Master's degree theses and internships

In the early stage of the definition of a space mission, it is often desirable to have a fast preliminary estimation of the DV cost of a low-thrust transfer. When the transfer is realised through a long spiral trajectory, a quick estimation of the total DV and transfer time can avoid lengthy calculations. For this reason, a number of authors have proposed simple control laws for the variation of specific orbital elements and/or analytical equations for the estimation of the DV associated to a given transfer. The usual approach to develop such approximations is that of integrating GPE using as integrand a fast anomaly variable while assuming the slow orbital elements as constants.

A recent approach developed in our lab proposes a more accurate analytical framework for integrating the GPE formulated as a system of linear differential equations. The thesis work consists of investigating the use of such a framework to develop analytical formulas for the secular orbital element variations, to be then used for the estimation of the DV associated to a given transfer.

Assessing and mitigating the risk of collisions between spacecraft in orbit, especially in LEO, is of paramount importance to ensure the safety of space assets, especially in the current context of the continuously increasing number of resident space objects. The common practice to evaluate the collision risk is by calculating the so-called 2D probability of collisions index, based on analysis of the closest approach between the two spacecraft in the so-called encounter plane. Although a 2D geometric analysis is suitable in most situations, some conjunctions require a full 3D analysis to assess the collision risk properly.

The thesis activity involves reviewing existing methods for 3D collision risk assessment and developing a detailed numerical simulator for assessing their performance. Extensive simulations in several representative conjunction scenarios will be performed to evaluate the respective advantages and drawbacks of the different methods. Areas of improvement shall also be identified.

In the context of Space Domain Awareness (SDA), tracking Resident Space Objects (RSO) in the Earth orbital environment, including space debris, allows to predict the trajectories of conjucting bodies in order to evaluate the risk of collision, and therefore to assess the need for a collision avoidance maneuver.  Autonomous space-based tracking of RSOs is becoming a popular research topic, due to the advantages it offers in terms of speed, coverage, revisit times, and reduction of human operators' burden.

The objective of this internship and thesis is to implement a validate an Extended Kalman Filter (EKF) in Matlab and Simulink to use it on-board a space-based observer in order to track target RSOs by processing astrometric positions from optical images.

The activity is structured as follows:

• Literature review of space-based SDA

• Simulation of the conjuction scenario
• Simulation of the astrometric measurements
• Implementation of the EKF
• Analysis of the results and validation

Requirements (actually not mandatory, but warmly recommended):

• Having attended the MSc courses in Spacecraft Orbital Dynamics and Control and/or in Spacecraft Deep Space Navigation
• Basic knowledge of Matlab and Simulink

Tutor: Stefano Palmiotto

The Earth orbital environment is getting increasingly crowded over the years. Among the different strategies to face this issue, Space Domain Awareness  (SDA) plays a role. SDA includes tracking Resident Space Objects (RSO) with optical sensors. A promising as well as challenging technology is Event-Based Vision, also known as neuromorphic vision, which takes inspiration from the functioning of the biological retina. Unlike a conventional frame-based camera, a, Event-Based Camera (EBC) only registers brightness variations, called events, in its field of view, generating a video-like output  known as event stream. Therefore, EBC technology is an attractive choice when it comes to reduce data rate, power consumption and redundancy of information in the scene, allowing to better handle the typically huge amount of data collected during observations of  RSOs.

Simulating event streams in a SDA context helps us to analyze the performances of an EBC and make decisions when designing a tracking system. Some studies have been conducted to model and simulate the output of an EBC, but further improvement is recommended. This thesis aims to explore and implement a model to simulate a realistic event stream in a SDA context. The activity is mainly divided in the following steps:

• Literature review of the models
• Software implementation
• Output analysis

Tutor: Stefano Palmiotto

The optimization of deep learning models for resource-constrained onboard satellite avionics is a critical area of research, as it facilitates autonomous vision-based operations in space while minimizing computational overhead, energy consumption, and latency — parameters that are fundamental to the success of missions constrained by the size, mass, and limited power budgets of modern satellites. 

Objectives:

• Investigate deep learning algorithms for vision-based spacecraft pose estimation, focusing on real-time onboard deployment.
• Develope and evaluate pruning techniques to compress and accelerate neural network models to meet memory and computational constraints of embedded hardware platforms.
• Explore hardware-aware optimization strategies to adapt pruned models for specific embedded devices, considering their unique capabilities and limitations.
• Validate the approach with experiments using synthetic and real spacecraft imagery, benchmarking pose estimation performance and resource usage before and after pruning.

Skills that will be acquired:

• Experience with Python /PyTorch for neural network design, training, pruning, and deployment
• Use of Ultralytics (YOLO) frameworks for object detection, keypoint regression, and leveraging built-in model pruning utilities.​
• Implementation of neural network pruning, quantization, and other model compression techniques for embedded hardware.​
• Image processing using OpenCV for spacecraft image analysis and data preparation.​
• Familiarity with embedded systems deployment, specifically for  NVIDIA Jetson Orin Nano.​
• Benchmarking and profiling deep learning models in constrained environments, including performance and efficiency analysis.​​
• Solid understanding of the spacecraft pose estimation problem, associated challenges, and relevant performance metrics.

The Microsatellites and Space Microsystems Lab has recently installed a new robotic testbench facility to simulate navigation scenarios in close proximity operations between a chaser and a target spacecraft. The facility features a 6-DoF robotic arm mounted on a linear rail (7-DoF system) and is controlled via ROS. The thesis focuses on developing a digital twin that accurately replicates the facility's physical properties and a GUI for trajectory planning. This enables software validation before hardware deployment preventing potential damage and reducing testing time.

Activities

• Setup simulation environment: Install and configure Gazebo/Ignition, integrate with lab ROS stack
• Digital twin refinement: Import existing URDF, refine CAD models, configure physics parameters and sensor models (cameras, OptiTrack)
• ROS integration: Ensure digital twin replicates MoveIt! kinematic solvers and control interfaces for transparent sim/real switching
• GUI development: Design graphical interface for trajectory planning that incorporates orbital rendezvous dynamics (Clohessy-- Wiltshire equations, natural motion) while accounting for workspace limits, joint limits, and collision avoidance
• Documentation and validation: test generated trajectories in simulation then deploy on real facility

Objectives

• Functional digital twin integrated with existing infrastructure
• GUI for physics-based trajectory planning (rendezvous dynamics + robotic constraints)
• Interface documentation for collaboration- Validated sim-to-real workflow