Site gratuit créé sur
Fallen Station is an enigmatic FPS in which players must unravel the mystery of the station they are exploring.
The game's main mechanic is the ability to switch between past and present, at any point in the station. In the present-day map, the station is filled with debris blocking most paths. In the past, defense systems are always on line, detecting the explorer as an intruder to be killed at all costs. Each level will feature its share of enemy AI, puzzles and boss fights.
MY CONTRIBUTION: I was mainly responsible for the AI behavior of the enemies and the construction of the last level with its final boss.
The project was to develop in C++ a physics engine whose objects are rigid bodies and whose collision resolution is iterative using pulses. The OpenGL API was used for the graphical interface.
Phase 2: manage objects made up of several masses by adding different forces and constraints (spring, archimed thrust, ground resistance, etc.).
Phase 1: building a basic particle management engine
Phase 4: addition of a collision resolution system (using an octree)
Phase 3: adding rotational physics
Bonus: Car accident simulation
The project consists in implementing new functionalities to a C++ simulation in which agents are directed by one or more goals: to go and take ammunition, to attack a target or to explore the environment.
MY CONTRIBUTION :
Implementing fuzzy logic in agent aiming and shooting
This allows concepts such as “far” or “slightly” to be expressed in code not by discrete intervals, but by fuzzy sets.
Add target behavior on a designated enemy agent
User can create a team of raven agents
Using a neural network to create a bot-learner
The bot, not knowing how to shoot at first, creates learning data by observing each bot action according to different variables, and measuring the performance of these variables.
Sensivise is a VR serious game made by the DOWiNO studio, which raise awareness of visual disabilities. Set in a company, the player moves around and acts with impaired vision, testing simple solutions to adapt the environment.
MY CONTRIBUTION: I wrote the entire design document that served as the foundation for the game's development, I took part in implementing the quest system and some mini-games, and I took care of assets research and their integration,
Chronomancers is a 2-player local strategy game set in a medieval fantasy world. Based on the time loop principle of Lemnis Gate, the aim is to control an area.
Players are assigned characters with their own attacks and characteristics (archer, barbarian, mage, etc.). Each player in turn chooses an action to perform on a character (move, attack, defend or use items). At the end of a certain number of ticks, the global state of the game is reset: all the actions the player has performed are recorded and will be replayed the following turn, so the opposing player will know them and everyone will try to change the course of events...
MY CONTRIBUTION: Construction of the map, sprite integration and all animations. I also managed the scoring system and the game loop.
The project consisted in creating an agent capable of learning to play Space Invaders. We used the deep Q-Learning algorithm for reinforcement learning, and a convolutional neural network to enable the agent to observe each frame of the game.
To put it simply, our neural network takes a game's fame as input, and as output returns a specific input to perform (move, shoot...).
To enable the network to learn, given a sequence of game states and actions, we'll associate a reward with them, linked to the loss generated by the action performed. The agent's objective is to learn a strategy that maximizes future rewards.
The aspirobot can capture the environment it's in, which is made up of continuously appearing dust and jewels. It can vacuum, pick up or move. Its internal state is in BDI “Beliefs-Desires-Intentions” form.
The aim is to get the mansion clean as quickly as possible, and to suck up as little jewelry as possible. The agent plans its actions using uninformed and informed exploration algorithms (with heuristics). Finally, after X iterations of learning, and thanks to its performance measurement, the agent optimizes its exploration frequency.
The work is the implementation of an agent capable of solving Sudokus, in the form of a constraint satisfaction problem.
The CSP uses the Arc Consistency Algorithm 3 (AC-3) and the minimum-remaining-values, degree heuristic and least constraining value heuristics. For its reasoning, the backtracking algorithm was used.
Just a Night is a horror game created as part of the Brackeys Game Jame 2022. The player takes on the role of a child who, awakened by a thunderstorm, searches for his parents in his house. He soon notices that strange things are happening...
MY CONTRIBUTION: I was mainly responsible for the animations, the player navigation and the whole soundscape.
Application developed in Python that calculates the stresses and strains of various structures using the finite element method.
Avenue Gustave eiffel, Paris 75007
|