Ricerca CERCA

15-Modellazione, simulazione e controllo di velivoli

Safe Reinforcement Learning

azienda Tesi esterna in azienda    


Riferimenti esterni Enrico Busto - ADDFOR

Gruppi di ricerca 15-Modellazione, simulazione e controllo di velivoli

Descrizione Reinforcement Learning (RL) [1] is a class of machine learning algorithms in which an agent interacts by
trial-and-error in an environment.
RL together with Deep Learning has obtained excellent results in a great number of simulated
environments, like video-games [2] or board-games [3].
The main obstacle to the application of RL in a real scenario is represented by the need of exploration: the
agent needs to acquire information about the environment. During the acquisition it can cause critical
damages to what surrounds itself, preventing a possible deployment as intelligent control system.
A promising approach to make RL applicable in the real world is Safe RL [4].
A lot of recent studies [5, 6, 7] have followed different directions in order to overcome the original RL
limitation. The goal of this thesis is to deepen the Safe RL approach in order to investigate the possibilities and the
limits of solutions proposed in the literature with respect to a real world application.

Planned Activities
1. Acquire strong theoretical basis on Deep Reinforcement Learning (DRL);
2. Deepen the approach of Safe RL applied to DRL algorithms;
3. Compare Safe RL solutions in a real world application.
Required Skills:
Good knowledge of machine learning from a probability perspective;
Good knowledge of linear algebra;
Good knowledge of algorithmic.
Optional Skills, considered as a plus:
Proficiency in at least one programming language (Python, Lua, Matlab, C++, Java);
Basic knowledge of Automatic Control.
Competencies to be acquired
Expertise on recent Deep Reinforcement Learning algorithms;
Application of automation control theory to recent Machine Learning backed control system;
Experience in algorithm design, analysis and comparison with respect to a real application

Vedi anche  thesis 16e - safe reinforcement learning.pdf 

Conoscenze richieste Students that are about to get their Master Degree in: mathematics, physics, computer science,
mathematical engineering, computer engineering, aerospace engineering, mechatronic engineering, mathematical
engineering, physics of complex systems.

Scadenza validita proposta 04/10/2020      PROPONI LA TUA CANDIDATURA

© Politecnico di Torino
Corso Duca degli Abruzzi, 24 - 10129 Torino, ITALY