Updates on Predator Evolution using a Genetic Algorithm

Updates Timeline

Presentation (5/7/14)

Our team of developers is quite proud of the work they did on this predator-prey simulation. While there are still some minor bugs with collaboration from our predator team, we have trained bots that can effectively catch our rather elusive prey. The presentation portion of the project went well, as many colleagues showed their interest and support in our results. Although our Head Developer, Jigar Dhimar will be graduating and furthering his coding career, we have two developers who show great promise in a possible future for Dumbodore and Sirius Bot. With some further adjustment to fitness, and more training, Marcus Vinicius Pinto Pereira Jr. and Michael Anthony Riley Jr. could bring this predator team to the top.


Ready For Training (5/4/18)

We observed a slight problem after running our tests. Since we decided to allow for one bot to be able to catch the prey when it comes within 100 units of it, our bots seemed to stray a bit from collaborative catches. One bot appears to die off early, while the other one chases the prey and eventually catches it. However, we are very proud that our predators can catch this advanced prey! We have adjusted our fitness function in order to incentivize both predators to stay alive longer, and try to catch the prey together. A two-predator catch will award more points than a one-predator catch. Hopefully with our adjusted fitness function, our robots will learn more of the importance of teamwork!


Integration Station! (5/1/18)

Our developers have now got our predator team sweeping the simple map! The individual predators go to opposite corners of the map, and begin checking each sector for the prey. They send out a "Cheking!" message when they are sweeping, and will send a "Caught!" message when they spot the prey. We have also created our own map using an Xpilot map editor. This map is a larger version of simple.xp. We will begin testing of our predator sweep with our prey in this map, as it will be the one we use for our training. Stay tuned!


What's Next? (4/29/18)

Now that the individual predator code is how we want it, we are ready for some integration! Our developers are hard at work integrating text-message communication, a collaborative sweep, and a collaborative chase method for our predator team of 2 bots. Once communication is up between our bots, the sweep and chase methods will be easier to integrate. With the help of our GA, we expect the predator team to become very efficient at catching the prey.


What's New! (4/26/18)

Our developers have mastered a "sweep" method for our predators. This method allows our predator to effectively search the map for the prey by splitting the map into sectors. Once it checks a sector, it marks it as visited, and moves on to another sector. The next step for our predators is communication. We need to implement this sweep method in a collaborative way, where both predators can search different sectors of the map, and be able to tell each other if the prey is spotted, or if the sector is empty. We are also currently training a prey bot on a genetic algorithm; it will stand as a worthy adversary to run from our predator team. Watch the video below to see some training!