Open positions

Master thesis

Application deadline: 26/01/2024

PIC4SeR Thesis

Robot UWB Trajectory Evaluation with precise optical tracking system

Supervisor:  Prof. Marcello Chiaberge, Alessandro Navone, Umberto Albertin, Gianluca Dara

Skills: Python Basic Level (Mandatory) – ROS2 (Recommended)

Code: 2024PIC01

Topic: Localization

Description: The goal of the thesis is to study and precisely evaluate robot localization algorithms based on Ultrawide-band sensors, collecting trajectory data with an advanced optical tracking system available at the PIC4SeR lab. The objective of the study is the collection of significant and reliable data to identify and characterize the UWB signal loss in Line-of-Sight (LOS) and NLOS conditions.

Robot Social Trajectory Evaluation with precise optical tracking system

Supervisor:  Prof. Marcello Chiaberge, Andrea Ostuni, Gianluca Dara, Andrea Eirale

Skills: Python Basic Level (Mandatory) – ROS2 (Recommended)

Code: 2024PIC02

Topic: Social Navigation

Description: The goal of the thesis is to study and precisely evaluate social navigation algorithms collecting trajectory data with an advanced optical tracking system available at the PIC4SeR lab. The candidate will study the algorithms at hand and test them computing relevant metrics of the task. Social navigation systems need a combined tracking of robot and people motion to be properly evaluated.

Learning Heuristics for Adaptive Path Planning in Social Scenarios

 Supervisor:  Prof. Marcello Chiaberge, Andrea Ostuni, Mauro Martini, Andrea Eirale

Skills: Python, ROS2 (Mandatory) – Path Planning Theory, Deep Learning (Recommended)

Code: 2024PIC03

Topic: Social Navigation

Description: The goal of the thesis is to develop social version of classic path planning algorithms (A*) by learning additional behaviours required to smoothly navigate in social contexts.

Efficient People 4D pose estimation and tracking for social controllers

Supervisor:  Prof. Marcello Chiaberge, Chiara Boretti, Andrea Ostuni, Martini, Andrea Eirale

Skills: Python (Mandatory) –  Deep Learning Basic Level, Pytorch/Tensorflow (Recommended)

Code: 2024PIC04

Topic: Perception Computer Vision

Description: The goal of the thesis is to study, apply and evaluate state of the art solutions for visual people detection and tracking. The thesis aims at setting up a precise performance benchmark to compare people detection and tracking solution with real-time performance to be executed on the onboard hardware of a mobile robot. A second step will aim at identifying or developing an efficient model for detection and tracking of people moving. The model should provide at least a 4D pose of the person (3D bounding box including yaw orientation) to feed and enable social navigation controllers.

Sensor Fusion for SLAM using FUSE in challenging environments

Supervisor:  Prof. Marcello Chiaberge, Giacomo Franchini, Andrea Ostuni, Mauro Martini

Skills: ROS 2, Robot Localization (Mandatory) – C++ (Recommended)

Code: 2024PIC05

Topic: Localization

Description: The goal of the thesis is to study, apply and evaluate state of the art solutions for visual people detection and tracking. The thesis aims at setting up a precise performance benchmark to compare people detection and tracking solution with real-time performance to be executed on the onboard hardware of a mobile robot. A second step will aim at identifying or developing an efficient model for detection and tracking of people moving. The model should provide at least a 4D pose of the person (3D bounding box including yaw orientation) to feed and enable social navigation controllers.

Semantic Obstacle Avoidance of Robotic Arm for Fruits Harvesting

Supervisor:  Prof. Marcello Chiaberge, Marco Ambrosio, Alessandro Navone, Luigi Mazzara

Skills: Python (Mandatory) – Deep Learning Basic Level, Pytorch/Tensorflow, ROS 2 (Recommended)

Code: 2024PIC06

Topic: Robotic Arm Control

Description: The goal of the thesis is to study and develop an obstacle avoidance algorithm for robotic fruit grasping. The work aims at integrating a semantic knowledge of the scene received from a semantic segmentation model with the control system of the robot. The candidate will develop a smart system to match semantic information obtained from visual data (fruits, branches, vegetation) to point cloud commonly used for obstacle detection to compute collision-free trajectory to grasp the target object.

Multi-robot localization: belief propagation on factor graph

Supervisor:  Prof. Marcello Chiaberge, Alessandro Navone, Mauro Martini, Marco Ambrosio, Giacomo Franchini

Skills: ROS 2, Robot Localization Theory, Object-Oriented Programming (Mandatory) – Probabilistic robotics background (Recommended)

Code: 2024PIC07

Topic: Localization

Description: Study and implement a Gaussian Belief Propagation method on a distributed multi-robot systems without the need of a centralized network node. The candidate will study the performance of such system on a set of multiple robots (from 3 to 15)

https://gaussianbp.github.io/

Optimization of large vision models for edge execution

Supervisor:  Prof. Marcello Chiaberge, Simone Angarano, Mauro Martini, Chiara Boretti

Skills: Python (Mandatory) – Deep Learning Basic Level, Pytorch/Tensorflow (Recommended)

Code: 2024PIC08

Topic: Perception Computer Vision

Description: The goal of the thesis is to study, evaluate and optimize state of the art large vision-language models for robotic perception. A wide range of robotic applications need an accurate but efficient perception system. Combining most recent foundation models with optimization techniques (model scaling, pruning, quantization, transfer learning) may lead to overall improved prediction performances

hkchengrex/Tracking-Anything-with-DEVA: [ICCV 2023] Tracking Anything with Decoupled Video Segmentation (github.com).

Robust Autonomous Navigation in Vineyard for complete path coverage

Supervisor:  Prof. Chiaberge, Marco Ambrosio, Alessandro Navone, Andrea Ostuni

Skills:  Python/C++, ROS2 (Mandatory) – Hands-on experience with hardware and robotic platforms (Recommended)

Code: 2024PIC09

Topic: Navigation Optimization

Description: The objective of the thesis is to integrate, improve and complete algorithms for robust full-field navigation in vineyards (straight, curved, plain, slope). The candidate will study both localization-based and position-agnostic solutions for vineyards and orchards traversal, identify critical points to improve in the starting version of the project. Testinng will be carried out in both simulation and real world the resulting complete pipeline (inter-row navigation and intra-row navigation). The output of the work will be a behaviour system to orchestrate the complete navigation task and the evaluated performance.

Spectral imaging for fruit health monitoring

Supervisor:  Prof. Marcello Chiaberge, Alessandro Navone, Luigi Mazzara, Francesco Messina, Gianluca Dara

Skills: Python (Mandatory) – Experience with spectral imaging sensors, Agronomy background (Recommended)

Code: 2024PIC10

Topic: Perception Computer Vision

Description: The goal of the thesis is to collect a structured dataset on field with a multi-spectral sensor and develop an efficient Deep Learning model to extract relevant feature to predict the desired fruit information. The thesis starts from a detailed study of agronomic patterns to detect in multi-spectral imaging and the realization of a dataset.

Fault-tolerant DNN for Space Applications

Supervisor:  Prof. Marcello Chiaberge, Simone Angarano, Mauro Martini, Umberto Albertin

Skills: Deep Learning Basic Level, Python (Mandatory) – Compression Techniques, Embedded Systems (Recommended)

Code: 2024PIC11

Topic: D.L. Space

Description: The thesis aims to develop methodologies for verifying the integrity of neural network parameters used for image analysis in the Edge-AI system. The idea is to avoid designing inherently robust neural networks for two reasons: 1) the robustness obtainable architecturally with appropriate training is never perfect or guaranteed, and 2) the networks thus developed pay for the acquired robustness with degradation in image analysis performance (e.g., the accuracy of a classifier). This WP will analyze approaches based on using a neural network not specifically designed to be robust, coupled with methodologies for the timely detection of parameter errors and their correction.

Knowledge Distillation for Domain Generalization

Supervisor:  Prof. Marcello Chiaberge, Simone Angarano

Skills: Python, Deep Learning Advanced Level (Mandatory) – Pytorch/Tensorflow (Recommended)

Code: 2024PIC12

Topic: Perception Computer Vision

Description: Study and explore knowledge distillation techniques for multi-class segmentation, pose detection, object detection and relevant computer vision tasks for robotic perception.

Solving the Primacy Bias of Deep Reinforcement Learning with Dimensional Growing Neural Networks

Supervisor:  Prof. Marcello Chiaberge, Andrea Eirale

Skills: Python, Deep Learning Advanced Level (Mandatory) – Machine Learning/Deep Learning Basic Level, Reinforcement Learning Basic Level (Recommended)

Code: 2024PIC13

Topic: R.L.

Description: Primacy bias is a well known problem in the reinforcement learning field. Some questionable solutions have been proposed during the years, but the problem has never been faced from its root. This thesis aim at validating the theoretical assumption according to which increasing the agent neural network(s) as experiences are gathered could avoid overfitting over early interactions, allowing the agent for a better and faster convergence.

Industrial Facilities Monitoring using autonomous UAV

Supervisor:  Prof. Marcello Chiaberge, Francesco Messina, Umberto Albertin

Skills: Python (Mandatory) – Matlab, Deep Learning Basic Level (Recommended)

Code: 2024PIC14

Topic: Computer Vision

Description: The goal of the thesis is to develop a Semantic Segmentation and Change Detection framework for multitemporal damage identification and classification from RGB, Thermal and Multispectral images of Industrial facilities roofs. The idea is to recognize objects from a reference image, extrapolate and analizying them in order to identify if a change or a damage occur over time. The final idea is the damage/change classification in order to gather a full dataset of the industrial facility’s roof.

Development of a Payload Dispenser for Lunar Exploration Rovers

Supervisor:  Prof. Marcello Chiaberge, Giacomo Franchini, Gianluca Dara

Skills:  Electro-Mechanic Design, CAD (Mandatory) – ROS2 + microROS (Recommended)

Code: 2024PIC15

Topic: EM Designing

Description: The goal of the thesis is to design and development of a payload dispenser prototype which will be able to release different types of payloads onto the surface autonomously. Design and development of dispenser interfaces with the rover. Analysis of the data collected during mission simulations, in order to assess the performance and payload positioning accuracy.

Development of a 2DOF Traction+Steering Wheel Corner for Outdoor Rovers

Supervisor:  Prof. Marcello Chiaberge, Giacomo Franchini, Gianluca Dara

Skills: Electro-Mechanic Design, CAD, FEM (Mandatory)

Code: 2024PIC16

Topic: EM Designing

Description: This master thesis focuses on the innovative development of an integrated assembly unit that combines traction and steering functionalities for an outdoor rover platform. By seamlessly integrating these components into a single unit, the research aims to enhance the rover’s maneuverability, stability, and overall performance in challenging terrains.

Fault Injection in Normalizing Flow Models for Space Applications

Supervisor:  Prof. Marcello Chiaberge, Carlo Cena

Skills: Python, Deep Learning (Mandatory) – Git, Time Series (Recommended)

Code: 2024PIC17

Topic: Deep Learning Space

Description: The goal of the thesis is to investigate the impact of different faults on a compact Real-NVP deep neural network designed for fault detection in time-series data. The research aims to compare the performance of a baseline model with that of an augmented version incorporating a physics-informed loss. Additionally, the study will explore variations in neural network sizes for both experimental setups.

Sensor fusion as a solution to the Kidnapped Robot Problem

Supervisor:  Prof. Chiaberge, Ostuni, Eirale, Ambrosio, Boretti

Skills: C++, Python, ROS2, Statistics

Code: 2023PIC01

Description: In robotics, the kidnapped robot problem is the situation where an autonomous robot in operation is carried to an arbitrary location. It is currently one of the most difficult and unsolved problems in the robotics field. This thesis project aims to develop a system able to relocalize the robot by exploiting several sensors (cameras, LiDAR, IMUs, Encoders, etc.) and techniques (traditional methods and learning-based) to solve the kidnapped robot problem.

Complex environment exploration

Supervisor:  Prof. Chiaberge, Eirale, Boretti, Martini

Skills: C++/Python, ROS2

Code: 2023PIC02

Description:  In order for the robotic platform to properly operate, the knowledge of the working environment is a fundamental aspect. Often this information is not previously provided and could be difficult or dangerous to retrieve. This thesis aims to develop a policy-based agent to fully explore a complex, unknown environment.

Edge Novelty Recognition

Supervisor:  Prof. Chiaberge, Albertin, Dara

Skills: C, Python

Code: 2023PIC03

Description:  Novelty recognition is a “parallel branch” of predictive maintenance that recognizes novelties from the trained data. It can be useful when no historical data about failures is collected (e.g., a new machine). The thesis concerns the study and development of a novelty detection algorithm to be deployed in an embedded device for predictive maintenance purposes.

Terrain Traversability Analysis for (Planetary Exploration) Rovers

Supervisor:  Prof. Chiaberge, Franchini, Messina

Skills: C++/Python, ROS2

Code: 2023PIC04

Description:  Novelty recognition is a “parallel branch” of predictive maintenance that recognizes novelties from the trained data. It can be useful when no historical data about failures is collected (e.g., a new machine). The thesis concerns the study and development of a novelty detection algorithm to be deployed in an embedded device for predictive maintenance purposes.

Deep Computer Vision for Social Robotics

Supervisor:  Prof. Chiaberge, Angarano, Martini, Eirale, Boretti

Skills: Python, Deep Learning, Linear Algebra

Code: 2023PIC05

Description:  Human-Robot Interaction (HRI) is a fundamental component of intelligent robots for well-being which aims to live beside us in the society. However, the nature of HRI can be different according to the context and the human behaviour. This thesis aims at investigating Deep Learning based robotic perception methods for extracting environmental features, in order to build a semantic representation of the human condition (available to interact/busy) and predict the best way of interacting with him/her (visually, talking, physically).

Navigation and Obstacle Avoidance in Vineyards

Supervisor:  Prof. Chiaberge, Martini, Mazzara, Ostuni

Skills: Python, Deep Learning, ROS2

Code: 2023PIC06

Description:  Low-cost position-agnostic navigation is the solution to GPS-denied conditions in outdoor precision agriculture applications. The two main tasks of a mobile robot in vineyard rows are following the plant rows and avoiding obstacles (boxes, people, etc.) on its trajectory. Multi-class semantic segmentation and multi-objective control strategy can be combined to control the robot’s motion and obtain the resulting final behavior. This thesis aims to improve existing visual-based navigation for vineyard rows.

Industrial Thesis

Optimised time domain response using motion planning innovative strategy

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: Python, System Control Theory

Code: 2023PIC07

Description:  The motion systems provide different response while submitted at diverse motion profiles (stroke, acceleration, speed, jerk, snap). The level of vibration and energy during motion change as a function of the those parameters inducing a different response of its elements (cables, mechanical parts, etc.). The goal of the thesis is to identify possible “strategies” (motion profile, frequency rejection technics, learning algorithm…) which limit the energy induced and optimise the motion systems responses. 

Heat flow management in motion systems for accuracy optimisation

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: CAD modelling

Code: 2023PIC08

Description: The accuracy of motion systems is affected by the thermal transient and gradient during operation. Those elements generate a “non linear” distortion of the structure which create accuracy losses at machine level. The capability to reduce or avoid such effect is then a key element to improve the motion systems performances. The goal of the thesis is to identify possible strategy of thermal management and/or evacuation (e.g. heat pipe) to minimize thermal gradient and transient of the systems during operation from cold to steady state condition.

Apply AI technics to optimise motion systems feed-forward​

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: Python, System Control Theory

Code: 2023PIC09

Description:  The motion systems performances are closely related to the feed-forward capabilities. The systems characteristics are not “always” constant and linear. For instance, the moving “apparent” mass of dynamic cable changes as a function of the position. In addition, non linearity, such as friction, could present different responses over time and based on specific positions and boundary conditions. The goal of the thesis is to analyse in details those behaviours and apply Artificial Intelligence technics to optimised and/or adapt the feed-forward of our advanced motion systems. 

Sensor fusion as a solution to the Kidnapped Robot Problem

Supervisor:  Prof. Chiaberge, Ostuni, Eirale, Ambrosio, Boretti

Skills: C++, Python, ROS2, Statistics

Code: 2023PIC01

Description: In robotics, the kidnapped robot problem is the situation where an autonomous robot in operation is carried to an arbitrary location. It is currently one of the most difficult and unsolved problems in the robotics field. This thesis project aims to develop a system able to relocalize the robot by exploiting several sensors (cameras, LiDAR, IMUs, Encoders, etc.) and techniques (traditional methods and learning-based) to solve the kidnapped robot problem.

Complex environment exploration

Supervisor:  Prof. Chiaberge, Eirale, Boretti, Martini

Skills: C++/Python, ROS2

Code: 2023PIC02

Description:  In order for the robotic platform to properly operate, the knowledge of the working environment is a fundamental aspect. Often this information is not previously provided and could be difficult or dangerous to retrieve. This thesis aims to develop a policy-based agent to fully explore a complex, unknown environment.

Edge Novelty Recognition

Supervisor:  Prof. Chiaberge, Albertin, Dara

Skills: C, Python

Code: 2023PIC03

Description:  Novelty recognition is a “parallel branch” of predictive maintenance that recognizes novelties from the trained data. It can be useful when no historical data about failures is collected (e.g., a new machine). The thesis concerns the study and development of a novelty detection algorithm to be deployed in an embedded device for predictive maintenance purposes.

Terrain Traversability Analysis for (Planetary Exploration) Rovers

Supervisor:  Prof. Chiaberge, Franchini, Messina

Skills: C++/Python, ROS2

Code: 2023PIC04

Description:  Novelty recognition is a “parallel branch” of predictive maintenance that recognizes novelties from the trained data. It can be useful when no historical data about failures is collected (e.g., a new machine). The thesis concerns the study and development of a novelty detection algorithm to be deployed in an embedded device for predictive maintenance purposes.

Deep Computer Vision for Social Robotics

Supervisor:  Prof. Chiaberge, Angarano, Martini, Eirale, Boretti

Skills: Python, Deep Learning, Linear Algebra

Code: 2023PIC05

Description:  Human-Robot Interaction (HRI) is a fundamental component of intelligent robots for well-being which aims to live beside us in the society. However, the nature of HRI can be different according to the context and the human behaviour. This thesis aims at investigating Deep Learning based robotic perception methods for extracting environmental features, in order to build a semantic representation of the human condition (available to interact/busy) and predict the best way of interacting with him/her (visually, talking, physically).

Navigation and Obstacle Avoidance in Vineyards

Supervisor:  Prof. Chiaberge, Martini, Mazzara, Ostuni

Skills: Python, Deep Learning, ROS2

Code: 2023PIC06

Description:  Low-cost position-agnostic navigation is the solution to GPS-denied conditions in outdoor precision agriculture applications. The two main tasks of a mobile robot in vineyard rows are following the plant rows and avoiding obstacles (boxes, people, etc.) on its trajectory. Multi-class semantic segmentation and multi-objective control strategy can be combined to control the robot’s motion and obtain the resulting final behavior. This thesis aims to improve existing visual-based navigation for vineyard rows.

Industrial Thesis

Optimised time domain response using motion planning innovative strategy

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: Python, System Control Theory

Code: 2023PIC07

Description:  The motion systems provide different response while submitted at diverse motion profiles (stroke, acceleration, speed, jerk, snap). The level of vibration and energy during motion change as a function of the those parameters inducing a different response of its elements (cables, mechanical parts, etc.). The goal of the thesis is to identify possible “strategies” (motion profile, frequency rejection technics, learning algorithm…) which limit the energy induced and optimise the motion systems responses. 

Heat flow management in motion systems for accuracy optimisation

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: CAD modelling

Code: 2023PIC08

Description: The accuracy of motion systems is affected by the thermal transient and gradient during operation. Those elements generate a “non linear” distortion of the structure which create accuracy losses at machine level. The capability to reduce or avoid such effect is then a key element to improve the motion systems performances. The goal of the thesis is to identify possible strategy of thermal management and/or evacuation (e.g. heat pipe) to minimize thermal gradient and transient of the systems during operation from cold to steady state condition.

Apply AI technics to optimise motion systems feed-forward​

Supervisor:  Prof. Chiaberge, Albertin, Lacagnina (Etel supervisor)

Collaboration: Etel SA

Skills: Python, System Control Theory

Code: 2023PIC09

Description:  The motion systems performances are closely related to the feed-forward capabilities. The systems characteristics are not “always” constant and linear. For instance, the moving “apparent” mass of dynamic cable changes as a function of the position. In addition, non linearity, such as friction, could present different responses over time and based on specific positions and boundary conditions. The goal of the thesis is to analyse in details those behaviours and apply Artificial Intelligence technics to optimised and/or adapt the feed-forward of our advanced motion systems.