DIAG Robotics Laboratory - Thesis topics


Many theoretical and/or experimental activities are available at the DIAG Robotics Lab as master theses. Typically, they are related to on-going research activities, especially in the context of european and italian projects in which we are involved. Occasionally, we propose theses to be done at companies, or at universities and research centers around the world. Self-proposed thesis arguments are also welcome.

Each thesis at the DIAG Robotics Lab is supervised by one of the faculty members. To apply for a thesis, contact Giuseppe Oriolo in advance; it is better to sign up when you have yet 3 or 4 exams to take. It is also desirable to have at least one control or robotics course in your curriculum.

A tentatively up-to-date list of theses is given below. Consider however that this is just a selection of the available topics; for further information check our research webpages.


Interaction force estimation for humanoid robots (added Jun 2017; contact: Vendittelli or De Luca)

The objective of this master thesis is to investigate the use of force/pressure sensors under the feet of humanoid robots to estimate interaction wrenches and to identify the point(s) on the robot structure where the interaction is occurring. Closely related results have been recently obtained by members of the Robotics Laboratory for serial manipulators with fixed base. This  study aims at generalizing these results to humanoid robots, both in static and dynamic conditions. Further details, thesis plan, and relevant bibliography can be found here.

The ideal candidate should be enrolled in Master studies in Robotics or Control Engineering, must have solid background in robotics, particularly for the dynamic aspects, good coding skills in C/C++, knowledge of V-REP.

Motion retargeting for human-humanoid collaboration (added Feb 2017; contact: Oriolo)

This project will take place at the INRIA Nancy (France) in the Larsen team, under the supervision of Serena Ivaldi and Karim Bouyarmane. The internship is for 6 months and it is paid around 600 euros/month.

Within the context of the European project H2020 AnD, the objective of this master thesis is to devise motion retargeting strategies for collaborative tasks involving a human operator and a humanoid robot. The human operator will be equipped with inertial and force sensors (XSens suit, shoes with force sensors) and the robot will use the motion capture data in real-time to perform its own part of the movement by imitating the human movement and readapting it to its own geometry, kinematics, and dynamics. A possible solution will be to apply the multi-robot whole-body control paradigm (Vaillant et al, 2016, Bouyarmane et al 2017) where a rigid-body dynamics model of the human and of the manipulated object will be available for the robot.  The robot can also use a database of the collaborative task performed by two humans, which will provide him with specific domain knowledge regarding the task to execute. The main scientific challenge will be to adapt the muti-robot whole-body controller to the situation in which the motion of one of the robot of the system is an input (constraint) rather than an output (command) of the controller, but till using the full coupled model of the multi-robot system {robot,human,object} to compute the motion of the robot.

Candidates should be enrolled in Master studies in Robotics, Computer Science or related field. Mathematical background, solid background in robotics, control and familiarity with C/C++ are required.

Development and integration of algorithms for manipulation with dual-arm robot systems (added Jan 2017; contact: Vendittelli)

This project will take place in the RIS group at LAAS-CNRS Toulouse (France).  The internship is for 6 months (preferred starting date: February, March at the latest) and it is paid around 500 euros/month.

Within the framework of the European project DualArmWorker, the LAAS-CNRS, Tecnalia and Airbus are working on the realization of a robotics workstation to carry out machining operations on wing ribs of the A380. Currently, the robot used (Nextage) can handle small parts using one arm. However, for larger parts, dual-arm manupulation is required. The first objective of this master thesis is to develop and implement algorithms for the manipulation with two arms of objects in a static environment on a PR2 robot at LAAS-CNRS. Then, the planning algorithms will be extended to avoid collisions with dynamic obstacles by using a Kinect camera. The design of the algorithms should be general enough to be easily ported to other robots. The ideal candidate should have good skills in C/C++, knowledge of ROS, Python and GIT.

Automatic feature selection for visual control of robots (added Jan 2017; contact: Oriolo)

This project will take place in the Lagadic group at Inria/Irisa Rennes (France).  The internship is for at least 6 months and it is paid around 550 euros/month.

Cameras are a widespread sensor modality used for controlling the motion of robots and/or understanding the surrounding environment. In a robotics context, a first necessary step for using cameras is to extract relevant "features" that will represent the actual "measured/controlled quantities" exploited in any subsequent control/estimation task. Features can either be geometrically-based (e.g., points, lines) or appearance-based (e.g., intensity levels of image patches). The choice of which feature to use is a crucial decision, which ultimately affects the convergence rate, stability domain and overall robustness of the control/estimation scheme. In general, features are selected by intuitive considerations or heuristic analyses. However, it would be more reasonable to rely on an autonomous selection algorithm able to understand which features are the most appropriate for the given task and perceived scene. Some recent work has given an initial contribution in this sense, by proposing an automatic selection of "integral" features (i.e., representative of whole image patches) for an optimal reconstruction of the scene. The goal of this internship is to extend this work by (1) considering other optimality criteria, (2) extending the methodology to the case of visual control of robots, and (3) by performing some experiments for validating the approach in real conditions. The ideal candidate should have good skills in Matlab, C/C++, control and estimation.


Analysis of forces exchanged between human and robot during an assembly task (added Jan 2016; contact: Oriolo)

This project will take place at the INRIA Nancy (France) in the Larsen team. The internship is for 5/6 months and it is paid around 500 euros/month.

To study the collaboration between humans and robots, we need to study the exchange of multimodal physical and social signals. During a collaborative task such as assembly, a human can communicate its intentions to the robot by a mix of physical signals (e.g., forces), verbal and non-verbal signals (e.g., touch, speech, gaze). The goal of this project is to analyse the forces exerted by humans on a humanoid robot during a collaborative assembly task. The goal is to analyse the forces, the arm trajectories, the contact locations, and perform a time-series analysis of such signals in synchrony with human speech, to identify automatically actions sequences. The project is in relation with the projects EDHHI and CODYCO, where we performed several experiments of human-robot interaction between 56 non-experts users and the humanoid robot iCub [Ivaldi et al., 2015]. The ideal candidate is good in Matlab or Python, C/C++, signal processing, statistics, robotics.

Learning and optimisation of task priorities for control of movement of a humanoid (added Jan 2016; contact: Oriolo)

This project will take place at the INRIA Nancy (France) in the Larsen team. The internship is for 5/6 months and it is paid around 500 euros/month.

In the context of the European project CODYCO, we are interested in studying the whole-body control of redundant robots in interaction with their environment. In [Modugno et al, ICRA 2016] we recently proposed a multi-task control framework for redundant robots, based on the soft prioritisation of tasks.  The goal of this thesis is to extend previous results to: 1) apply the controller on a humanoid robot such as the iCub 2) explore how the system scales and which machine learning algorithms we can use to learn efficiently offline and online. A promising method will be to rely on bayesian optimisation guided by priors from simulation, as recently proposed in [Cully et al, Nature 2015]. The ideal candidate is very good in C++, Matlab, optimisation, machine learning, robotics.




Random walks with local interaction bias (added Oct 2015; contact: Oriolo)

Collective exploration and exploitation of an unknown environment by swarms of autonomous agents (e.g., ants, robots) requires adaptive strategies that can deal with the probabilistic distribution of resources and that can take advantage of stochastic interactions among agents. The objective is to enhance random walk exploration patterns with information polled from the local interaction network, and the implementation and test of exploration and exploration strategies with a swarm of robots (mainly in simulation). This objective can be pursued in different ways, each exploring some particular aspect of the problem (analytical modelling, relation with specific random walk patterns, relation with resource distribution), hence opening to different thesis topics on the subject. Each topic will have both a theoretical component (literature background, characterisation of different exploration/exploitation patterns) and an experimental component (implementation and verification in multi-agent and/or robotics simulations). The thesis will be carried out within the framework of the DICE project lead by Dr. Vito Trianni at the Institute of Cognitive Sciences and Technologies of the National Research Council in Rome.



A case study of safe human/robot coexistence (added Mar 2015; contact: Oriolo)

This project will take place in collaboration with the Research for Innovation department of Loccioni Group, Angeli di Rosora, Ancona. The objective is to use 3D sensors or equivalent technology to detect operators that are close to a fixed or mobile robot manipulator, and ensure their safety while the robot is performing a task. In particular, the focus will be on industrial robotic arms (6-DOFs anthropomorphic manipulators) and sensor data fusion. Several 3D sensors will be used to perceive the robot workspace, and data from each sensor will be merged in order to reconstruct the whole working environment; information about humans, and in general obstacles, will be inferred and used to control robot motion. Experience with 3D sensors and vision technologies, knowledge of C/C++, expertise with the NI Labview environment, familiarity with Linux are desirable characteristics. The project is scheduled to start in April 2015 and the main activities should be completed by June 2015. A small contribution to logistics and accommodation expenses (350 €/month) is available.







Agonist/antagonist control of cable-sheath transmission for bendable systems - Application to flexible endoscopy (contact: Vendittelli)



This project will take place at the
Icube Laboratory, Control, Vision and Robotics team (AVR), University of Strasbourg, France. The goal of the project is to investigate how the use of two motors placed in an agonist/antagonist configuration and combined with force sensors at the proximal side could improve the distal position and velocity control of a flexible endoscopic system developed by the AVR team. Details are available here.





Development of perception, planning and control algorithms for the humanoid robot NAO (contact: Oriolo)

icub.jpgThe Robotics Laboratory has recently purchased a NAO robot from Aldebaran Robotics. We intend to use it as an experimental platform for designing and validating innovative algorithms for perception, planning and control of humanoid robots. In the immediate future, we plan to start the following activities:

- developing an appearance-based visual controller for robust locomotion in indoor environments (e.g., following a corner or entering a door);
- devising a visual odometry system for keeping track of NAO's motion with respect to the environment;
- planning robot motions for manipulation tasks under collision avoidance and dynamic equilibrium constraints;
- using visual feedback to improve walking pattern generators based on the zero moment point concept;

Master theses are available on all the above topics, plus many others. Self-proposed topics are also welcome.




Navigazione autonoma di UAV (Unmanned Aerial Vehicles) in ambienti indoor (contact: Oriolo)

uav.jpgLa tesi ha come obiettivo la definizione del corredo sensoriale e degli algoritmi di guida necessari per la navigazione di UAV in ambienti indoor. In particolare, si intende dotare un velivolo quadrirotore delle funzionalità necessarie a eseguire autonomamente una obstacle avoidance di basso livello, in modo da consentire al pilota (che opera in remoto) di concentrarsi su quanto sta osservando. Inoltre, ciò renderebbe il sistema più robusto rispetto a eventuali perdite di comunicazione, sia telemetrica che video. La prima parte della tesi sarà rivolta all'individuazione del "minimo" (dal punto di vista di peso, costo e consumo) insieme di sensori necessari per tale scopo. La seconda parte dovrà proporre degli algoritmi di guida che usino le misure sensoriali per realizzare un'obstacle avoidance di basso livello. Tali algoritmi dovranno essere i più semplici possibili, in un'ottica di implementazione e successivo sviluppo.




Tracking visivo di oggetto in movimento per UAV (Unmanned Aerial Vehicles) (contact: Oriolo)

GlobalHawkUAV.jpgLa tesi si propone di studiare il problema del tracking di un oggetto al suolo da parte di un UAV (un quadrirotore) equipaggiato con una telecamera pan-tilt. In particolare, l'obiettivo è quello di sviluppare un algoritmo che permetta sia il tracking visivo dell'oggetto identificato da un operatore al suolo, che la stima della posizione e velocità dell'oggetto stesso basandosi sulle informazioni della piattaforma inerziale imbarcata e sugli angoli di assetto della telecamera stessa. Dallo studio e dal confronto tra le varie tecniche possibili dovrà emergere l'algoritmo che meglio si adatta alle esigenze di calcolo real-time su piattaforme dalla capacità computazionale limitata, come sono le avioniche di bordo.





Controllo collettivo del moto (contact: Oriolo)

team.jpgLa tesi riguarda il settore della Swarm Robotics (letteralmente: robotica degli sciami), parte di una recente disciplina più generale chiamata Swarm Intelligence e dedicata allo studio di processi collettivi auto-organizzanti sia naturali (sciami di insetti, stormi di uccelli, mandrie di bovini) che artificiali (squadre di robot, plotoni di veicoli). L'aspetto più interessante di questi sistemi sta nell'emergere di un comportamento globale (ad esempio, una formazione di volo) come risultato esclusivo dell'interazione a basso livello tra i componenti elementari, senza alcuna azione di supervisione. L'analisi dell'instaurarsi di tali comportamenti può essere portata avanti con i metodi tipici della teoria del controllo, come il metodo di Lyapunov. In particolare, la tesi in questione ha come obiettivo lo sviluppo di un metodo di controllo del moto collettivo per la squadra di mini robot mobili Khepera III disponibile presso il Laboratorio di Robotica. In particolare, si intende considerare il problema dell'entrapment di un bersaglio in movimento da parte della squadra.
 





Esplorazione cooperativa di ambienti ignoti (contact: Oriolo)

coopexpl.jpgL'obiettivo della tesi è quello di sviluppare un metodo SLAM (Simultaneous Localization And Map-building) cooperativo per una squadra robot mobili impegnata in un compito di esplorazione. In particolare, il compito assegnato alla squadra è quello di esplorare un ambiente ignoto, costruendone una mappa e al tempo stesso mantenendo una stima accurata della propria posizione all'interno del mondo. In particolare, si cercherà di integrare l'algoritmo SLAM sviluppato con il metodo di esplorazione SRG (descritto qui) recentemente proposto dal nostro gruppo. Oltre allo sviluppo metodologico, la tesi prevede una consistente attività sperimentale sulla squadra di mini robot mobili Khepera III disponibile presso il Laboratorio di Robotica






Manipolazione non prensile per il robot Butterfly (contact: Oriolo)

butterfly.jpgIl robot Butterfly (descritto qui) è un particolare sistema sottoattuato (cioè con meno ingressi di controllo che variabili controllate) che è stato sviluppato nel nostro laboratorio. Intendiamo utilizzarlo per studiare modalità innovative di manipolazione dinamica, cioè non basate su una presa "ferma" dell'oggetto da manipolare. In questo caso, un possibile obiettivo di controllo è quello di comandare il motore del robot in modo da "lanciare" e "riprendere" la pallina sul profilo del Butterfly. Si tratta di un problema di controllo in feedback che richiede l'adozione di tecniche innovative di controllo non lineare. La tesi, che si avvarrà della presenza di un sistema di visione per realizzare uno schema di "visual servoing", ha dunque una componente metodologica e una sperimentale.



Robotics Laboratory Homepage