Description: Developement of a LLM-based Pepper robot therapist using LLaMa and Gemma language models throgh Groq API. The system is based on NAOqi library and Choregraphe software.
Description: Implementation of two reinforcement learning algorithm called Behavioral Cloning and DAgger in Reacher and Franka Kitchen Environments
Description: Implementation of the reinforcement learning algorithm PER DDQN (Prioritized Experience Replay Double Deep Q-Networks) to solve the CarRacing-v2 gym environment.
Description: Modeling, planning and control of a autonomous racing car using the Learning Model Predictive Control (LMPC) algorithm to improve lap time.
Description: Rest-to-rest planning and control for a single flexible link robotic manipulator with elastic joint through inverse Dynamic method.
Description: Using of Convolutional Neural Networks algorithm to solve the CarRacing-v2 Environment as an image classification task.
Description: Machine translation task using LLMs and Transformer-based architecture to translate old Italian authors such as Dante Alighieri, Ludovico Ariosto to modern Italian.
Description: Cultural classification task using Transformer and Graph-based methods of Wikidata items.
Description: Development of a deep neural network to classify chest X-ray images as pneumonia or normal, architecture based on a convolutional backbone (DenseNet-169), a module for selecting the most relevant channels called the "Fuzzy Channel Selective Spatial Attention Module" and a KAN (Kolmogorov-Arnold Network) classifier.
Description: Development of a deep neural network for the recognition of emotions from audio: the network was inspired by an existing network called TIM-net (Temporal-aware bI-direction Multi-scale network) in which convolutions were replaced with continuous convolutions.
Description: Design and development of a mobile robot called LIDARino, using Arduino for low-level firmware and Raspberry + ROS for high-level localization.