Keio 0 Lectures Introduction

The 21th Century COE Program at Keio University:“System Design Paradigm Shift from Intelligence to Life”A Series of Lectures on:Planning Cooperative Behaviors for Multi-robot SystemsbyEnrico PagelloPresident of the International IAS-SocietyIntroduction to the Lectures: 

The 21th Century COE Program at Keio University: “System Design Paradigm Shift from Intelligence to Life” A Series of Lectures on: Planning Cooperative Behaviors for Multi-robot Systems by Enrico Pagello President of the International IAS-Society Introduction to the Lectures Intelligent Autonomous System Laboratory (IAS-Lab) The University of Padua, Italy

Slide2: 

Motivation of our research work If we consider as a New application area: Introducing robots in people’s ordinary life we need to deal with Multi-robot Systems (MRS) But moving from a single-robot theory_and_practice to a multi-robot science_and_technology requires: Developing a cooperative distributed vision system Designing distributed real-time software tools Balancing social deliberation and reactivity Using emergent behaviors Understanding better what collective behaviors are

Slide3: 

Distributed Vision Systems (DVS) What is a DVS? a set of vision systems embedded in the environment and connected by a network Which are the DVS tasks? Real time wide area safety surveillance Home assistance for young and old people

Slide4: 

Integrating Vision Agents on a Robot Goal: from simple tasks to cooperative tasks! Cooperative tasks for Multi-robot Systems require powerful omnidirectional specialized sensors Introducing the concept of an Omnidirectional Vision Agent Enhancing Image-based Localisation approach Integrating a Monte Carlo Localisation method Introducing learning methods for calibration Sharing and distributing sensor information through the net Developing an Omnidirectional Distributed Vision System

Slide5: 

Introducing an Emergent Behavior Engineering Traditional AI failed for real-time systems, but offers explanation and allows control Pure reactivity systems allows using real-robots in real-world, but only for simple tasks. Hybrid systems are needed Using Stigmergy to make emerging collective behaviors Stigmergy = observing other’s work Two robots interact through a perceptual flow, so that the action of the former affects the behavior of the latter A Role Assignement schema is a basic mechanism to make emerging a collective behavior inside a MRS

We start our series of Lectures by considering the localization problem for mobile robots, a basic problem for developing autonomous MRS.We discuss the Monte-Carlo Localization (MCL) approach, a probabilistic method, in which the current location of the robot is modelled as a posterior distribution on the sensor data represented by a set of weighted particles. We use an omnidirectional vision sensor as a range finder (like a laser or a sonar) sensitive to colors transitions to detect the nearest obstacles. E. Menegatti, A. Pretto, and E. Pagello Testing omnidirectional vision-based Monte-Carlo Localization under occlusion. Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS-2004), September 2004, Sendai, Japan: 

We start our series of Lectures by considering the localization problem for mobile robots, a basic problem for developing autonomous MRS. We discuss the Monte-Carlo Localization (MCL) approach, a probabilistic method, in which the current location of the robot is modelled as a posterior distribution on the sensor data represented by a set of weighted particles. We use an omnidirectional vision sensor as a range finder (like a laser or a sonar) sensitive to colors transitions to detect the nearest obstacles. E. Menegatti, A. Pretto, and E. Pagello Testing omnidirectional vision-based Monte-Carlo Localization under occlusion. Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS-2004), September 2004, Sendai, Japan First Lecture, May 24, 2005 - A Monte-Carlo Localization Method applied to a Omnidirectional-Vision Range Finder

We discuss two issues in the area of processing sensor information problems. First, we present a way to merge an image-based localisation approach with the MCL. In an image-based approach, the robot tries to match the current view of the environment with the reference views stored in its the visual memory to calculate its position. In MCL, the posterior probability density of the robot’s pose is represented with a set of weighted samples. Our method combines the advantages of both, overcoming their limitations when used separately. To work in in environments with spatial periodicity (perceptual aliasing). we avoid to use a metrical map. We exploits the properties of the Fourier Transform of the omnidirectional images, and uses the similarity between images to weights the samples.E. Menegatti, M. Zoccarato, E. Pagello and H. Ishiguro. Image-Based Monte-Carlo Localisation with Omnidirectional Images. Robotics and Autonomous Systems, Vol. 48, No. 1 , 2004 : 

We discuss two issues in the area of processing sensor information problems. First, we present a way to merge an image-based localisation approach with the MCL. In an image-based approach, the robot tries to match the current view of the environment with the reference views stored in its the visual memory to calculate its position. In MCL, the posterior probability density of the robot’s pose is represented with a set of weighted samples. Our method combines the advantages of both, overcoming their limitations when used separately. To work in in environments with spatial periodicity (perceptual aliasing). we avoid to use a metrical map. We exploits the properties of the Fourier Transform of the omnidirectional images, and uses the similarity between images to weights the samples. E. Menegatti, M. Zoccarato, E. Pagello and H. Ishiguro. Image-Based Monte-Carlo Localisation with Omnidirectional Images. Robotics and Autonomous Systems, Vol. 48, No. 1 , 2004 Second Lecture, May 31, 2005 - First part. Using the Fast Fourier Transform in an Omnidirectional Distributed Vision System

Then we discuss the problem that traditional image-based localization methods do not work when the robot is moving in an environment whose appearance is changing in time. Thus, we propose a new approach that enables the system to work also in highly dynamic environments by using several omnidirectional cameras installed in the environment and an omnidirectional camera mounted on a mobile robot. The localization of the robot is achieved by comparing the current image grabbed by the robot with the images grabbed at the same time by the DVS. E. Menegatti, G. Gatto, and E. Pagello, Takashi Minato and Hiroshi Ishiguro Distributed vision system for robot localisation in indoor environment. Submitted to ECMR’05 Conference: 

Then we discuss the problem that traditional image-based localization methods do not work when the robot is moving in an environment whose appearance is changing in time. Thus, we propose a new approach that enables the system to work also in highly dynamic environments by using several omnidirectional cameras installed in the environment and an omnidirectional camera mounted on a mobile robot. The localization of the robot is achieved by comparing the current image grabbed by the robot with the images grabbed at the same time by the DVS. E. Menegatti, G. Gatto, and E. Pagello, Takashi Minato and Hiroshi Ishiguro Distributed vision system for robot localisation in indoor environment. Submitted to ECMR’05 Conference Second Lecture, May 31, 2005 - Second part. Using the Image-Based Localisation in a higly dynamic environment

Lectures continue by considering the problem of distributing the Knowledge through an Omnidirectional DVS (Distributed Vision System). To this aim, two experiments are presented. The first deals with an Omnidirectional DVS that learns how to navigate a service-robot in an office-like environment without any knowledge about the calibration of the cameras or the robot control law. One Vision Agent (VA) learns to control the robot with a SARSA(lambda) reinforcement learning technique, using the LEM strategy to speed-up learning. Once the Agent learnt the correct policy, it transfers its knowledge to another Agents. E. Menegatti, C. Simionato, S. Tonello, G. Cicirelli, A.Distante, H. Ishiguro, E. Pagello Knowledge Propagation in a Distributed Omnidirectional Vision System. Submitted to a Special Issue in Memory of Marco Somalvico, Int. Jour. of Intelligent and Fuzzy Systems: 

Lectures continue by considering the problem of distributing the Knowledge through an Omnidirectional DVS (Distributed Vision System). To this aim, two experiments are presented. The first deals with an Omnidirectional DVS that learns how to navigate a service-robot in an office-like environment without any knowledge about the calibration of the cameras or the robot control law. One Vision Agent (VA) learns to control the robot with a SARSA(lambda) reinforcement learning technique, using the LEM strategy to speed-up learning. Once the Agent learnt the correct policy, it transfers its knowledge to another Agents. E. Menegatti, C. Simionato, S. Tonello, G. Cicirelli, A.Distante, H. Ishiguro, E. Pagello Knowledge Propagation in a Distributed Omnidirectional Vision System. Submitted to a Special Issue in Memory of Marco Somalvico, Int. Jour. of Intelligent and Fuzzy Systems Third Lecture, June 7, 2005 - First part. Sensoring the environment by an Omnidirectional Distributed Vision System

Then, we consider how to cooperatively track and share the information about moving objects using a multi-robot team, when every robot of the team is fitted with a different omnidirectional vision system running at different frame rates. The information gathered from every robot is broadcast to all the other robots and every robot fuses its own measurements with the information received from the teammates, building its own “vision of the world”. The cooperation among the several vision sensors enhances the capabilities of each single vision sensor allowing a team of robot to track multiple objects E. Menegatti, A. Scarpa, D. Massarin, E. Ros, E. Pagello Omnidirectional Distributed Vision System for a Team of Heterogeneous Robots. Proc. of IEEE Workshop on Omnidirectional Vision (Omnivis'03), in conjunction with Computer Vision and Pattern Recognition (CVPR 2003) Madison, Wisconsin (USA) 2003: 

Then, we consider how to cooperatively track and share the information about moving objects using a multi-robot team, when every robot of the team is fitted with a different omnidirectional vision system running at different frame rates. The information gathered from every robot is broadcast to all the other robots and every robot fuses its own measurements with the information received from the teammates, building its own “vision of the world”. The cooperation among the several vision sensors enhances the capabilities of each single vision sensor allowing a team of robot to track multiple objects E. Menegatti, A. Scarpa, D. Massarin, E. Ros, E. Pagello Omnidirectional Distributed Vision System for a Team of Heterogeneous Robots. Proc. of IEEE Workshop on Omnidirectional Vision (Omnivis'03), in conjunction with Computer Vision and Pattern Recognition (CVPR 2003) Madison, Wisconsin (USA) 2003 Third Lecture, June 7, 2005 - Second part. Sharing the Distributed Knowledge in an Omnidirectional Distributed Vision System

Task Allocation Problem is a key issue in the MRS domain. The scientific community has presented several interesting solutions to this classical AI planning problem. Applying it to a team of robots means to switch to a dynamic role allocation approach. Gerkey and Mataric showed it is similar to task allocation problem for MRS in order to cooperatively achieve the goal, where a time-extended role concept replace that of a transient task. We discuss also how an hybrid architecture can take advantages from the interaction between the deliberative part and the reactive one and viceversa, to make a MRS to exhibit the assigned cooperative task. A. D’angelo, E. Menegatti, E. Pagello How a cooperative behaviour can emerge from a robot team. R. Alami, H. Asama, R. Chatila Eds., Proceedings of Int. Conf. on Distributed Autonomous Systems (DARS-2004), Toulouse (Fr), 23-25 June, pp. 71-80: 

Task Allocation Problem is a key issue in the MRS domain. The scientific community has presented several interesting solutions to this classical AI planning problem. Applying it to a team of robots means to switch to a dynamic role allocation approach. Gerkey and Mataric showed it is similar to task allocation problem for MRS in order to cooperatively achieve the goal, where a time-extended role concept replace that of a transient task. We discuss also how an hybrid architecture can take advantages from the interaction between the deliberative part and the reactive one and viceversa, to make a MRS to exhibit the assigned cooperative task. A. D’angelo, E. Menegatti, E. Pagello How a cooperative behaviour can emerge from a robot team. R. Alami, H. Asama, R. Chatila Eds., Proceedings of Int. Conf. on Distributed Autonomous Systems (DARS-2004), Toulouse (Fr), 23-25 June, pp. 71-80 Fourth Lecture, June 21, 2005 - Cooperation Issues for Multi-Agent Systems

Since the execution of any cooperative robotic tasks starts with a cooperative solution of some motion planning problems, we consider to the possibility of dealing with MRS’s high dimensional C-spaces by using a centralized approach based on efficient randomized motion-planning algorithms. We show how to construct a probabilistic roadmap, as the transition from subsequent free configurations for the MRS Then, we propose a distributed algorithm for the classical problem of motion planning for multi-robot systems. Our algorithm uses randomization and solution quality indexes to efficiently find a suitable solution. S. Carpin, E. Pagello Exploiting Multi-robot Geometry for Efficient Randomized Motion Planning. Proc. of the 7th Int. Conf. on Intelligent Autonomous Systems (IAS-7), LosAngeles, USA, March 2002S. Carpin, E. Pagello A Distributed Algorithm for Multi-robot Motion Planning. Proc. of the Fourth European Workshop on Advanced Mobile Robots (EUROBOT01), Lund (Sweden), September 2001, pp. 207-214: 

Since the execution of any cooperative robotic tasks starts with a cooperative solution of some motion planning problems, we consider to the possibility of dealing with MRS’s high dimensional C-spaces by using a centralized approach based on efficient randomized motion-planning algorithms. We show how to construct a probabilistic roadmap, as the transition from subsequent free configurations for the MRS Then, we propose a distributed algorithm for the classical problem of motion planning for multi-robot systems. Our algorithm uses randomization and solution quality indexes to efficiently find a suitable solution. S. Carpin, E. Pagello Exploiting Multi-robot Geometry for Efficient Randomized Motion Planning. Proc. of the 7th Int. Conf. on Intelligent Autonomous Systems (IAS-7), LosAngeles, USA, March 2002 S. Carpin, E. Pagello A Distributed Algorithm for Multi-robot Motion Planning. Proc. of the Fourth European Workshop on Advanced Mobile Robots (EUROBOT01), Lund (Sweden), September 2001, pp. 207-214 Fifth Lecture, June 28, 2005 - Motion Planning Strategies for Multi-Robot Systems

E. Pagello, A. D'Angelo, F. Montesello, F. Garelli, C. Ferrari. Cooperative behaviors in multi-robot systems through implicit communication. Robotics and Autonomous Systems, Vol. 29, No. 1, 1999, pp. 65-77S. Carpin, C. Ferrari, E. Pagello. Map Focus: a way to reconcile reactivity and deliberation in multi-robot systes. Robotics and Autonomous Systems. 2002, Vol. 41, pp. 245-255: 

E. Pagello, A. D'Angelo, F. Montesello, F. Garelli, C. Ferrari. Cooperative behaviors in multi-robot systems through implicit communication. Robotics and Autonomous Systems, Vol. 29, No. 1, 1999, pp. 65-77 S. Carpin, C. Ferrari, E. Pagello. Map Focus: a way to reconcile reactivity and deliberation in multi-robot systes. Robotics and Autonomous Systems. 2002, Vol. 41, pp. 245-255 Sixth Lecture, July 5, 2005 - First Part Stigmergy, Emerging Behaviors, and Hybrid Architectures We illustrate the Cooperation through Implicit Communication behavior-based approach used for developing The University of Padua Simulated Soccer Robot Team for RoboCup Simulation League. The configuration of the environment, namely the robots' relative positions depending on both the global task and the game dynamics, provides a source of implicit information about the robots' intention to be involved in collective actions, making them able to cooperate implicitly. The soccer-team performance can be tuned by triggering the arbitration module of any single robot to generate, as many as possible, suitable situations which hint to the team the action of scoring the goal. Some macroscopic parameters have been usefully introduced to evaluate the evolution of the whole multi-robot software system.

We investigate how stigmergic information allow each individual of a group of autonomous robots to take advantages from other individual behaviors. Without explicit communication, the collective behavior of a group of teammates can be forced only if the robot designer makes each robot to become aware of distinguishing configuration patterns in the environment.  We model the environment dynamics by considering some parameters that express the ability of each robot to perform its task. The proposed analysis is based on the roboticle model where sensor data and effector commands are treated as energy exchange between the robot and its environment, eventually populated by other robots.A. D'Angelo, J. Ota, E. Pagello How intelligent behavior can emerge from a group of roboticles moving around Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS-2003), Las Vegas (USA), October 2003 D’Angelo, E. Pagello Making Collective Behaviours to work through Implicit Communication Proc. of the 2005 IEEE Int. Conf. on Robotics and Automation (ICRA-2005), Barcelona (Spain), April 2005: 

We investigate how stigmergic information allow each individual of a group of autonomous robots to take advantages from other individual behaviors. Without explicit communication, the collective behavior of a group of teammates can be forced only if the robot designer makes each robot to become aware of distinguishing configuration patterns in the environment. We model the environment dynamics by considering some parameters that express the ability of each robot to perform its task. The proposed analysis is based on the roboticle model where sensor data and effector commands are treated as energy exchange between the robot and its environment, eventually populated by other robots. A. D'Angelo, J. Ota, E. Pagello How intelligent behavior can emerge from a group of roboticles moving around Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS-2003), Las Vegas (USA), October 2003 D’Angelo, E. Pagello Making Collective Behaviours to work through Implicit Communication Proc. of the 2005 IEEE Int. Conf. on Robotics and Automation (ICRA-2005), Barcelona (Spain), April 2005 Sixth Lecture, July 5, 2005 - Second part. Implicit Communication in a Dynamical System-based Framework

Slide15: 

Thanks for Your Attention! The Artisti Veneti Team www.dei.unipd.it/~robocup