Bartolomeo Montrucchio

Politecnico di Torino recently collaborated with internationally renowned astronomers to create AETHER, a software developed in Godot that allows the behavior of cometary comas to be modeled in three dimensions by applying known physical principles such as the effect of core rotation and solar radiation pressure.

At present, the variables for modeling cometary comas are partially defined as input by the operator using a trial-and-test procedure, following the analysis of two-dimensional digital telescopic images. The numerical model generated (or, rather, the synthetic image obtained from its orthographic projection onto the two-dimensional plane) is then superimposed on the image to allow a comparison of the correspondence of details. However, since this comparison is purely visual, the verification of the match is subjective and therefore subject to error.

The purpose of this thesis proposal is to modify the existing software, AETHER, in order to automate the comparison and objectively evaluate the synthetic results obtainable from the numerical model with respect to its real counterparts, captured through a telescope. This objective can be achieved using various techniques: for example, by identifying key points on the image that can serve as references for the development of the numerical model, followed by the software generating a series of models within user-definable ranges of the variables that affect the calculation of dust trajectories (Monte Carlo method). Following this operation, the software can verify the correlation between the trajectories generated by the model and those traced on the real image, choosing the best match on a statistical-mathematical basis.

For further information, contact Bartolomeo Montrucchio and Antonio Marceddu at the following email addresses:

bartolomeo.montrucchio@polito.it

antonio.marceddu@polito.it

The goal of this thesis proposal is to create an Artificial Intelligence (AI) system for an autonomous spacecraft. Specifically, the goal is to develop a monoscopic or stereoscopic vision system capable of recognizing the distance to an object and its edges so that it can be grasped. The ability to grasp an object depends on how the robotic arm is constructed, as well as the shape of the gripper at the ends of the manipulator.

Regarding the monoscopic or stereoscopic vision system, one way to determine the distance to an object could be through depth estimation systems, which require significant computing power. These, combined with a system for segmenting the edges of objects in the scene, pose estimation, and other techniques, could enable the creation of a prototype capable of manipulating objects of unknown shape. Contact bartolomeo.montrucchio@polito.it antonio.marceddu@polito.it matteo.dallavedova@polito.it stefano.mauro@polito.it paolo.maggiore@polito.it

The thesis work will focus on the development of a hybrid quantum-classical reinforcement learning agent and its subsequent application to optimize energy consumption within a simulated data-center environment. The student will work with well-established frameworks for classical reinforcement learning and machine learning, such as Gymnasium and PyTorch, as well as with frameworks for quantum programming, including Qiskit and PennyLane.

Description (EN)

The thesis is developed within the SentinelPV research project, which aims to design innovative methodologies for large-scale photovoltaic power forecasting using a limited number of representative “sentinel” plants.

The objective of the thesis is the development and analysis of machine learning models for multi-site photovoltaic forecasting, with a focus on scalability and generalization across distributed PV systems.

The work will be structured into the following main activities.

1. Data analysis and preprocessing

The student will work with real-world datasets including:

photovoltaic production data,

meteorological variables (irradiance, temperature, etc.),

potentially geospatial data.

The activity will include:

data cleaning and missing data handling,

feature engineering,

dimensionality reduction,

analysis of spatial and temporal correlations.

2. Multi-site machine learning modeling

The core of the thesis will focus on developing predictive models for multi-site PV forecasting, addressing the challenge of limited observability in real systems.

Different approaches will be explored, including:

traditional machine learning models (Random Forest, Gradient Boosting);

deep learning models for time series (LSTM, CNN, Transformers);

advanced multi-site approaches such as spatio-temporal models and Graph Neural Networks.

Special attention will be given to:

modeling spatial correlations among PV sites;

using a subset of monitored systems to infer the behavior of a larger fleet;

comparison between single-site and multi-site approaches.

These aspects are central in modern PV forecasting, where machine learning enables capturing nonlinear and spatio-temporal dependencies in complex systems .

3. Model validation and performance analysis

The developed models will be evaluated using standard metrics (RMSE, MAE, etc.) and compared against baseline methods.

Additional analyses will include:

robustness to missing or noisy data,

performance across different forecasting horizons,

generalization capability to unseen sites.

Expected outcomes

The thesis will result in:

the development of advanced multi-site forecasting models;

a comparative analysis of machine learning techniques;

a contribution to the SentinelPV forecasting framework.

This thesis will concentrate on defining and analysing figures of merit for job scheduling in hybrid HPC–quantum systems. The student will review performance metrics from classical HPC (throughput, utilisation, fairness, energy) and quantum computing (fidelity, error rates, queue time, success probability). The work will propose a coherent set of metrics and parameters tailored to hybrid workflows where quantum accelerators are accessed on‑demand. Several composite objectives will be designed, for instance combining user‑perceived turnaround time with expected quality of quantum results. Using a simplified scheduling simulator or analytical models, the thesis will explore trade‑offs between different metrics and how scheduler policies influence them. The expected contribution is a clear metric framework that operators and researchers can use to evaluate hybrid scheduling strategies.. Please contact also dr. Edoardo Giusto

Politecnico di Torino recently collaborated with internationally renowned astronomers to create AETHER, a software developed in Godot that allows the behavior of cometary comas to be modeled in three dimensions by applying known physical principles such as the effect of core rotation and solar radiation pressure.

The following thesis proposal, therefore, focuses on improving this software in order to add new features, including the creation of a small dynamic Solar System capable of correctly positioning the comet at a certain time of year based on the ephemerides obtained from the NASA website.

During the creation of the software, the student will be supported by researchers from the Asiago Astrophysical Observatory and the Politecnico di Torino. The thesis can be carried out remotely or in person at the Politecnico di Torino.

For further information, contact Bartolomeo Montrucchio and Antonio Marceddu at the following email addresses:

bartolomeo.montrucchio@polito.it

antonio.marceddu@polito.it

This thesis will develop a mathematical optimisation model for the scheduling of hybrid HPC–quantum workloads with on‑demand quantum access. The student will formalise hybrid jobs as sequences or DAGs of classical and quantum stages, with constraints capturing resource capacities, network latency to cloud quantum devices, and priority or fairness policies. The scheduling problem will be expressed, for example, as a mixed‑integer or stochastic programming model, potentially including multi‑stage decisions and uncertainty in job arrivals. Different objective functions will be examined, such as minimising makespan, average waiting time, or weighted combinations of performance and cost. Where possible, the model will be solved or approximated using standard solvers and heuristics, and evaluated on synthetic workloads. The thesis aims to provide a rigorous formulation that can guide the design of practical hybrid schedulers. Please contact also dr. Edoardo Giusto

This thesis proposal involves the development, implementation, and fine-tuning of an intelligent system capable of combining text and visual understanding to answer complex queries on technical documentation in PDF format. This system can be based on the Model Context Protocol (MCP) proposed by Anthropic, which represents a significant evolution over current document assistants as it makes them modularly connected. Therefore, it enables the creation of Artificial Intelligence (AI) systems capable of dynamically responding to external stimuli, such as voice and images. Therefore, the thesis will involve developing a system of this type that can be interfaced using natural language and images from Augmented Reality (AR) devices or smartphone apps. Contact antonio.marceddu@polito.it and/or bartolomeo.montrucchio@polito.it

This thesis will investigate job scheduling strategies for hybrid computing systems where an on‑premises HPC cluster interacts with remotely accessible, on‑demand quantum devices. The work will start from a detailed analysis of existing heterogeneous and hybrid scheduling approaches in production and experimental environments. The student will define models for hybrid jobs that alternate between classical and quantum phases, including queueing and latency aspects of remote quantum access. Different scheduling policies (e.g., FCFS, priority‑based, workflow‑aware) will be designed and compared in a simulated hybrid environment. The evaluation will focus on metrics such as job turnaround time, quantum device utilisation, and classical resource idle time. The expected result is a set of guidelines and a prototype scheduler policy for efficient on‑demand quantum access in HPC workflows. Please contact also dr. Edoardo Giusto

This thesis will design and implement a reusable proof‑of‑concept scheduler component for hybrid HPC–quantum environments. The scheduler will accept job descriptions including classical phases and one or more quantum calls, and it will interact with a configurable back‑end representing remote quantum services. The component will be architected to integrate, where available, external quantum resource estimators to inform scheduling decisions such as device selection or expected runtime. Different pluggable scheduling policies (e.g., basic FCFS, resource‑aware, or workflow‑aware) will be implemented and exposed via a clean API. A simulation harness will generate representative hybrid workloads and collect performance data under each policy. The thesis will deliver both the software artefact and an evaluation demonstrating how such a component can be integrated into larger HPC or workflow management systems. Please contact also dr. Edoardo Giusto

Procedural generation can, to varying degrees, automatically create photorealistic, annotated synthetic 3D scenes based on predefined rules. This is very useful for synthetically increasing the number of images to be fed to Artificial Intelligence (AI) systems, such as Object Detection (DE) and Depth Estimation (DE). Currently, several frameworks capable of this capability exist, such as Infinigen, Nvidia Omniverse, and BlenderProc. Most of them focus specifically on the procedural generation of Earth-like environments, such as urban, industrial, and natural scenes, for applications primarily related to the development of autonomous vehicles and robots.

At the same time, there are few applications for the procedural generation of annotated photorealistic space environments, which could help create autonomous robots capable of orbiting Earth to remove space debris or navigate on satellite and planetary surfaces.

The goal of this thesis proposal is to explore solutions for the procedural generation of photorealistic, annotated space environments that could be used for the aforementioned tasks.

For further information, contact Bartolomeo Montrucchio and Antonio Marceddu at the following email addresses:

bartolomeo.montrucchio@polito.it

antonio.marceddu@polito.it

This thesis will focus on integrating quantum resource estimation into hybrid job scheduling for HPC–quantum systems. The student will study existing resource estimators and architectural models to extract parameters such as logical and physical qubits, circuit depth, expected runtime, and error rates under different target technologies. The main goal is to define an interface and workflow that allows schedulers to query these estimators before dispatching quantum jobs. Using this information, the thesis will explore policies for backend selection, batching, and admission control of quantum tasks. A simulation framework will be developed to compare scheduling decisions with and without resource‑aware information. The outcome will be design recommendations and a reference implementation showing how resource estimation can improve utilisation and predictability in hybrid environments. Please contact also dr. Edoardo Giusto

This thesis will concentrate on defining, analyzing, and validating

figures of merit for advanced sensor fusion architectures within smart

environments. The student will review performance and reliability

metrics combining contactless spatial data (millimeter-wave radar

sensors) with biometric and motion streams from low-power wearable

devices (BLE/Zigbee). The work will propose a coherent framework of

metrics tailored to real-time edge computing workflows, optimizing

event detection while minimizing false alarm rates. Several composite

objectives will be designed, for instance balancing system latency, data

consistency, and user privacy constraints. Using simplified sensor

simulators, data fusion models, or hardware testbeds, the thesis will

explore trade-offs between different filtering policies and how edge

computing architectures influence detection accuracy. The expected

contribution is a clear metric and algorithmic framework to evaluate

multi-sensor fusion strategies in home automation.

Notes: The thesis will be carried out in a hybrid mode (Smart Working

for software and code development) with periodic in-person testing

sessions and laboratory activities at the Alessandria campus.

Contact Patrik Resta resta.patrik83@gmail.com