RoboCup-99 Team Descriptions 0 NIL League, Team French-Team, pages 0–0 http://www.ep.liu.se/ea/cis/1999/NIL/2/

The French Robocup Team

French-Team

Kamel BOUCHEFRA, Vincent HUGEL, Patrick BONNIN, Pierre BLAZEVIC, Dominique DUHAUT

DD: 6 University PB: Versailles Saint Quentin En Yvelines University VH: Laboratoire de Paris KB, PB: Paris 13 University

Abstract. This paper presents the software components designed by the LRP team and which are intended to make the Sony Pet behave as powerful organized soccer players. These components comprise a locomo- tion module, a vision module and a strategy module. The article explains the personal background of the authors, this includes the experience from earlier projects. The key words of this work are the following : Pet Robots, auton- omy, perception, concurrency, strategy. The French RoboCup Team is composed of Dominique DUHAUT, Pierre BLAZEVIC, Patrick BONNIN, Vincent HUGEL and Kamel BOUCHEFRA. Pierre BLAZEVIC and Vincent HUGEL are both in charge of the part of the project. They are the roboticians of our team. Pierre and Vincent have been involved in many competitions and exhibitions, among which the last RoboCup held in july 1998 in Paris. Pierre BLAZEVIC is an Associate Professor at Versailles Saint Quentin En Yvelines University. Vincent HUGEL is a PhD student at LRP (Laboratoire de Paris). Patrick BONNIN is in charge of the vision system. He has been involved in the last RoboCup held in july 1998 in Paris. Patrick BONNIN is an Associate Professor at Paris 13 university. Kamel BOUCHEFRA is a new member in the team, he is in charge of the strategy level within the subject. Kamel is an Associate Professor at Paris 13 University. Dominique DUHAUT is at the origin of the participation of the French RoboCup Team in the Sony legged Project. Special acknowledgement is given to him from all the team members. Dominique DUHAUT is an Associate Professor at Paris 6 University.

1 Introduction

The design of autonomous robots, operating in real world environment, is the subject of research programs in the main industrialized countries. Inter- 1 est for this research field has grown specially during the last decade. This interest is due to various conditions : an advantageous attention of a large public, the development in circuits design that lead to powerful devices and processors at continuously cheaper prices, ... In this frame of mind, the Sony Pet Robots, described in paragraph 2, are powerful autonomous plat- forms, and an exceptional opportunity to consider various research aspects of great importance. These robots are expected to play soccer, according to an organization of three robots per team. The application determines an en- vironment which is the source of events usually uncertain, partially accurate and always unpredictable. Consequently, the targeted robots need the de- sign of a software level architecture that makes the robots able to fulfil real time constraints, and produce adequate actions, often reactive, sometimes deliberative and perhaps opportunistic in certain situations. The design of the targeted software architecture requires complementary skills. Thus our research group associates roboticians who deal with locomotion aspects (see paragraph 3), specialists in vision who focus on image processing aspects (see 4) and AI specialists who deal with strategy considerations (see 5).

2 Sony Pet Robot’s short description

The four legged Sony Pet Robots are equipped with adequate hardware and software that allow autonomous motion. The hardware equipment includes a supply battery and a camera which is the main sensor. The legs may also be used to kick a ball. The software part is based on Open-R, an API (Application Programming Interface) between the user level and the Operating System, it allows the control of the different devices according to programs written in C and/or C++.

3 Locomotion

The pet robot has been provided with an enhanced walking control sys- tem allowing it to move in any direction keeping balance using quasidy- namic gaits (regular symmetric gaits with duty factor 0.75). Varying the duty factor allows to change the robot’s velocity (in rotation, translation and turns), according to the situation. Also a self-recovering behavior has been implemented in case of fall down during competition using onboard accelerometers, [1; 2].

4 Vision

The vision system is the main sensor. Its aim is to detect, to identify and to spot the different elements of the observed scene during the play. Detection corresponds to the extraction of all connected components, the set of pixels in the color images that represent the objects observed with the camera. Identification is processed in two steps. The first step consists in finding the set of connected components representing an object within the scenery. The second step consists in associating each connected component with a symbolic label. The set of defined labels is the following : ball; beacon; own or opponent goal; partner or opponent player; edge of the 2 soccer field. Spotting beacons or goals means localizating them in terms of an horizontal angle (with respect to the head’s direction), and of a rough measurement of the distance between the head and the targeted object. The main characteristic of the vision system is the design of an image processing software package that allows object retrievals, processed at about 14 or 15 frames per second, according to queries issued from the strategy level (see 5).

5 Strategy

A model of agents is the basis for our design of the software level of the Sony Pet. Each robot is considered as an agent. The design of our agents’ model leads us to consider both aspects of algorithms design (each algorithm cor- responds to a particular process), and process control [3; 4]. The proposed process control is event-driven and is based on considering processes level of priority. The organization of soccer teams leads us to the elaboration of different agent-to-agent interactions schemes [5; 6; 7]. These interaction schemes are based on agent’s objective identification. The arising problem, in designing these interaction schemes, concerns conflicts identification be- tween different agents’ objectives. This difficulty is due to uncertainty and the inaccurate nature of data issued by the environment, making acquain- tance’s behavior interpretation difficult. The design of agent’s processes includes data uncertainty and inaccuracy management which is based on possibility theory.In the proposed schemes, agents that belong to a same team, cooperate if their objectives differ, they are concurrent when their objectives are the same. The agents are antagonists, if they belong to dif- ferent teams and if they share a common objective.

6 Conclusion

This paper presents a research in both fields of multiagents systems and robotics. The focus, from agent’s model design, is given to aspects related to the autonomy and interactions points. This is, in our opinion, a key element in order to make explicit agents’ behavior. From a vision system designer point of view, this research allowed the integration of the vision system on a real robot, leading the designer to take into account variable lighting conditions and images that lack stability.

7 Bibliography

[1] Vincent Hugel, Pierre Blazevic. Towards Efficient Implementation of Quadruped Gaits With Duty Factor of 0.75. IEEE International Conference On Robotics and Automation. Detroit, Michigan, USA. May 1999. [2] M. Fujita, S. Zrehen, H. Kitano. A Quadruped Robot for RoboCup Legged Robot Challenge. Proc of the second RoboCup Workshop, Paris, July 1998. 3

[3] Kamel Bouchefra, Roger Reynaud, Thierry. Maurin. Basis for Intelligent Interactions. IFAC Conference on Intelligent Systems (CIS’97). Belfort , mai 1997. [4] Kamel BOUCHEFRA, Patrick AUGE, Thierry MAURIN, Brigitte RO- ZOY, Roger REYNAUD. Multi-agents Based Architecture Specification and Verification. KAW Conference on Intelligent Systems (KAW’98). Banff , april 1998. [5] B. A. Huberman. The performance of cooperative processes. Physica D42. North Holland. Elsevier Science Publisher. 1990. [6] J. O. Kephart, T. Hogg, B. A. Huberman. Collective behavior of pre- dictive agents. Physica D42. North Holland. Elsevier Science Publisher. 1990. [7] P. J. Gmytrasiewicz. On reasoning about other agents. M. Wooldridge, J. P. M¨uller, M. Tambe (Ed’s). Intelligent Agents II. Agents Theories, Architectures, and Languauges. IJCAI’95 Workshop Proceedings. Springer Verlag. 1995.