Rabbit HabitsFor an academic project, we created an environment for a pack of NPC wolves with a mix of two machine learning algorithms, to achieve the goal of more effectively hunting a rabbit before it reached it's home.
The Wolves had re-enforced learning applied during the simulation, as well as genetic learning to breed the superior wolves for the next round. |
Results
Even though pack hunting was not introduced with the original values (each wolf was randomized). After many generations, the wolves took on a group strategy. 3 wolves became trackers, and would scour the map for sight or scent of the rabbit. Upon finding it, they would howl and attract more wolves, but would rarely engage the rabbit. One wolf however had no sense of smell or tracking, this hyper aggressive wolf was attracted by the howls. Upon reaching the site, it would attack the rabbit, joined by any wolves that were currently stalking it.
We believe the introduction of more triggers and states would create many more unique pack hunting strategies
Below are my roles in the project, as well as a video showcasing the features of the simulation/game.
We believe the introduction of more triggers and states would create many more unique pack hunting strategies
Below are my roles in the project, as well as a video showcasing the features of the simulation/game.
|
Automated RabbitAutomated Rabbit to allow many generations of wolves to learn before a player plays.
|
VideoLink to video on youtube displaying the features
|