Research
Writings
Teaching
Personal
Microb1 (1994-1997)

Under construction - Last update : Sun, Jul 2, 2000

Table of contents

Description

The aim of the MICROB project was to investigate collective phenomena of organisation in societies of robots. It has begun as a joint project between the LIP6 and the LRP (Laboratoire de Robotique de Paris) . We decided in 1994 to use the game of soccer as a testbed, well before the hype that has accompanied the RoboCup competitions. The project relied on the use of a simple testbed, with semi-autonomous robots playing against human players (which were remotely controlling cars), which has enabled us to conduct experiments focused on the organisational features of the teams. The first instantiation of this project has taken place in a proprietary environment, before being involved within the RoboCup and MiroSot competitions of soccer-playing robots, both on simulated and real playing fields. The simulation experiments have been conducted on both a proprietary simulator and the RoboCup official simulator, SoccerServer, while the real ones have relied on two different types of robots.

1. 1994-1996

Our purpose with Microb was to physically (through real experiments) demonstrate that a set of robots is able to achieve an "intelligent" behaviour without being provided with the general solution of the problem it is supposed to solve. To put it in another words, we were interested in the emergence of self-organized collective phenomenons in societies of robots, where each member is provided with minimal individual capacitites, in terms of communication, interaction or cooperation.

Microb was a collaboration between the Miriad team of the LIP6 (at this time called Laforia) and the Laboratoire de Robotique de Paris (LRP); the latter was in charge of building the physical platform (the robots, their playground, etc.) , while the task of the former was to build a multi-agent system for controlling the robots and implement a simulator of the platform. Microb has been initiated, and is co-managed, by Alexis Drogoul (LAFORIA) and Dominique Duhaut (LRP).

The first application we have been working on to be implemented using Microb, is the programming of a reduced team of robots able to play a soccer-like game (a simplified version of the game) against a human team that remotely controls other vehicles. The number of robots simultaneously playing in the same team was limited to four. The rules were as follows: each team has its own goal, and contends with the other one for a ball, in order to shoot goals. The game takes place in an environment bounded by a wall (on which the ball can bounce). In that way, we do not take the penalties usually applied to a real soccer game (throw-ins, corners,..) into account.

The Microb project included a computer simulation platform that allows the user to create, test and modify the behaviors of the robots. The simulator (named SIEME) has been designed by Laurent Magnin (now at the CRIM). In a simulated environment defined using the Sieme syntax, the simulated robots have been provided with a set of behaviors, in order to allow them to (efficiently) cooperate within a "soccer" team .

The environment defined in the Sieme Simulator contains:The description of the system entities: the bounds surrounding the playground, the ball, the goals as well as the "physical" external appearance of the cars. The rules of interaction between these entities (like, for example, the bounce of the ball on the bounds) have also been implemented (using Sieme specific rules). Of course, the environment modelled here appears to be simplified, compared to the real one, but it has allowed us to implement and test a lot of behaviors, intended to be directly usable with the real robots.


Smalltalk / Sieme Browser

The robots have been provided with four basic behaviors (we have used them to compose all the -possibly more complex- sequences of behaviors they play with, using the Cassiopeia multi-agent design methodology):

In the shooting behvior, the car tries to place itself correctly in order to shoot the ball towards the opponents' goal. In practice, it follows a field of rotational vectors centered on the ball.

In order to distribute, within a team, the basic behaviors among the players, cars enter into contracts with the other ones. In fact, only one car can be the "shooter" at a given time. It is its duty to decide whether to give up or not (allowing another car to shoot), and with which robot to play (who is going to be the blocker, etc.). This kind of collaboration is implemented by a kind of contract-passing between the agents that drive the robots. Agents outside this "contract-net" automatically become defenders.

In order to reduce the communication between the agents, we have also implemented agents that possess (and reason about) a very simple model of the other robots. Whenever they have to take a decision, they enter into contracts with these virtual agents, as if they really were contract-based robots, and they take the role that they would have had in the previous case. Such cars can be integrated into heterogeneous teams (with purely reactive agents, or human-driven ones, ...). In these cases, they continuously interpret the actions of their partners and try to adapt (without communicating with them). See, for example, an experiment showing one of these agents collaborating with a human player.


Collaboration between a human-driven car and a model-based car

2. Physical Robots

The "real" experiments with "real" robots have made use of two different architectures. The first one, built after remote-controlled toy cars provided by the toy manufacturer TOMY (see picture) never entered a competition, but was mainly used for testing different pieces of the global system (i.e., the vision system, the remote controllers, etc.). That said, these small trucks were a lot sexier than today's RoboCup robots...

The second hardware architecture on which Microb1 was tested has been entirely built by the LRP and entered the first MiroSot competition, in June 1995 in Korea.

back to top

Software

Some of the softwares available can be found on Laurent Magnin's page.

back to top

People

List of the people that have worked or still work on this project :

back to top

Supports

This project has been supported by the CNRS and the University of Paris 6. TOMY (toy maufacturer) has provided us with the first set of vehicles.

back to top

Related papers

You might be interested to take a look at these papers for further information :

back to top

Related links

You might be interested to take a look at these links for further information :

back to top