Home | News | Hacking | Sciences | Technology | Ti 92 | Programming | Free articles | Links | Webmaster

SIMULATION OF PLANTS

 

1. Introduction


One of the most demanding tasks when you try to simulate a natural environment (for example for games or rendering aplications) is the simulation of plants. How can you create a realistic plant that interacts with its environment as a real plant would interact? There have been a couple of different approaches but only one of them has been really sucessfull in simulating a broad variety of different plants - the so called L-systems.

2. The Idea Behind L-Systems


The L-systems - or Lindenmayer systems - have been developed by the biologist Aristid Lindenmayer. His idea was to simulate plants by a model that corresponds as accurately as reasonably possible to the real growth of plants. If we take a look at the growth of real plants, they start with some sort of seed (in the L-system terminology this is called an axiom). The cells of this seed reproduce themselves according to some rules which are determined by the DNA (=desoxyribonucleic acid) and finally create the entire plant. Of course this is a pretty oversimplified view of what is really happening, but for creating an usefull model this is already enough accurate.
Let's start with one of the most prominent examples, that is used in all books about this topic to show the power of L-systems:

The beauty of this L-system is, that it is extremely simple. There are only two different elements (in this example represented by the blue and the red line) : the red line represents the younger parts of the plant, while the blue line represents the older part of the plant. It is therefore quite reasonable to start with the younger part (i.e. use the red line as axiom). The first reproduction rule (the one for the red line) states, that with each iteration the young part grows by a factor of two and becomes an old part (the red line is replaced by two blue lines). Additionally three new young branches (the three red lines) grow from this old part. The second reproduction rule tells, what happens to old parts: they simply grow, but no new branches are formed (one blue line is replaced by two blue lines). This behaviour is also what you would expect from a real plant: it is highly unlikely that after some time a new leave emerges from the stem of an old tree - the probability that the leaves starts at some new branch is much higher.
OK, after we now basically understand how this L-system works, lets take a look at the result:

 

In the image above I have drawn the red and blue lines as green, so that it reminds a bit more to a plant. The starting point (n=0) is just the axiom (i.e. a straight line corresponding to the red line of the axiom which represents the new part of the grass). In the first iteration (n=1) the replacement rule for the red line is applied (there is no blue line, that could be replaced because the axiom was only one single red line) and we get an old stem from which 3 young branches emerge. In the second iteration (n=2) the two blue lines of the stem are replaced by 4 blue lines (i.e. the old stem grew) and each branch is replaced by a stem from which 3 new branches emerge. If you do this for another 6 iterations you end up with something that looks pretty similar to some sort of grass.
As you can see we can already get very nice results with quite simple models. The only three things we have asumed about this grass were, that

  1. it starts with a young part (axiom: red line)
  2. each young part becomes old but while it becomes old it produces 3 young branches
  3. old parts can still grow, but don't produce any branches

However if you really want to generate these plants on your computer, you have to use a little bit more formal aproach (although the colorful lines look nice, the computer is usually not to happy with them ;-) Lindenmayer suggested to represent each different element of the plant by a different letter (for example the red line would be represented by F and the blue one by G). To represent the actual branching structure you need additionally some geometrical representation. Lindenmayer sugested to use turtle graphics for this representation. Originally turtle graphics has been used in an old programming language (called LOGO), that probably most of us haven't seen any more. Nevertheless the idea behind is very intuitive and most people will recognize it immediately because many games use it to control the movements of the actors or some rendering programs use it also for their realtime preview.
The idea is, that you sit on the turtle and tell her to go straight ahead (default mode, each capital letter means that a straight line of unit length has to be drawn), turn left (in the L-system represented by +) or turn right (in the L-system represented by -). When you turn left or right, you just change the orientation of the turtle but you don't draw a line in that direction. For example to draw a line to the left, you have to use the sequence +F instead of just +. If you want to have branches, you need an additional symbol (in L-systems represented by [ ) to put the current position of the turtle into the stack and a symbol (in L-systems represented by ] ) to get the position of the turtle back from the stack after the branch has been drawn.
With these conventions the L-system above would look like this:

derivation length: 8
Axiom: F
F --> G[+F][-F]GF
G --> GG

Beside the actual L-system there are a few additional informations needed on how the L-system has to be interpreted. In the L-system above we have to specify the angle by which the orientation is changed with + or -. As you can see this affects the final result quite significantly:

 

3. Symbols used in L-Systems

The symbols mentioned above are already a good starting point, but if you want to create more complex models (3D models, flowers,...) you need a little bit more symbols. In the following list you can find the most important symbols for L-systems with a short description of each (see also the paper "Visual models of plant growth" from Prusinkiewicz et al.):

Symbols that cause the turtle to move and draw:

F(s), G(s) Move forward a step of length s and draw a line segment from the original to the new position of the turtle.
f(s), g(s) Move forward a step of length s without drawing a line.
@O(r) Draw a sphere of radius r at the current position.

 

Symbols that control turtle orientation in space:

+(a) Turn left by angle a around the z axis. The rotation matrix is Rz(a).
-(a) Turn right by angle -a around the z axis. The rotation matrix is Rz(-a).
&(a) Pitch down by angle a around the y axis. The rotation matrix is Ry(a).
/\(a) Pitch up by angle -a around the y axis. The rotation matrix is Ry(-a).
/(a) Roll left by angle a around the x axis. The rotation matrix is Rx(a).
\(a) Roll right by angle -a around the x axis. The rotation matrix is Rx(-a).
| Turns 180 around the z axis. This is equivalent to +(180) or -(180).

 

Symbols for modeling structures with branches:

[ Push the current state of the turtle (position, orientation and drawing attributes) onto a pushdown stack.
] Pop a state from the stack and make it the current state of the turtle. No line is drawn, although in general the position and orientation of the turtle are changed.

 

Symbols for creating and incorporating surfaces:

{ Start saving the subsequent positions of the turtle as the vertices of a polygon to be filled.
} Fill the saved polygon.
~X(s) Draw the surface identified by symbol X, scaled by s, at the turtle's current location and orientation. Such a surface is usually defined as a bicubic patch.

 

Symbols that change the drawing attributes:

#(w) Set line width to w, or increase the value of the current line width by the default width increment if no parameter is given.
!(w) Set line width to w, or decrease the value of the current line width by the default width decrement if no parameter is given.
;(n) Set the index of the color map to n, or increase the value of the current index by the default color increment if no parameter is given.
,(n) Set the index of the color map to n, or decrease the value of the current index by the default color decrement if no parameter is given.

The rotation matrices mentioned in the above definitions are defined as follows:

The definitions above represent the symbol definitions according to Lindenmayer (and extended by Prusinkiewicz). It is a good idea to use these definitions for your own implementations too, because it is much easier this way to exchange the L-systems with fellow programmers/scientists.

 

4. Let's start growing ...

Well now you know everything you need to let the plants grow on your desktop. When you want to actualy grow plants now, you have to either implement the string replacement and graphic output routines or you can use one of the available programs. The best programm you can use is LinSys3D developed by Andrea Esuli.
All three programs are very similar and pretty powerfull. I will now briefly describe how you can create L-systems in one of these programs. Typically the L-system would look like this:

 

derivation length: 8     //the number of derivation steps 
Axiom: F                 //the axiom 
F --> G[+F][-F]GF        //one (or more) productions
G --> GG
    

Additionally to the actual L-system you need some parameters for the graphical interpretation. For the L-system above, these are:

angle increment: 15              //the turning angle, in degrees
initial line width: 1 pixels     // self-explanatory
initial color: 2                 //index to the color map (2=green)
scale factor: 0.9000             //controls the size on the screen

The commands above will produce the simple gras we discussed earlier. By modifying the code above you can easily represent any L-system you want.

 

5. Parametric L-Systems

 

As we have seen L-systems are a very powerfull tool to model plants. Nevertheless I have to admit that finding a proper model for a certain plant is usually quite time consuming (at least for non-biologists like me ;-) When you want to model a new plant you should have a couple of photos (or even better: the plants itself) in different growth phases. Then you have to compare from which part which other parts are developed and then you can asign to each element a letter and you must try to find the rules that govern the development. Additionally you should try to keep the axiom as simple as possible (don't put all your knowledge into the axiom - the axiom should not be the final plant but the seed for the plant! ).
Let's try to create some sort of fern as an example:

If you look at the final plant it is usually much to complex to get a good idea on how to design the L-system. Therefore you should try to get a few images of the plant as it looks in an early state. There is also another principle, that might help you sometimes: selfsimilarity. Put simply, this means that if you take a small part of the plant, the structure will be similar to the structure of the entire plant (this principle has been discussed very extensively in the context of fractals).
In most plants you don't find this phenomenon, but in some plants - like ferns - this principle is obvious and helps you designing the L-system: you don't necesarily need images of all the different states of the plant, but it is suficcient to concentrate on a small part of the plant. Let's take a look at a small part of a single leaf :

 

Based on what we know so far, the most obvious approach would be something like this:

Axiom: F      
F --> FF[+FFFF][-FFFF]FF[+FFF][-FFF]FF[+FF][-FF]FF[+F][-F]FFFF

Sure this type of L-system would reproduce the leaf above, but as you might guess it is not very desireable to put all your knowledge in one universal super reproduction rule. The development should be governed by some simple principles. If we take a closer look at the leaf above, we see, that the lower branches could be described as older parts, that had more time to grow, while the upper branches are the newer ones. Of course this could also be implemented with the ideas we have discussed so far (just use a couple of different letters for the different growth phases).
However I would like the chance to show you a new concept, that is very useful in such cases: parametric L-systems. The idea is, that you determine the age of the element by an additional numerical parameter. With this notation the L-system might look like the following:

Lsystem: 0
derivation length: 30
Axiom: A(0)
A(d) : d>0 --> A(d-1)
A(d) : d==0 --> F(1)[+A(4)][-A(4)]F(1)A(0)
F(a) --> F(1.25*a)
endlsystem
    

On the first glance this might look a bit confusing, but on the second glance you'll see, that it is much simpler and more elegant than the brute force approach above. As we would do in the classical case, we introduce two different letters for the two different parts we have in the plant - the old parts, which are just growing straight but are not capable of making any new branching strucutres (represented by F) and the young parts, that are responsible for the branching (represented by F).
The parameter behind F represents the length of the straight element. By multiplying it each time with the factor 1.25 we simulate the growth of this element. The higher this parameter, the faster this element grows and vice versa. The actual value has to be adjusted later during the simulation. Until you know (from the simulation) what value makes sense, you should start with an average growth rate of 25% (i.e. a multiplication factor of 1.25). If this value is too small, all the leaves will overlap, while a too big value creates large empty spaces (it will remind you to a tree in winter ;-)
For the branching element we need two rules the first one is responsible for the aging (A(d) : d>0 --> A(d-1)), while the second one takes care of the actual branching: if the A-parameter is equal to 0 (i.e.: the element is old) it is replaced by the actual branching structure. As you can see we have used 4 as new A-parameter for the branches in the second reproduction rule. Keep in mind that A is responsible for the branching and that the A-parameter is therefore proportional to the density of the branches. If you don't have a feeling which parameter might be usefull, you should start with a big value like 10 and decrease this value until the density of the branches is like you want it to be (if you start with a small value, the calculation of the L-system might take eternities :-) In this case I know from Lindenmeyers book "Algorithmic beauty of plants", that 4 is a reasonable parameter in this case ;-)

If we simulate this L-system we get the following image:

As you can see this image looks already pretty similar to a fern. Nevertheless the spacing between the leaves is a little to big. If we therefore reduce the growth factor from 1.25 to say 1.23 we get a more realistic picture:

The corresponding L-system would be:

derivation length: 30
Axiom: A(0)
A(d) : d>0 --> A(d-1)
A(d) : d==0 --> F(1)[+A(4)][-A(4)]F(1)A(0)
F(a) --> F(1.23*a)    

and as viewing parameters I used the following (in L-Studio again):

angle factor: 15 
initial color: 3 
initial line width: 1

I hope that this tutorial gives you a starting point for your own experiments with L-systems and helps you creating your own plant simulations. In another tutorial we will discuss the more advanced aspects of L-systems (i.e.: how can we create asymmetrical plants? How can we avoid, that all the plants of one type look identical? How can we simulate environmental influences on the plant? ...).

If you need some inspiration, you should check out the demo of
Vue d'Esprit: they have a very good rendering tool, which is capable of rendering highly realistic plants!

When you have created a good plant and you would like to share your work with others, send your L-system (including the viewing parameters) and I will add it here.