September 4th: The Triumph of Light!
The Fourth Lucia PhD School on "Artificial Intelligence and Robotics"

Lectures and Teachers

Semantic mapping: overview and benchmarking: A Tutorial (Monday and Wednesday)

Semantic mapping is the process that creates a representation of the environment suitable for the implementation of complex tasks and behaviors that require both geometric and semantic knowledge. The tutorial will present an overview of the approaches to Semantic Mapping, and then, it will focus on the interactive approach developed at RoCoCo laboratory. Subsequently, a novel system for semantic mapping (that will be used in the practicals) will be introduced together with a methodology for systematic benchmarking of semantic mapping approaches. The key distinguishing elements of this methodology are the following: (i) it allows for the evaluation of both metric and semantic information; (ii) it takes into account the navigation of the robot in the environment; (iii) it is accompanied by a simulation environment, that will be used for the practical activity. In the second half of the tutorial, students will be using the system for semantic mapping within the simulation environment. This activity will take the form of a competition, with the aim of providing the "best" semantic map built with the given tools. The tutorial will end with a discussion on the limitations of the proposed approach to semantic mapping, and on the factors that influence the performance of a semantic mapping system.

Prerequisites: Students are expected to have gone thrugh the reading assignments and software to be installed prior to the event (see the private area section here).

Teachers

Photo Daniele Nardi is Full Professor at Sapienza University of Rome, where he was employed since 1986. His current research interests are in the field of artificial intelligence and cognitive robotics (h-index 45 [Google scholar]). He is Trustee of RoboCup and past president of the RoboCup Federation, participating in sevral RoboCup leagues since1998. In 2006 he was the promoter of the RoboCup@Home league. From 2005-2008, he served as the co-chair of IEEE Technical Committee on Safety, Security and Rescue Robotics. Daniele Nardi received the "IJCAI-91 Publisher's Prize" and the prize "Intelligenza Artificiale 1993" from the Associazione Italiana per l'Intelligenza Artificiale (AI*IA). Currently, he is principal investigator of RoCKEU2 (FP7 CA), FLUORISH (H2020).

Contact: www.dis.uniroma1.it/~nardi

Photo Federico Nardi is PhD student at Sapienza University of Rome, where he graduated in 2014. His current research interests are in the field of Autonomous Navigation of wheeled robots and Semantic Mapping. During his master thesis he participated to the international research project TRADR (H2020). He's currently a member of the "Socialis robot PopulusQue Romanus et Lindensis" (SPQReL) team with which has participated to the European Robotics League Service Robots and will participate to the RoboCup@Home Social Standard Platform League 2017.

Contact: https://www.dis.uniroma1.it/~dottoratoii/students/306

 

Photo Roberto Capobianco is a Post-Doc Researcher at Sapienza University of Rome and Research Scientist at Cogitai, Inc. His current research interests span from autonomous and interactive robot learning, reinforcement learning, knowledge representation and semantic mapping. Roberto received his Ph.D. in Computer Engineering from Sapienza University of Rome in 2017, and he was a visiting Research Scholar at the Robotics Institute of the Carnegie Mellon University in 2015. He was awarded a research starting grant from Sapienza University of Rome in 2015, and a Robotics Fellowship in 2016 from AAAI.

Contact: http://robertocapobianco.com

 

Probabilistic Planning with Formal Guarantees for Mobile Robots: A Tutorial (Teusday)

In this tutorial, we will discuss two requirements for robot applications that are to be deployed for long periods of time in everyday environments: reasoning about the inherent environmental uncertainty emerging both from the robots noisy sensor and actuators and the presence of humans that change the state of the environment; and providing formal guarantees on the robots behaviour, ensuring safe and efficient behaviour.

In the first theoretical session, Nick will provide motivation for the need of the use of such techniques, and an overview of state-of-the-art approaches for planning for robots. In the second theoretical session, Bruno will focus on our work on the use of probabilistic model checking for the generation of policies for mobile robots with attached formal guarantees.

In the practical sessions, the students will be arranged in small groups and have the opportunity to apply the presented techniques in simulation. Groups that finish the simulation exercises will have the opportunity to apply the techniques on one of ISRs MBOTs, implementing behaviours that use our approach for planning the execution of tasks based on the MBOTs available skills.

Prerequisites: Students are expected to have gone thrugh the reading assignments and software to be installed prior to the event (see the private area section here).

Teachers

Photo Nick Hawes is a Reader in Autonomous Intelligent Robotics in the School of Computer Science at the University of Birmingham (BHAM). His expertise lies in the application of Aritifical Intelligence (AI) techniques (e.g. planning, spatial reasoning, probabilistic inference) to robots in order to generate intelligent, robust behaviour in challenging real world environments. He is the project coordinator of the £10M EU FP7 STRANDS project on long-term autonomy in everyday environments, PI on the CHIST-ERA project ALOOF, and has been PI or Co-I on other projects, to a total of £1.9M. Hawes has significant experience leading teams of researchers to successful outcomes in collaborative projects, and also in public engagement around robotics and AI.

Contact: http://www.cs.bham.ac.uk/~nah

Photo Bruno Lacerda received his Ph.D. in Electrical and Computing Engineering from the Instituto Superior Técnico, University of Lisbon, Portugal, in 2013. Currently, he is a research fellow at the School of Computer Science of the University of Birmingham, UK, working on the EU FP7 project STRANDS, where he is investigating the long term deployment of autonomous mobile robot systems. His research interests are in the use of formal approaches to specify and synthesise high-level controllers for robot systems, with attached meaningful formal guarantees. To achieve this goal, he his particularly interested in using formal verification, temporal logics, and planning under uncertainty approaches.

Contact: http://www.cs.bham.ac.uk/~lacerdab

 

Integration of AI and Robotics

Teachers

Photo Alessandro Saffiotti is full professor of Computer Science at ORU, where he heads the AASS Cognitive Robotic Systems Lab. He holds a MSc in Computer Science from the University of Pisa, Italy, and a PhD in Applied Science from the Universite Libre de Bruxelles, Belgium. His research interests encompass artificial intelligence, autonomous robotics, and technology for elderly people. He is the inventor of the notion of "Ecology of physically embedded intelligent systems", a new approach to include robotic technologies in everyday life, which has been applied in several projects. He has published more than 180 papers in international journals and conferences, his h-index is 40 in Google Scholar and 16 in the Web of Science. He has organized a large number of international events, and is the coordinator of the euRobotics topic group on "AI and Cognition in Robotics". He has participated in 14 EU projects, several EU networks, and many national projects. He is on the editorial board of Artificial Intelligence and of the International Journal on Social Robotics. He is a member of AAAI, a senior member of IEEE, and an EurAI fellow.

Contact: http://aass.oru.se/~asaffio

 

Multirobot Cooperative Perception (Thursday)

In this mini-tutorial, I will discuss how multiple robots with on-board vision sensors can cooperate to improve the perception of themselves (self-localization) as well as their surroundings (object tracking, mapping, etc.). In general, cooperative perception (CP) methods in multirobot systems increase overall state estimation accuracy, provide robustness to individual sensor failures, allow greater area coverage and, most importantly, enable better coordination among robots engaged in a cooperative task. In this tutorial, I will focus on two different classes of methods for CP. The first employs a particle filter (PF) and allows online estimation, while the other uses a pose graph optimization technique and is usually suited for offline estimation. I will discuss the advantages and drawbacks of both classes. Subsequently, I will describe a hybrid approach that attempts to combine the advantages from both classes, while being applicable in realtime. Finally, I will present some results on mid-size omnidirectional robots playing soccer as well as multi-rotor UAVs.

The practical part of this mini-tutorial will be designed as a 45 minute hackathon. The problem statement will be a variant of cooperate localization and target tracking. The exact statement will be given on the spot. As a prerequisite, students will need a few ROS packages and gazebo installed in their computer. One of the ROS packages (to be disclosed later), will serve as a coding point where the students would simply need to write snippets of code as a part of their 45-minute hackathon. Familiarity with c++, ROS and g2o will be required.

Prerequisites: Students are expected to have gone thrugh the reading assignments and software to be installed prior to the event (see the private area section here).

Teacher

Photo Aamir Ahmad is currently a research scientist at the Max Planck Institute for Intelligent Systems in Tübingen, Germany, where he is leading the multi-UAV aerial outdoor motion capture project. Up to August '16 he was a Marie-Curie postdoctoral fellow at the Max Planck Institute for biological Cybernetics in Tübingen, Germany. He received his Ph.D. degree (with merit and European Doctorate) in Electrical and Computer Engineering from Instituto Superior Técnico (IST), University of Lisbon, Portugal in April 2013. He received his Bachelors degree, B-Tech.(with Honors), in Civil Engineering from the Indian Institute of Technology (IIT) Kharagpur, India in July 2008. From April 2013 to August 2014 he was a postdoctoral researcher at the Institute for Systems and Robotics (ISR), IST, Lisbon. His main research focuses on the fields of Bayesian sensor fusion, multi-robot systems, robot localization, object tracking, cooperative methods for tracking and localization and 3D object detection. He was a member of the ISR scientific board and a program committee member of the international conference on autonomous agents and multiagent systems (AAMAS) 2014, 2016 and the European Conference on Mobile Robotics (ECMR) 2013. He has served as a reviewer of various prestigious conferences and journals, e.g., ICRA, IROS, ECMR, Robotics and Autonomous Systems Journal, Autonomous Robots and International Journal of Social Robotics.

Contact: https://ps.is.tuebingen.mpg.de/person/aahmad

 

Planning as an Optimal Control Problem (Thursday)

In this lecture, I will attempt to give a holistic view on the historically separated subjects of robot motion planning and control. In particular, viewing motion behavior generation as an Optimal Control Problem (OCP) allows for a unified formulation that is uncluttered by a-priori domain assumptions and simplified solution strategies. As an underpinning example, the problem of mobile robot fleet task assignment, motion planning and control will be used. Treating it in a holistic manner removes the necessity of, e. g., spatio-temporal reasoning about navigation along paths planned for each robot in isolation (i. e., the coordination problem). Essentially, optimal robot motion behavior generation can be seen as finding a policy as the solution of a highdimensional Markov Decision Process (MDP). However, in practice, the global solution of such a MDP is intractable. Therefore, this lecture will focus on locally optimal open-loop solution strategies. Specifically, we will discuss numerical trajectory optimization methods, where the underlying OCP is transcribed to a Nonlinear Programming Problem (NLP) which then is solved by Newton-type methods. Fundamental topics treated in this lecture include the Karush-Kuhn-Tucker (KKT) first-order optimality conditions, as well as direct trajectory optimization methods based on single-shooting, multiple shooting and collocation.

Prerequisites: Students are expected to have gone thrugh the reading assignments prior to the event (see the private area section here).

Teacher

Photo Robert Krug received his diploma degree (MSc) in Mechatronics in Mechanical Engineering from the University of Technology in Graz, Austria in 2009 and a PhD in Control Theory from Örebro, University, in Örebro, Sweden in 2014. He currently holds a shared post-doc position at the Applied Autonomous Sensor System (AASS) Research Center, Örebro, Sweden and the Robotics, Perception and Learning lab, KTH Stockholm, Sweden. His main research focus lies on simultaneous motion planning and control for autonomous robotic systems. Here, applications of interest range from grasping and manipulation to mobile robot navigation.

Contact: http://www.aass.oru.se/Research/Learning/rtkg

 

Learning by Self-Exploration (Thursday)

The use of robots in unstructured environments demands for methods to adapt to novel situations and environments. In this talk we will describe recent methods for robot learning complex models and skills from its own actions. First, we introduce the concept of object affordances and how to learn them. Object affordances contain information about the relations between objects, robot actions and effects of the actions, and provide rich models for planning, recognition and inference. Then, we will introduce Bayesian Optimization, a recent technique for sample efficient global optimization of expensive cost functions. Because sampling from the environment can be a costly process (in terms of energy, risk of damage and time) the exploration process must be efficient. Bayesian Optimization actively controls the robots learning process and sample from the environment the information that maximizes its experience. While classical reinforcement learning techniques reason about the expected value of the reward, Bayesian Optimization imposes a prior on the distribution of the function to optimize, thus explicitly representing uncertainty. The robot strategy will try to optimize the function that represents its task by looking not only at the regions where the functions expected value is high, but also for promising underexplored regions.

Prerequisites: Students are expected to have gone thrugh the reading assignments prior to the event (see the private area section here).

Teacher

Photo Alexandre Bernardino is an Associate Professor at the Dept. of Electrical and Computer Engineering of IST-Lisboa and Senior Researcher at the Computer and Robot Vision Laboratory of the Institute for Systems and Robotics of IST-Lisboa. He has participated in several national and international research projects as principal investigator and technical manager. He published more than one hundred research papers on top journals and peer-reviewed conferences in the field of robotics, vision and cognitive systems. He is associate editor of the journal Frontiers in Robotics and AI and of major robotics conferences (ICRA, IROS). He is the chair or the IEEE Portugal RAS Chapter. His main research interests focus on the application of computer vision, machine learning, cognitive science and control theory to advanced robotics and automation systems.

Contact: http://users.isr.ist.utl.pt/~alex

 

Can Robot Competitions Evaluate the Performance of Artificially Intelligent Robots? (Thursday)

Evaluating the performance of industrial robots through metrics like repeatability and accuracy has been a natural choice since decades. But with the growing development of all kinds of service robots, where Artificial Intelligence (AI) and Robotics interplay, the question of how to evaluate their performance has become crucial. AI evaluation is a challenging task which has been performed by the competition paradigm since its early years (e.g., Turing test, chess playing). In the last two decades, the same has happened in Robotics (e.g., the RoboCup 2050 challenge, DARPA Challenge, etc.). However, not all robot competitions can be considered proper instruments for repeatable experimental evaluation and benchmarking. In this talk we will discuss what does it mean to set up a benchmarking competition, i.e., a competition aimed at artificially intelligent robot benchmarking, and to what extent it is possible to evaluate robot functionalities, as well as complete robot systems. We will start by presenting a diversity of AI robot competitions that were or still are active in the last twenty years, and then focus on those which have significant potential for benchmarking. To establish how the performance of each functionality impacts on the performance of the complete system is still an open question. Some proposals to tackle it will be discussed in the realm of current EU funded projects and the selected robot competitions.

Prerequisites: Students are expected to have gone thrugh the reading assignments prior to the event (see the private area section here).

Teachers

Photo Matteo Matteucci (PhD in Computer Engineering and Automation, 2003, Politecnico di Milano) is an Associate Professor at POLIMI. He got a Master of Science in Knowledge Discovery and Data Mining at Carnegie Mellon University (Pittsburgh, PA), and a PhD in Computer Engineering and Automation at POLIMI. He has published more than 30 (peer-reviewed) papers on international journals and more than 100 (peer-reviewed) contributions to international conferences and book chapters. He is part of the Program Committee of several conferences on Artificial Intelligence and Robotics, he is in the Technical Committee of Intelligent Autonomous Vehicles of the International Federation of Automatic Control, and he serves as reviewer for international journals and main conferences in his field of expertise. He is deeply involved in the field of Robot Benchmarking; he participated as benchmarking expert in the FP7 EU funded RoSta Project; he has been an active participant to the Special Interest Group on Good Experimental Methodologies and Benchmarking of EURON, and he is one of the co-authors of the Review guidelines produced by the Special Interest Group; he has been one of the proposers of the euRobotics "Topic group on Benchmarking and Competitions" and currently member of the euRobotics Topic Group on "Experiment Replication, Benchmarking, Challenges and Competitions". He has been the Coordinator of the European project RAWSEEDS (2006-2009, http://www.rawseeds.org) a Specific Support Action in the FP6 for the development of a benchmarking toolkit for multi-sensor SLAM algorithms. He has been the National Scientific Coordinator (Principal Investigator) of the ROAMFREE project (2009-2013, http://roamfree.dei.polimi.it) for the development of method for the robust estimation of robot odometry by sensor fusion funded by the Italian Ministry for the University and the Research (MIUR) under the PRIN 2009 program. He has been the Principal Investigator for Politecnico di Milano (Partner) of the FP7 project RoCKIn (2013-2015, http://www.rockinrobotchallenge.eu/) for the design and execution of two international competitions for the benchmarking of autonomous robots in the home environment (RoCKIn@Home) and at work (RoCKIn@Work). He is currently the Principal Investigator for Politecnico di Milano in the H2020 RockEU2 (2016-2018) Coordinated Action within the workpackages devoted to the development of the European Robotics League and the establishment of distributed benchmarking competitions in Europe.

Contact: http://chrome.ws.dei.polimi.it

Photo Pedro U. Lima (Ph.D., Associate Professor) received the Licenciatura (5 years) and M.Sc degrees in Electrical and Computer Engineering from IST in 1984 and 1989, respectively, and the Ph.D. (1994) in Electrical Engineering from the Rensselaer Polytechnic Institute, NY, USA. Currently, he is a Professor at IST, Universidade de Lisboa, and a researcher of the Institute for Systems and Robotics, where he is the coordinator of the Intelligent Robots and Systems group. He is the co-author of two books, and Associate Editor of the Elsevier's Journal of Robotics and Autonomous Systems. His research interests lie in the areas of discrete event models of robot tasks and planning under uncertainty, with applications to networked robot systems. Pedro Lima is a Trustee of the RoboCup Federation (2003-2012, 2016-), and was the General Chair of RoboCup2004, held in Lisbon. He was President and founding member of the Portuguese Robotics Society, was National Delegate to EU and ESA Space Robotics programs and was awarded a 6-month Chair of Excellence at the Universidad Carlos III de Madrid, Spain in 2010. He was also the Coordinator of the FP7 Coordination Action RoCKIn and member of the Advisory Board of the Mohamed Bin Zayed International Robotics Challenge (MBZIRC), whose first edition took place in the United Arab Emirates in March 2017.

Contact: http://users.isr.ist.utl.pt/~pal

 

Building Robots for Social Interaction (Friday)

Social Robotics is a rapidly expanding field of research, but long-term results in real-world environments are still limited. In recent years, IDMind has been developing robots for different social environments. Robots to play specific social roles, interact with humans under tight constraints and coping with the uncertainty common in social environments. The presentation will focus the constraints involved in the design and operation of our social robots, the details related with their development to accommodate scientific goals while satisfying different environmental and technical constraints.

Prerequisites: Students are expected to have gone thrugh the reading assignments prior to the event (see the private area section here).

Teachers

Photo Paulo Alvito, IDMind's CEO, CTO and co-founder. Received the Licenciatura (5 year degree) in Electrical and Computer Engineering from Instituto Superior Técnico, Tech. Univ. of Lisbon, Portugal, in 1995. Lectured Control, Robotics and Industrial Instrumentation at the School of Technology from the Polytech. Inst. of Setúbal, from 1997 to 2007. Previous R&D activities include: Robotic Vision applied to Mobile Robotic Navigation at the Milan Polytech. Univ. - Italy, under the European Research Network (ERNET); implementation of control architectures applied to mobile robots at the Intelligent Systems Lab of the Institute for Systems and Robotics (ISR/IST), Lisbon. Managing IDMind's activities in FP6, FP7 and H2020 research projects. Representing IDMind at the euRobotics AISBL and Lisboa Robotics cluster.

Contact: http://www.idmind.pt

Social Robots and the Societal Applications: An industrial perspective at the cross-section of AI, Robotics and the Society (Friday)

We are evolving, so as our society, lifestyle and the needs. AI has been with us for decades, and now embodied in robots, penetrating more in our day-to-day life. All these are converging towards creating a smarter eco-system of living, where robots will coexist with us in harmony, for a smarter, healthier, safer and happier life. Social Intelligence of such consumer robots, enable the robots to behave in socially expected and accepted manners, will be the key technology and the next big R&D challenge. The talk will reinforce that robots have a range of potential societal applications, and that as a robotics industry, SoftBank Robotics R&D and Innovation is around the centrality of wellbeing of people. The first part of the talk will illustrate some of the use cases and the market analysis. The second part will present the feedback and needs from the real users, highlighting some of the immediate R&D challenges from industrial perspective. The talk will conclude with some open and grand challenges ahead, including social and ethical issues. Through this the young graduates will know the must/should have skills to be the part of this next generation of robotics revolution.

Photo Amit Kumar Pandey is Head Principal Scientist (Chief Scientist), also serving as the scientific coordinator (R&D) of its various collaborative projects. Earlier for 6 years he worked as researcher in Robotics and AI at LAAS-CNRS (French National Center for Scientific Research), Toulouse, France. His Ph.D. thesis in Robotics (title: Towards Socially Intelligent Robots in Human Centered Environment), is the second prize winner (tie) of the prestigious Georges Giralt Award for the best Ph.D. Thesis in Robotics in Europe, awarded by euRobotics (the European Union Robotics Community). His current research interest includes Socially Intelligent Robots, Human Robot Interaction (HRI), Robot's Cognitive Architecture and Lifelong Learning. On these aspects, he has been actively contributing as principal investigator, researcher, and industrial scientific coordinator in various national and European Union (EU) projects, as well as involved in their design and proposal. Among other responsibilities, he is the founding coordinator of Socially Intelligent Robots and Societal Applications (SIRo-SA) Topic Group (TG) of euRobotics, and an active contributor in the Multi- Annual Roadmap (MAR) and Strategic Research Agenda (SRA) of euRobotics, which aim to shape the future of robotics in Europe in collaboration with European Commission (EC) through PPP SPARC (the largest civilian-funded robotics innovation programme in the world). He is also the recipient of Pravashi Bihari Samman Puruskar 2014 (Non Residential Bihari Honour Award), for Science, Technology and Education, one of the highest level civilian honors, awarded by the state of Bihar, India.

Contact: http://www.amitkpandey.com

 

 

[ School Home ]

------------------
Website hosted by ISR/IST
Last updated on July 10, 2017