For Students

Available topics for MSc thesis projects are available in the following.

Safety strategies for collaborative industrial manipulators

A robot is used to palletise boxes and is also responsible for placing the empty pallet.
An operator, in turn, is responsible for bring away the load with a pallet jack. His/her safety must be guaranteed at any time.
The work will explore an approach based on the separating plane (Support Vector Machine) between the robot and the operator tracked with a 3D vision system.

Contact: Prof. Andrea Maria Zanchettin


Localisation and local planning for mobile robots

Localisation and local planning are two important parts of a mobile robot navigation system.
The first one is in charge of localising the robot in the environment, i.e., estimating its pose, the second one uses this estimate to control the robot so that it follows a desired path/trajectory.
Though localisation and local planning are tightly interacting, they are usually studied and implemented as two separated blocks.

The aim of this work is to study the possible interactions between the two subsystems, and to create a new navigation framework where localisation and local planning tightly interact, helping each other, in order to increase the safety and performance of the navigation system.

The work include a methodological part (theoretical analysis and redesign of the localisation/control systems), and a validation of the navigation system in simulation or an experimental validation.

Contact: Prof. Luca Bascetta


Development of a Modelica package for modeling and simulation of discrete event systems

The object-oriented language Modelica, designed for the modeling and simulation of complex engineering systems, is essentially based on the description of continuous-time dynamic phenomena, but still allows an accurate description of events, i.e. discontinuous variations. On the other hand, the design of engineering systems is initially based on functional descriptions, to which a discrete events modeling approach is applied. Exploiting the language mechanism for event management and applying appropriate theoretical formalisms (time Petri nets), the thesis will have as its objective the design and implementation of a Modelica package for modeling and simulation of discrete event systems, naturally integrated with the description of the continuous dynamics of the process. Application to an experimental production system is envisaged.

Contact: Prof. Gianni Ferretti


Online identification of cornering stiffnesses

In a previous thesis work, the problem of offline estimation of a vehicle’s cornering stiffnesses was tackled by an LFT (Linear Fractional Transform) reformulation of the single track model and the subsequent application of a maximum likelihood method. The good results obtained suggested the extension of the approach to an online estimate. The thesis work will therefore have as objective the implementation of the online estimation method, considering both an experimental type vehicle and, possibly, a suitably instrumented real vehicle.

Contact: Prof. Gianni Ferretti


Autonomous navigation in human crowded scenarios

Thesis proposals are available on the design of autonomous navigation algorithms, including local and global planning, for personal mobility devices in human crowded scenarios.
The aim is to design an autonomous navigation system that allows a personal mobility device (or a robot) to be perceived as socially acceptable.
This implies that the trajectories planned and followed by the vehicle should be as much as possible human like, and social constraints, like proxemics, should be enforced.
The works include a methodological part (theoretical analysis and design of the control system), and an experimental part (on a commercial electric wheelchair).

Contact: Prof. Luca Bascetta


Autonomous car control

Thesis proposals are available in the area of autonomous car control, mainly, but not only, focusing on the following topics: trajectory tracking control systems for driving a car at the limits of handling, lane-change stabilisation (system analysis and advanced control design).
The aim is to design control systems for controlling vehicles at the limits of handling, either in order to support autonomous driving or to increase safety of human-driven vehicles during emergency manoeuvres.
The works include a methodological part (theoretical analysis and design of the control system), and an experimental part (on a scaled car model).

Contact: Prof. Luca Bascetta


Multiple robot motion planning with communication constraints

 

The goal of this thesis is to develop a motion planning algorithm (for example based on sampling-based planning methods) to compute a coordinated path for multiple robots such that the robots always stay within a connection range with at least another one to transmit/receive information.
We also consider a specific scenario where there are two different groups of robots: while the first group of robots need to execute an assigned task (for example surveillance), the second group needs to behave as a relay between the first group of robots and a decision centre.

Contact: Prof. Luca Bascetta


 

Calculation of working sequences for double-head wire bending machines

 

In the metalworking industry a very important part concerns the modeling of metal wires, or rods, which, through successive bends, make it possible to create numerous objects for the most disparate uses. The bending of the wire takes place through specialized machinery, which can bend metal rods up to 10 mm in diameter with extreme ease, thanks to the versatility of the variable bending radius. The thesis work therefore has as its objective the determination, automatically, of all the admissible bending sequences, that is, without collisions with the machine components and with the wire itself, given the geometry of the piece to be made. The developed solution is based on the modeling of the geometric configuration assumed by the wire during processing as a manipulator with cylindrical links, whose configuration, in fact, evolves during the manufacturing process. The complexity of the task grows considerably in the case of a double-head machine, that is equipped with two folding devices on both ends of the wire.

Contact: Prof. Gianni Ferretti


 

Collaborative robotics for assembly operations

 

We want to explore the introduction of collaborative robots in the assembly line to perform assembly operations (e.g. screwing), as well as machine tending ones (e.g. loading/unloading a pneumatic press for high-force assembly steps).

The topic is especially tailored to the assembly problem of double effect pneumatic cylinders, which also entails manipulation of deformable objects (as, for examples, o-rings and gaskets) and will be developed in collaboration with a company.

Contact: Prof. Andrea Maria Zanchettin or Prof. Paolo Rocco


 

 

Use of AI to manage agile manufacturing systems

 

We want to use state-of-the-art AI mechanisms to control the operations (scheduling and dispatching of activities) of an agile manufacturing system populated by collaborative robots and human workers.

In particular, mobile manipulators will be adopted so to be easily relocated to other stations of the shop-floor.
The production will be then quickly reconfigurable and able to respond to demand fluctuations.

Contact: Prof. Andrea Maria Zanchettin or Prof. Paolo Rocco



Robust monitoring of human task advancement

In collaborative robotics applications, predicting the evolution of the current human activity allows for a better coordination among agents. Tasks however can be executed in multiple ways and humans can make mistakes.

Can we develop a robust method for real-time monitoring of long human operations with variants?

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Flexible collaborative assembly: multiple products

We have already developed a method for flexible task allocation and scheduling in a collaborative robotic cell. However the optimization in case of multiple products can become computationally intensive.

Can we learn an intelligent policy to determine which product to start at any time?

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Flexible collaborative assembly: adapt to human errors

We have already developed a method for flexible task allocation and scheduling in a collaborative robotic cell, however it is prone to errors.

What if the human performs the wrong action?
What if the human performs the right action wrong?

Can we develop an algorithm that is robust against human errors?

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Flexible collaborative assembly: adapt to human preferences

We have already developed a method for flexible task allocation and scheduling in a collaborative robotic cell, however it is prone to errors, however receiving strict commands may induce discomfort in the human.

Can we develop an algorithm that allows the human to chose its task among the feasible ones?

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Exploiting semantic and geometrical similarities for human activity recognition

Recognizing the actions performed by an operator is crucial for providing the proper cobot assistance.
Factor graphs efficiently describe the correlation among: detected objects in the scene, human postures and sequence of human actions in time.

How to characterize such correlations? In this thesis the possibility to use geometrical similarities (for objects) and semantic similarities (for the actions) will be explored

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Modeling human’s reactive strategy during collaboration with the robot

During human-robot collaboration (HRC) the robot can predict what the human is intended to do.

In our model of HRC we could also take into account the coupling between the robot action and the human reaction in terms of action.

We want to model how the human strategy during cooperation is influenced by the robot behavior and how the human adapts to the robot behavior

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Game-theory approach to human-robot collaboration

The human-robot collaboration can be interpreted as a two-players game where the human utility function is not known a priori and must be estimated online

At each time step the robot must decide whether to chose the strategy that corresponds to the Nash equilibrium (best compromise for both agents) or show the human the optimal strategy, based on the human’s estimated attitude

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Robot manual guidance : collaborative guided insertion

The problem here is a collaborative guided insertion of a generic object inside a partially open box

The objective is to devise a method that helps (guides) the human to insert an object mounted on the end-effector inside a box which has some holes of generic shape.

The remaining part of the robot arm has not to collide with the box during the insertion (e.g. some insertion constraints could be relaxed).

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Robot manual guidance with obstacle avoidance

The framework of this thesis is manual guidance.

We assume that the human is guiding the robot towards the goal position and that an obstacle exists along the path.

The purpose of this thesis is to develop an algorithm to help human avoiding to collide against the potential obstacle

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Safe homing for heavy-load collaborative robots

Within a project on the design of a new collaborative robot for heavy loads, our objective is to develop a system capable of detecting the human presence while the robot is coming back to the home position after the accomplishment of a certain task (e.g. lifting a heavy object).

Two main problems are of interest:
-choice of external sensors
-collision avoidance criteria

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin


Applications and case studies in collaborative robotics

The MERLIN Lab has been active for several years in research on human-robot interaction and collaborative robotics

A spin-off company, Smart Robots, has been recently created. It will bring a new device for human-robot collaboration to the market

Theses will be available on development of case studies and applications of collaborative robotics with different robotic platforms

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Manipulating laundry appliances doors

To open or close the door of a washing machine or a dryer, a combination of computer vision and machine learning needs to be integrated for developing the required functionality.

We want to give the ability to our system to robustly detect the door and manipulate it to achieve the required tas

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Soft bodies grasping modelling

The modelling of soft bodies can be done using different physics engines, but this functionality is unfortunately unavailable in robotics simulation platforms.

The purpose  of this project is to model soft bodies with particular attention to clothes and their interaction with grippers. We aim to import the relevant module as a plugin or develop our own

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin


Self-organized aggregation of robots without computation

A method has been developed for self-organized aggregation of embodied robots that requires no arithmetic computation. The robots have no memory and are equipped with one binary sensor, which informs them whether or not there is another robot in their line of sight.

Robots rotate on the spot when they perceive another robot, and move backwards along a circular trajectory otherwise.

We want to study the properties of this method

Contact: Prof. Paolo Rocco or Prof. Patrizio Colaneri



Dynamic motion planning of a legged humanoid robot

We want to plan dynamic motions for a legged humanoid robot.
Dynamic motions may include jumping or pushing exploiting contacts with the environment.
Experimental validation on CENTAURO and COMAN+.
Methodology: Optimal control (CasADi)

Thesis to be developped at IIT (Italian Institute of Technology), Genova

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Manipulation tasks with a humanoid robot

We want to perform manipulation tasks with a humanoid (e.g. pick a box from the ground) while avoiding self-collisions and collisions with the environment, and simultaneously keeping balance.
Experimental validation on CENTAURO and COMAN+.
Methodology: Sampling-based motion planning for floating-base systems.

Thesis to be developped at IIT (Italian Institute of Technology), Genova

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Balancing control of a humanoid robot

We need a control layer that can enforce balance during locomotion and manipulation tasks, based on a simplified model (e.g. linear inverted pendulum).
Experimental validation on CENTAURO and COMAN+.
Methodology: Inverse dynamics, admittance/impedance control

Thesis to be developped at IIT (Italian Institute of Technology), Genova

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Locomotion of a humanoid robot

We are able to make CENTAURO crawl (3 feet are always in contact with the ground) and COMAN+ walk statically.
Yet we would like to make CENTAURO trot (2 feet are in contact with ground) and COMAN+ walk dynamically, to speed up locomotion.
Experimental validation on CENTAURO and COMAN+.
Methodology: Dynamic walking

Thesis to be developped at IIT (Italian Institute of Technology), Genova

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Human-robot collaboration control based on passive velocity field control and MPC

This thesis aims to define a safe controller for physical human-robot collaboration. Based on passive velocity field controller, MPC techniques will be applied to define the reference for the lower level controller, actively assisting and collaborating with the human.
An interaction dynamics model aiming to predict the collaboration state will be design and exploited by the MPC. The proposed controller will be stable, while empowering the human.
Work with a FRANKA robot for the experimental evaluation of the proposed approach.

Thesis to be developped at USI/SUPSI IDSIA, Manno (Switzerland)

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



AI techniques for interaction control under dynamics uncertainty

Manipulators have to identify the context in which they are working, adapting their controller to the properties of the environment. However, it is not possible to tune a controller for each working scenario. Therefore, the manipulator has to implement a procedure to autonomously identify the working conditions and tune its controller consequently.
Artificial intelligence plays a fundamental role in such a field. The presented thesis aims to define a machine learning approach in order to give such degree of autonomy to the manipulator, being able to face unforeseen situations adapting it controller.
Work with a FRANKA robot for the experimental evaluation of the proposed approach.

Thesis to be developped at USI/SUPSI IDSIA, Manno (Switzerland)

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Intelligent task learning and optimization through minimum human demonstration

Robots are nowadays required to learn tasks quickly, adapting to uncertainties. Learning has to be efficient, requiring minimum trials and iterations.In order to efficiently teach a task to a manipulator exploiting human demonstrations, this thesis aims to investigate artificial intelligence techniques.
Reinforcement learning will be applied in order to quickly learn a specific task. In addition, control techniques will guarantee that the compensation of task uncertainties will be faced, maintaining active the robot learning capabilities.
Work with a FRANKA robot for the experimental evaluation of the proposed approach.

Thesis to be developped at USI/SUPSI IDSIA, Manno (Switzerland)

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Haptic-guided shared control tele-manipulation

Tele-manipulation causes a high cognitive load on the operator
Autonomous system are not sufficiently trusted by conservative industries
Haptic-guide teleoperation (HGT) aims at making teleoperation easier
It is proofed HGT significantly reduces the cognitive load on human operator

Thesis to be developped at the University of Lincoln (UK)

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin



Learning the robot controller from samples

Controlling a robot in a complex environment or to perform a difficult task is cumbersome.
These tasks are still performed via teleoperation.
AI can be used a full (or partial) model for the controller.
This project involves Deep time series to learn a complex task.
E.g., we have prepared a dataset of needle insertion task in robotic surgery context. We aim at learn the controller which performs this task using Deep time series, e.g. Recurrent Convolutional Neural Network or Long Short Term Memory.

Thesis to be developped at the University of Lincoln (UK)

Contact: Prof. Paolo Rocco or Prof. Andrea Maria Zanchettin