September 29, 2004
Eric L. Goldstein
Peter J. Ottomanelli
Anya L. Roberson
|
|
|
The "Organisms" problem is an extremely open-ended one with seemingly infinite possible approaches to creating a successful, stable organism
population. However, probability plays a large role in increasing the difficulty of sculpting a strong, agile player because any particular
organism can choose from at most nine actions, namely sitting still, moving in one of the four cardinal directions, or reproducing in one of
those four directions. Therefore, any group attempting to tackle the problem of engineering a good player must implement
the maximum number of nondestructive, simultaneously successful strategies in order to increase an organism's probability of a beneficial move.
Most games that require probabilistic models to predict their outcomes involve a substantial amount of chance, including Organisms II. However,
the more highly populated the game board and/or the harsher the board environment, the more this game becomes based on chance rather than on
a successful probabilistic model.
In a game like Organisms II, however, concrete probabilistic methods are difficult to develop because many variable factors affect what actions are beneficial or detrimental to an organism's survival. Therefore, after each move, one can adapt an organism's behavior to its new surroundings rather than rely on a probabilistic prescription. We had two main strategies in light of these observations. One involved the communication between neighboring friends in order to pass information about our organisms' local surroundings and adjust behavior to fit the local conditions. Because gathering information about the conditions on the entire board is incredibly difficult, if not impossible, our communication method would at least give an organism an idea about what the board looked like locally. Our other main strategy was modeled after one widely known and utilized in the chess world, wherein a skilled chess player never moves the same piece twice. We took that strategy and adapted it to fit the rules of the game. Generally speaking, an organism's chance of performing a particular action twice consecutively is very low. These "actions" include anything an organism can possibly do including moving, sitting still, or reproducing. One key benefit to this strategy is that it partially prevents overpopulation, thus increasing the chances of a higher average energy level per organism. This strategy also draws out the life-cycle of the game, decreasing the chances of our organism burning out. This and other strategic implementations will be discussed in depth later in this document. |
|
|
|
A sufficient understanding of the game's fundamentals is required before one can truly conceptualize and create a strong, versatile organism
able to survive in a maximum number of different conditions. Versatility is a key element in tournament success because its game conditions
varied, sometimes considerably, from one game to the next. Overall, the general trend observed during multiplayer games can be divided into
four unique phases.
The Game's Four Phases
|
|
|
 Screenshot of our success against Black Plague. Our character was initially designed to exist independently. The goal was for the player to survive under the most circumstances possible within the single player game. Once we were satisfied with its performance alone, we introduced Black Plague, a high-performing group from the last year (2003). Our initial results were highly discouraging. Our organism exploded and swarmed well, but Black Plague would take over around turn 2000. This did not bode well, and the initial multi-player tests against the rest of the class had the organism "What are ya, Yella?" (v0.5) dying within 500 turns. The lesson quickly learned was to temper the frequency of movement as well as reproduction. Thus, "Mello Yellow" was born. The initial test results were positive, but the organism was still suffering from a form of "burn-out," where Black Plague had a higher energy to unit ratio than any of our organisms throughout the game. When the game approaches a steady state level, this proved disastrous as Black Plague could continue expanding beyond our steady state and eventually choke out our player. Through a series of fine-tuning tests, additional timers were introduced to work in conjunction with tolerances already in place, leading to our final player. This player performed two important actions. One, it was not the worst player in the multi-player tests. Two, it was defeating Black Plague in two-player games either to extinction or to a point where it could not even exceed a twenty-percent share of the map. |
|
|
General
Communication
Movement
The next movement threshold was a timer. Our organism was designed to survive in a wide range of conditions. To this end, frequent movement was not beneficial for all values of p and q. To ensure our organism did not move too frequently, the organism has a move "timer" that is set at 5 turns after a move not towards food. This helps to ensure the organism does not move wildly about the board, wasting energy, even if it has high energy. Reproduction
 Number of Organisms (y) vs. p Value (x): shows the unimportance of q with respect to population. This is a key benefit of our organism. Weaknesses
|
|
|
|
Our organisms produced solid and moderately healthy populations in all of the single player games in the tournament. This reflects the fact that our
organism was originally designed to survive in a single player game under a variety of conditions. While the total energy was not always high for our
organisms, compared to others with similar populations, our populations were always sizeable and stable.
In multi-player games our populations usually died off when p was low. This is most likely due to the fact that p is more important than q to our players, since they consume any food they find, rather than waiting for it to grow. The only situations under which our Player1 survived were those with a low movement penalty v. The main difference between our Player1 and Player2 is that Player2 tempers its movement frequency, so with high v and many players on the board, Player1 would move too much and burn its energy before finding enough food to sustain itself. This also accounts for Player2 performing almost uniformly better than Player1. Under extremely harsh conditions (p <= 0.005, q <= 0.02), Player1 performed better (lived longer) than Player2, though both still became extinct. Player1's better performance is perhaps because Player1 is more prone to moving, and therefore increased its chances of finding food on a harsh board. Player2 survived in multi-player games, as long as p > 0.005. Again, since our players consumed all food found, we were at a disadvantage under harsh board conditions, as we never allowed the food to grow, nor did we have any food protection mechanisms. Player2 did particularly well under the default board conditions, regardless of board size. This makes sense, as most of our testing was done under the default conditions. Player2 fared better under the default conditions with a larger board size, most likely because this reduced the possibility of overpopulation starving the player. |
|
|
| Our overall impression was that, despite our particular approach to solving the "Organisms" game problem, any well thought out set of strategies that follow our general guidelines as well as adding and improving to them will result in a successful player. This open-endedness made the project challenging yet extremely rewarding regardless of the fact that we did not win the tournament. |
|
|
| We would like to acknowledge the contributions of the following groups and their organisms towards our work. Black Plague - This organism provided us with an enemy to initially try and defeat. Their strategy of age and waiting allowed us to come up with a counter strategy of initial expansion and then more waiting. The Net - Their idea of using communication between units to keep them from clustering up. This led to our deciding of movement for a unit based on friends present. |