Davide De Vita

 Hold Close

Android drag-and-throw game, based on a popular silent story on  internet. You play as either the fire or water spirit, and try to reunite with your love.

As the fire spirit, you can burn your way and get fueled up by the free flames sparse in the forest; but the rain is deadly to you so you must hurry!
As the water spirit, you are a more gentle soul so you must find another way across; the rain heals you but the bare walking on the dry leaves may drain you.
Also, beware of the operous ants, in order to preserve their home they will flee from fire and gather water to extinguish it, so if you are fire you may want to dodge their drops, but if you are water you may want to keep your distance from them!

The game recyles its resources, and all the levels, character, npcs and events are data driven, so anyone could create a new scenario with an XML.
The ants, accept either a Decision Tree or a State graph as AI module for their decisions, and every damage affects both the hit entity and the hitting entity!
The general rule is: the less HP something has, the smaller it becomes! If you are small enough you may go through narrow crevices, but you are one step away from oblivium!

Adaptive Pac Man

This was the project I implemented alongside my MSc dissertation.

 I feel strongly about the potential of Dynamic Difficulty Adjustment, and about how ML can help with a customized experience.
That is why I implemented a fair-replica  of the original Pac-Man game, studying the documents left behind by the original creators, and upgraded it adding a learning module.

The game observes the player and decides wether they need a bit more challenge, or if it needs to cut them some slack!
The changes were made to never give a free win or certain loss to players, and follow various policies according to the adapting style we desire to adopt.

The learing process was achieved creating many  player-simulating bots, that would play countless games against different game setups in order to feed a Reinfrocement Learning model. Such model shall, during the game, evaluate the player and propose the best combination to meet their expertise level!

Results were tested on many players of different age, gender, expertise and background in order to understand which game they enjoyed most (between non-adaptive and two adaptive versions), if they felt that the game was changing (they did not know the purpose of the experimet beforehand) and their general comments!