PPO-Humanoid

Proximal Policy Optimization for continuous control. The agent learns to walk in the Humanoid environment from OpenAI Gym, mastering complex locomotion through reinforcement learning.

Adversarial Attacks on Neural Networks for Radar Perception

Research project on adversarial attacks against neural networks used for FMCW radar perception. It investigates their vulnerability to adversarial perturbations and whether such attacks can be physically realized in the real world.

Stable Diffusion Real-Time Game Engine

A diffusion model generates the next video frame in real-time, trained on Super Mario Kart. Training data was collected by an AI agent learning to play the game, combining RL-based data collection with generative modeling.

SPH-CUDA

Smoothed Particle Hydrodynamics simulation using CUDA for parallel computation and raylib for real-time visualization. Uses a uniform grid for neighbor searches, scaling to thousands of particles.