *have* to be, since the physical forces are non-linear.
: Actually, no. See the National Institute of Health's introduction
: to molecular modelling. They use Hooke's Law linear springs.
: http://cmm.info.nih.gov/modeling/guide_documents/molecular_mechanics_document.html
Use of one citation to claim that an entire field does not use non-linear
forces between molcules?
Look at the page you cite, even: "Non-bonded atoms (greater than two bonds
apart) interact through van der Waals attraction, steric repulsion, and
electrostatic attraction/repulsion."
Electrostatic attraction/repulsion to mention just one force decreases
with an inverse square law depending on the distance between the
particles. This is a non-linear relationship.
If "a non-linear force" is your only novely - and it seems to be the only
remotely unusual part of the first claim in the patent - it's been being
done for donkey's years by the molecular modellers. Also it's blindingly
obvious, and completely necessary in order to produce any realistic
molecular modelling.
I'm not saying that you don't have something patentable in your
technology. However the current patent *seems* to encompass *far*
too much existing work to be meaningful - at least in claim 1.
: Besides, molecular modelling doesn't have general collision
: detection.
Of course it does. Even the introduction you cite mentions that
"non-bonded atoms (greater than two bonds apart) interact..."
How do you think you can do any sort of useful molecular simulation
without modelling interactions between particles?
: All the connections between the atoms are predefined.
*Sometimes* they are. Sometimes they are not. In my "crystal
garden" system - or my "molecular membranes" - for example, bonds between
particles form and dissolve dynamically.
That some bonds may be predefined doesn't mean that molecules that aren't
connected don't interact. They can - and do.
: And they don't have the huge mass and velocity ranges that cause
: trouble in animation and game work.
Nor is this mentioned in your patent.
As it stands, claim 1 of your patent appears to cover large volumes of
existing work by the molecular modeling folk - not to mention other
groups.
I would have thought that any half-decent laywer could use this material
to demolish any patent infringment claims that are made relating to that
claim.
--
__________ Lotus Artificial Life http://alife.co.uk/ t...@cryogen.com
|im |yler The Mandala Centre http://mandala.co.uk/ Namaste.
Tim Tyler
:> :> Pierre Terdiman wrote:
:> :> > I recently heard rumors about a new patent regarding collision
:> :> > detection/collision response. [...]
:> :> >
:> :> > I'd really like to have more information about that one.
:> :>
:> :> > [...] Crazy things are said. For example one reported it was
:> :> > covering most collision detection methods, which have been used in
:> :> > games for years. Which is, let me find the appropriate word....
:> :> > oh yeah: grotesque. And shameful as well.
:> :>
:> :> Sure. See http://www.animats.com/topics/patents.html [...]
:>
:> : I definitely have a problem with it. Not because I think your algorithm
:> : is obvious or the patent is invalid (although I would argue that as
:> : well), but because I'm against software patents in general. [...]
:>
:> From the claims, it looks *very* much like common-or-garden collision
:> detection, as described in the textbooks.
: The "clever twist" which is claimed is using non-linear springs in the
: collision response.
From claim 1:
``a nonlinear relationship with respect to said closest-points vector such
that said force increases sufficiently rapidly as said closest-points
vector goes to zero to overcome motions causing said collisions between
bodies;''
Alas, the inverse square law fits this description absolutely precisely.
The inverse square law goverens a huge number of inter-molecular forces -
and there are no shortage of people who have modelled them.
:> I have /plenty/ of collision detection in my programs - for example, see
:> *all* the applets at http://atoms.org.uk/
: Fun :)
;-)
:> The molecular modelling folk appear to have been doing this exact same
:> thing for donkey's years - see, everything beaneath http://www.msi.com/
:> for example.
: I can't tell at first glance if these examples use non-linear springs for
: collison response. But for molecular simulation, they'd pretty much
: *have* to be, since the physical forces are non-linear.
These models appear very sophisticated. They model adhesion, surface
tension, diffusion through polymers, crystallisation, gelation and
abot a million other things. The idea that they've managed this witout
bothering with minor things like electrostatic forces appears ridiculous.
For some pretty picures of MSI's work, see:
http://www.msi.com/materials/studio/images/
MSI are not the only folk in this field. For other molecular modelling
companies, see the huge list at:
http://cmm.info.nih.gov/modeling/universal_software.html
Dave Eberly
news:39A6C1A9...@animats.com...
> Yes. With linear springs, either the system needs incredibly
> stiff springs that lead to incredibly stiff numerical systems, or
> a hard collision results in too much interpenetration. With
> nonlinear springs, you still get numerical stiffness, but it's
> manageable. Read the patent to see how I did it.
Fortunately for you, Bill Gear and others did not seek a
patent on numerical analysis/methods for handling stiff
systems of differential equations. That is how technology
advances--you build on the ideas of people before you.
"Students, pay attention now. Before I start the lecture,
please sign the NDA being circulated about the room.
And by the way, what I am about to teach you may not
be used by you in any way, otherwise I'll have to sue you.
Your class project is due at the end of the semester.
Grade of B if it is patentable, grade of A if you finish it
early and get it patented."
Sound extreme? It does to me, but that is the direction
you and others appear to be taking with "mathematics
patents" and "algorithm patents". Sadly enough, the
Universities themselves are jumping on this bandwagon,
too. I served on a PhD dissertation committee a couple
of years ago, topic was about tracking in an augmented
reality system. I thought it was a reasonable enough
result for the student to earn the PhD, but it was just a
start and there were a number of technical issues that
needed to be explored and understood. I was very
surprised to see a few months later that a patent had
been applied for and granted. Even more surprised to see
that one of the folks on the patent was a very recent
Turing award winner.
Where will it end? Hard to say, but I do claim that
William Gibson is not a sci-fi/cyberpunk-fantasy writer,
he is a prophet.
> (One advantage
> of patenting something like this is that I can disclose how I do it.)
I believe that Jeff Lander and I stated an opinion on this some
time ago. If you have a great idea that you want to make a lot
of money with, do not tell anyone about it. Create your product,
license the binaries, and make your money. If someone just
as smart as you are comes along and develops the same idea,
that's life. And the probability that this will happen is quite
large, because... (redundant statement to occur...) That is how
technology advances--you build on the ideas of people before
you.
Now go find your old teachers and thank them for the
blessing they have given you.
--
Dave Eberly
ebe...@magic-software.com
http://www.magic-software.com
John Nagle
> :> Tim Tyler wrote:
> If "a non-linear force" is your only novely - and it seems to be the only
> remotely unusual part of the first claim in the patent - it's been being
> done for donkey's years by the molecular modellers. Also it's blindingly
> obvious, and completely necessary in order to produce any realistic
> molecular modelling.
All the claims have to be read together against prior art.
Claim interpretation is a strict "AND". I recommend reading
"Patent It Yourself", by David Pressman, for an introduction
to US claim interpretation. I'm not going to argue claim
interpretation here; that's a specialist topic and we pay
lawyers to do that.
We will generally offer no-cost licenses for 100% GPL'd
free software projects. So this shouldn't interfere with
non-commercial or academic work. See the Animats web site
on how to request a license. I'd personally like to see a good
GPL'd rigid-body physics library; there are some decent collision
libraries now, but the available physics libraries mostly suck.
Most implementors knock off the easy cases, hit the stiffness wall,
and give up.
If someone really has a problem with this, Animats is
represented by Townsend and Townsend and Crew in Palo Alto,
California, Silicon Valley's largest intellectual property
law firm. Please have your attorney contact Gary Aka or
Bill Bohler there. Thanks.
John Nagle
Animats
www.animats.com
John Nagle
> I believe that Jeff Lander and I stated an opinion on this some
> time ago. If you have a great idea that you want to make a lot
> of money with, do not tell anyone about it. Create your product,
> license the binaries, and make your money.
Now that's backwards. Public policy, all the way back to
the Constitution, encourages patents and discourages trade secrets.
A patent is a monopoly for a limited time. You have to disclose
how it's done, and in practice most patents can be licensed
by competitors. Trade secrets are potentially a monopoly forever.
Nothing gets disclosed, and trade secrets are hard to license
because of the risk of leaks. Patents are thus much more open.
The idea is to get away from "secrets hidden in the binaries".
We've had too much of that. That's the major antitrust complaint
with Microsoft.
Followups to "misc.int-property", where this discussion
has been done before.
John Nagle
Animats
Thatcher Ulrich
> [snip]
> I wanted to do character work with real physics, so I had
> to actually solve the problem. The results are on
>
> http://www.animats.com
>
> You can play the videos, and if you have Softimage, can download
> a plug-in and try this stuff yourself. Go look at the demos.
> Characters falling downstairs, being kicked into a limo by
> a giant foot, and other tough situations. They look right,
> too. Then go look at the competition. Mostly balls and
> blocks banging around.
Looks good, and the technical explanations make sense.
> Any questions?
Yes. Question 1: if Falling Bodies is a superior product, why do you
need patent protection?
Question 2: who do you expect to collect fees from; is it from developers
of products that compete directly with Animats (e.g. modeling package
plug-ins), or from developers who don't compete directly (e.g. game
developers)?
Thatcher Ulrich
> Dave Eberly wrote:
> > I believe that Jeff Lander and I stated an opinion on this some
> > time ago. If you have a great idea that you want to make a lot
> > of money with, do not tell anyone about it. Create your product,
> > license the binaries, and make your money.
>
> Now that's backwards. Public policy, all the way back to
> the Constitution, encourages patents and discourages trade secrets.
> A patent is a monopoly for a limited time. You have to disclose
> how it's done, and in practice most patents can be licensed
> by competitors. Trade secrets are potentially a monopoly forever.
> Nothing gets disclosed, and trade secrets are hard to license
> because of the risk of leaks. Patents are thus much more open.
The public policy stretching back to the US Constitution was to exclude
mathematical formulas and algorithms from patent protection. Mathematics
seems to have survived the centuries without patents just fine.
> The idea is to get away from "secrets hidden in the binaries".
I've never seen a software patent that contained source code. Are you
planning on opening the Falling Bodies source as a result of your
patents issuing?
> We've had too much of that. That's the major antitrust complaint
> with Microsoft.
Um, no it isn't. The major antitrust complaint againt Microsoft is that
they leveraged their market position to extend their monopoly. It has
nothing to do with IP. And MS has plenty of patents, but that doesn't
seem to have encouraged them to be more open.
Tim Tyler
: Tim Tyler wrote:
[collision detection patent objections]
: All the claims have to be read together against prior art.
: Claim interpretation is a strict "AND". [...]
Claims 2 and 3 appear to be mutually contradictory - as do claims 7 and 8
... and so on. They can be ORred together without becomming meaningless -
but the "AND" of the claims in the patent appears likely to be
pretty-much the null set.
*If* this interpretation is correct, I don't have any problem with
the patent - except my usual distaste in seeing tax-payers' money
spent on the evil patent office, of course ;-)
Billy Bissette
> All the claims have to be read together against prior art.
> Claim interpretation is a strict "AND". I recommend reading
> "Patent It Yourself", by David Pressman, for an introduction
> to US claim interpretation. I'm not going to argue claim
> interpretation here; that's a specialist topic and we pay
> lawyers to do that.
There are US patents which have been issued despite ample prior
art. The patent system just isn't designed to properly handle
certain types of applications, particularly in areas of quick
improvements.
It is even more true when those involved in the patent process
beyond the applicants themselves are the only ones versed in
the topic area. And the only way to overturn a patent, no
matter how unfounded its issue, is through costly legal battles.
> The idea is to get away from "secrets hidden in the binaries".
> We've had too much of that. That's the major antitrust complaint
> with Microsoft.
Someone else has already corrected you on the antitrust issue.
But I would also like to ask whether you will open source all of
your patented works? The patent itself will most likely make no
code publically available, so it will be up to you to get away from
"secrets hidden in the binaries".
Of course, you could just GPL the code. And if you have such
faith in the patent system, surely your prior art claim would
prevent someone else from profitting from your work with their
own patent application.
In the end, you are only hurting technological advancement. If
you hadn't done it, and patented it, someone else probably would
have within the next couple of years. That is where the patent
system breaks down, it simply cannot deal with the rates of
advancement in certain fields. All you are doing is encouraging
others to restrict their own works. Perhaps the 'methods' will
be public, but what will it matter when people aren't allowed to
use them?
Dave Eberly
> Now that's backwards. Public policy, all the way back to
> the Constitution, encourages patents and discourages trade secrets.
> A patent is a monopoly for a limited time. You have to disclose
> how it's done, and in practice most patents can be licensed
> by competitors. Trade secrets are potentially a monopoly forever.
> Nothing gets disclosed, and trade secrets are hard to license
> because of the risk of leaks. Patents are thus much more open.
Let me make my point clearer. I want neither a short-term nor
long-term monopoly on "mathematical/software algorithms".
I want *no* constraints on them. If everyone who develops an
algorithm of this type were to patent it, technology is stifled. Do
you think folks really want to pay to use something mathematically
obvious? I certainly would not want to pay Animats just because you
happened to learn what a stiff system is and how to deal with
it numerically. Numerical analysts having been dealing with these
for a long time. I had to deal with them 20 years ago in modeling
combustion.
I have not had to time to study physically-based animations.
When I decide to do so, I guarantee you that with my mathematics
and computer science background, I will discover the same
things you did *without reading a thing you have ever done*.
Why is that? Because I will have studied the same prior
knowledge and used the same analytical reasoning skills and
deductive logic to get to the same conclusion. I would find it
quite distasteful to have corporations and intellectual property
lawyers threatening to sue me just because I decided to think.
Take pride if you believe that you are the first to solve such a
difficult problem. But enough of this Highlander attitude of
"there can only be one."
> Followups to "misc.int-property", where this discussion
>has been done before.
I hate it when someone rudely changes the 'followups'. Stay here
and enjoy the heat for awhile.
Dave Eberly
> I'm going to drop out of this discussion, since it's turning
> into the Generic Anti-Patent Rant. Thanks.
Just what you did a long time ago in a similar thread.
At the same time the Generic Anti-Patent Rant ends,
so will the Generic Patent Rant. Thanks for your time
and insight into the issues.
John Nagle
> John Nagle wrote:
> > We've had too much of that. That's the major antitrust complaint
> > with Microsoft.
>
> they leveraged their market position to extend their monopoly. It has
> nothing to do with IP. And MS has plenty of patents, but that doesn't
> seem to have encouraged them to be more open.
The "secret APIs" are a big issue. Read Judge Jackson's
findings of fact. I don't really have time to get into that mess
here.
John Nagle
Animats
John Nagle
> I have no knowledge of the niceties of patent law, nor of how this
> point bears on the particular patent under discussion. But it seems
> to me that this point is closely related to a deep philosophical
> question which has been debated by mathematicians and philosophers of
> science for centuries, and without definitive resolution AFAIK.
I'm going to drop out of this discussion, since it's turning
into the Generic Anti-Patent Rant. Thanks.
John Nagle
Animats
www.animats.com
Gerry Quinn
>I have not had to time to study physically-based animations.
>When I decide to do so, I guarantee you that with my mathematics
>and computer science background, I will discover the same
>things you did *without reading a thing you have ever done*.
>Why is that? Because I will have studied the same prior
>knowledge and used the same analytical reasoning skills and
>deductive logic to get to the same conclusion. I would find it
>quite distasteful to have corporations and intellectual property
>lawyers threatening to sue me just because I decided to think.
>Take pride if you believe that you are the first to solve such a
>difficult problem. But enough of this Highlander attitude of
>"there can only be one."
If you can demonstrate the above, you should have nothing to fear. The
'novelty' requirement of patent law means that a 'person well versed in
the art' would not be expected to discover the published invention.
- Gerry Quinn
Thatcher Ulrich
In any case, it was a red herring argument in the first place. API
secrets have no relevance to algorithm patents, that I can see.
Thatcher Ulrich
>
> [collision detection patent objections]
>
> : All the claims have to be read together against prior art.
> : Claim interpretation is a strict "AND". [...]
>
> Claims 2 and 3 appear to be mutually contradictory - as do claims 7 and 8
> ... and so on. They can be ORred together without becomming meaningless -
> but the "AND" of the claims in the patent appears likely to be
> pretty-much the null set.
>
> *If* this interpretation is correct, I don't have any problem with
> the patent - except my usual distaste in seeing tax-payers' money
> spent on the evil patent office, of course ;-)
Hopefully this clarifies things:
>>>
4. Validity is assessed separately for every claim in a patent. This
creates an incentive for applicants to submit multiple claims.
5. Infringement of one valid claim constitutes infringement of the
patent.
<<<
(from "A Brief Overview of Patent Law for the University Research
Administrator" by Kenneth D. Sibley at
http://www.carolinapatents.com/pat_articles/pat_article14.htm )
This jibes with what I've been told by patent lawyers.
Disclaimer: I am not a lawyer, nor do I want to be.
Tim Tyler
:> [collision detection patent objections]
:>
:> : Claim interpretation is a strict "AND". [...]
[snip problems with this]
: Hopefully this clarifies things:
[quote start]
: 4. Validity is assessed separately for every claim in a patent. This
: creates an incentive for applicants to submit multiple claims.
: 5. Infringement of one valid claim constitutes infringement of the
: patent.
[quote end]
That's what I figured originally. I'm convinced claim 1 is worthless.
However some of the other numerous claims may succeed in being so specific
that they are not easy to dismiss on grounds of originality alone -
and they're also correspondingly less likely to be infringed.
Tim Tyler
: One advantage of patenting something like this is that I can disclose
: how I do it.
*Hardly* an advantage of patenting - since you can disclose how you do it
anyway - with or without a patent.
To be clear, the advantages of patenting primarily include the ability to
extract money from others who use related methods, by either suing them,
or licensing the technology under coercion provided by the threat of
lawsuits.
Patents can also contribute to a display to ward off potential
competitors - by hinting at the possibility of legal proceedings -
and to hinder them by forcing them to use second-rate techniques.
Patrice Ponce
"Tim Tyler" <t...@cryogen.com> wrote in message news:G013p...@bath.ac.uk...
> John Nagle <na...@animats.com> wrote:
>
> : One advantage of patenting something like this is that I can disclose
> : how I do it.
>
> *Hardly* an advantage of patenting - since you can disclose how you do it
> anyway - with or without a patent.
>
> To be clear, the advantages of patenting primarily include the ability to
> extract money from others who use related methods, by either suing them,
> or licensing the technology under coercion provided by the threat of
> lawsuits.
>
> Patents can also contribute to a display to ward off potential
> competitors - by hinting at the possibility of legal proceedings -
> and to hinder them by forcing them to use second-rate techniques.
May I suggest that patents are not bad things for themselves. It is even
kind
of written in the declaration of the rights of man: one's inventions should
be protected... I think thats also a part of business : small companies must
be able to protect their industrial property, otherwise it is quite
difficult to
face the bigger ones.
The real problem is in what can be patented : too many patents cover really
obvious "inventions"; there should be a clear process to be sure that's not
common sense for someone versed in the art.
I did not read carefuly this collision detection patent, but when we look at
all physics engines, it appears to be really one of the worst problems...
Antonio Martini
> collision response.
Inthe CGDC'98(Computer Games Developers' Conference) at pag.693 I read:
"McKenna has suggested the following joint limits:
Fj = -bV + L
where L = a(exp( b(qmin - q) ) -1) if q < qmin
-a(exp( b(q-qmax) ) -1) if q > qmax
"
as we can see an exponential function used for the collision response.
The reference cited in the paper is:
McKenna, M. and , Zelter D., "Dynamic Simulation of a Complex Human Figure
Model
with Low Level Behavior Control",Presence,Vol 5. No. 4, Fall 1996. 431-456.
just an observation
Antonio.
Thatcher Ulrich
That looks like pretty solid prior art. I went looking for the reference
in the ACM Digital Library, and didn't find it... I guess _Presence_
hasn't made it into the digital archives. However, I did find another
paper by McKenna and Zeltzer (note spelling):
Michael McKenna and David Zeltzer, "Dynamic simulation of autonomous
legged locomotion", International Conference on Computer Graphics and
Interactive Techniques, Conference proceedings on Computer graphics,
August 6 - 10, 1990, Dallas, TX USA, Pages 29-38.
<quote>
Collision and contact in corpus is handled through spring forc-
es. When a collision is detected between two bodies, forces are
applied to each body as a function of penetration depth in the
direction of the collision normal. Linear or exponential spring
functions can be used to create the forces; exponential springs
are typically employed because deep penetration is strongly re-
sisted by the exponential rise in force.
</quote>
IANAL, but that alone looks like the death-blow to the broadest claims in
John's patent.
Jeff Lander
One difference with the McKenna paper, which I have, is that they use the
exponential springs largely for muscle based motors to drive the dynamic
animation. I do not believe Nagle's work has any dynamic controller to
create biologic motion. At least from what I have seen, Falling Bodies just
simulates the object colliding in an enviroment If John had a robust
controller that created realistic motion for a dynamic biped character over
a variety of terrains, he wouldn't need to patent it, many would gladly buy
that solution.
So it is exponential springs used for penalty method collision detection.
It is not clear exactly how the collision response in this paper is
addressed but the more important quote as it applies to this patent is:
"During stance, the feet would stick to the ground using exponential
springs."
It really says no more that can I see on the contact model but there are
tons of refs in the paper to chase down as well as talking to McKenna about
it. I will pursue that a bit.
In Nagle's summary of the invention you will see the formula used for
calculation of the force
F=ae^b(c-x) -d where a,b,c,d are constants and x is the distance
In McKenna's paper he gives the formula f = a(e^(b|x|)-1) suspiciously
close but perhaps legally different enough. I would avoid that formulation
for the force by slightly modifying it or using McKenna's. I am surprised
the McKenna paper is not mentioned as a prior art source since it is a very
well known paper and is very close in formulation and discussion, but I am
no lawyer.
I believe there is nothing in this patent to worry about. He will likely
try to bully MathEngine and Havok into licensing some tech, won't go after
any game developers despite the posturing (not any money in it), and maybe
eventually make a real-time physics library to license. For developers,
stick to the documented research, support public research projects, publish
your own work, and refuse to buy products and use tech from people who try
and patent mathematical algorithms.
-Jeff
"Thatcher Ulrich" <t...@tulrich.com> wrote in message
news:MPG.1415db016...@news.std.com...
Jeff Lander
http://www.media.mit.edu/~dsmall/mikey/index.html#Publications
If you want to read it yourself.
"Jeff Lander" <jeffl@darwin'three'd.com> wrote in message
news:vi1r5.7267$p5.2...@newsread03.prod.itd.earthlink.net...
Thatcher Ulrich
> BTW: Paper is available from Michael's site at:
> http://www.media.mit.edu/~dsmall/mikey/index.html#Publications
>
> If you want to read it yourself.
>
> One difference with the McKenna paper, which I have, is that they use the
> exponential springs largely for muscle based motors to drive the dynamic
>
> So it is exponential springs used for penalty method collision detection.
> It is not clear exactly how the collision response in this paper is
> addressed but the more important quote as it applies to this patent is:
>
> "During stance, the feet would stick to the ground using exponential
> springs."
> It really says no more that can I see on the contact model but there are
> tons of refs in the paper to chase down as well as talking to McKenna about
> it. I will pursue that a bit.
Thanks for the link. Are we talking about the same paper? I was looking at
"Dynamic Simulation of Legged Autonomous Locomotion", the SIGGRAPH '90 paper
that's on McKenna's site, and it's also in the ACM Digital Library which is
where I found it. Most of the paper seems to be about locomotion control,
but the specific quote in my last post (from section 3.1 of the paper, second
to last paragraph in the section) is talking explicitly about collision and
contact response in the same sense as the Nagle patent. At least that's how
I read it. It seems to me that claim 1 of the Nagle patent is almost a direct
paraphrase of the beginning of that paragraph.
As I read more of the paper, I even notice that McKenna writes a couple
sentences on the numerical problems of stiff springs, and the use of
variable-length time steps (Section 3.3, last paragraph).
I find it appalling to see minor concepts clearly outlined in a SIGGRAPH
'90 paper show up as primary claims in a 1998 patent. What a sorry waste
of everybody's time.
Jeff Lander
me on. An algorithm is not patentable. Only a particular unique process.
The patent must describe an inventive effect on how a machine operates.
Assuming, and I haven't had a lawyer look at this particular patent, that
the crux is the formulation of the exponential spring force and the process
of how it is appied to the object. What is not clear to me and would require
a lawyer is how different would your process need to be in order to not
infringe.
I am not sure just changing the force formula would do it. If it were
applied using the same method, they could certainly try and claim you were
infringing. Maybe not win but probably have a case. A different process as
well as formula would most certainly not infringe. Easy to do in a
simulation such as this.
-Jeff
"Ron Levine" <r...@dorianresearch.com> wrote in message
news:39b10cde...@news2.psn.net...
Jeff Lander wrote
>In Nagle's summary of the invention you will see the formula used for
>calculation of the force
>F=ae^b(c-x) -d where a,b,c,d are constants and x is the distance
>
>In McKenna's paper he gives the formula f = a(e^(b|x|)-1) suspiciously
>close but perhaps legally different enough. I would avoid that formulation
>for the force by slightly modifying it or using McKenna's.
What's "legally different enough"?
Why not use some polynomial approximation to the exponential, such as
a truncated Taylor series? That is effectively what the library exp
routine does anyway. Would that be legally different enough?
Computation is only approximate, so some numerically small
perturbation of the above formula may be mathematically distinct yet
yield effectively the same results, perhaps even better. Would that
be legally different enough?
Jeff Lander
collision response other then the ground sticking bit. Nothing about object
to object collision handling. It clearly covers use of non-linear springs
for DOF restrictions and joint limits. That was one of the patents claims.
McKenna's thesis goes into more detail about the collision response and it
is clear to me from looking at it that it is a prior art source for
non-linear and exponential springs for use in collision response using the
penalty method as well as for joint limits.
Other sound prior art for penalty method exists from Moore and Wilhelms at
Siggraph 88 "Collision Detection and Response for Computer Animation". They
use linear springs but claim in the paper that it is a widely used method
for collision response. Searching their refs should get you back earlier.
In my opinion (obviously not my lawyers :) I agree with you and believe that
established clear prior art exists for the broadest of the claims (namely
claim 1, 6, and 19). That is no big deal for the patent since I am certain
they were probably designed to be very broad. It is the more detailed
claims that are trickier. I think the compressible/incompressible stuff
would have good prior art though would need to follow up. If you use the
force calculation in claims 2 or 3 and the rest of the process you may be in
trouble. Etc. If you are worried and designing a system you would probably
need to avoid the most specific claims. Some probably are unique and have
legal merit. If John wants to debate why the broad claims should be covered
by prior art, I would love to. I certainly don't automatically discount the
whole thing.
But as I said before it is most likely a non-issue designed to build up IP
in the company, possibly some licenses, and get some press (even if it is
bad... which we are feeding). Animats doesn't have a product to sell for
real-time physics and I doubt this issue makes a lick of difference to
people who would want to buy the Softimage plugin.
I find companies like Unisys who buy patents then use a ton of legal muscle
to bully people much more scary. Not that I don't think we should speak out
against any software patent. Obviously I think we should :).
-Jeff
"Thatcher Ulrich" <t...@tulrich.com> wrote in message
news:MPG.1416f4adb...@news.std.com...
> Jeff Lander <jeffl@darwin'three'd.com> wrote:
> > BTW: Paper is available from Michael's site at:
> > http://www.media.mit.edu/~dsmall/mikey/index.html#Publications
> >
> > If you want to read it yourself.
> >
> > [prior post]
> > One difference with the McKenna paper, which I have, is that they use
the
> > exponential springs largely for muscle based motors to drive the dynamic
> > animation. [...]
> >
> > So it is exponential springs used for penalty method collision
detection.
> > It is not clear exactly how the collision response in this paper is
> > addressed but the more important quote as it applies to this patent is:
> >
> > "During stance, the feet would stick to the ground using exponential
> > springs."
> > It really says no more that can I see on the contact model but there are
> > tons of refs in the paper to chase down as well as talking to McKenna
about
> > it. I will pursue that a bit.
>
> Thanks for the link. Are we talking about the same paper? I was looking
at
> "Dynamic Simulation of Legged Autonomous Locomotion", the SIGGRAPH '90
paper
> that's on McKenna's site, and it's also in the ACM Digital Library which
is
> where I found it. Most of the paper seems to be about locomotion control,
> but the specific quote in my last post (from section 3.1 of the paper,
second
> to last paragraph in the section) is talking explicitly about collision
and
> contact response in the same sense as the Nagle patent. At least that's
how
> I read it. It seems to me that claim 1 of the Nagle patent is almost a
direct
> paraphrase of the beginning of that paragraph.
>
> As I read more of the paper, I even notice that McKenna writes a couple
> sentences on the numerical problems of stiff springs, and the use of
> variable-length time steps (Section 3.3, last paragraph).
>
> I find it appalling to see minor concepts clearly outlined in a SIGGRAPH
> '90 paper show up as primary claims in a 1998 patent. What a sorry waste
> of everybody's time.
>
BuschnicK
using a patented algorithm, how can someone say if they don't have the
source code? Judging, hmm, that looks similar to what my patented algorithm
does, so you are legally forced to show me the source? Are there any
examples of this happening? I know about the mp3 and gif compression stuff,
but that's more specific, i.e. if I can read a certain file format it's
obvious I must be using the patented decoder stuff...
BuschnicK
--
RealName: Sören Meyer-Eppler | ICQ: 45724082
email: Busc...@SPAM.ngi.de
Life would be much easier if I had the source code.