Right-based Coordination for Environments with Changing

Complexity

Results in Traffic Simulation

Eduardo Alonso and Peter Kristofersson-Izmajlow

Department of Computer Science, City University London, London EC1V 0HB, U.K.

Keywords: Rights, Norms, Rational Choice Theory, Co-ordination, Efficiency, Traffic Simulation.

Abstract: In this paper we analyse three different approaches to multi-agent co-ordination, with an accent on the

concept of “right-based” agents as opposed to free-rider agents and normative agents. We claim that a

balance between unrestricted behaviour and regulatory systems would make collections of agents perform

more efficiently, particularly when the complexity and the dynamicity of the environment increases. We

present preliminary results on a set of experiments using a traffic simulator.

1 INTRODUCTION

How a collection of agents gets coordinated is the

object of study in various disciplines including

organisation theory, political science, social

psychology, anthropology, law, sociology, and

recently computing. Indeed, to ensure that agents

function in a social environment some type of

coordination mechanism is required. Such

mechanisms allow all agents to co-exist, and in some

cases to cooperate, so that they can perform their

tasks without clashes with other agents. More

specifically, in the age of the Internet and of cloud

computing, it is paramount that we develop

coordination mechanisms so that systems of

software agents, the so-called Multi-Agent Systems

(MAS), work efficiently. There are two main

schools of coordination in MAS, one that extends

shamelessly Rational Choice Theory (RCT) from

single-agent scenarios to MAS, and a

complementary one based on social norms, which

directly addresses multi-agent interactions.

The dilemma is that the agents either do

whatever they feel like or are constrained by the

designer and thus lack the freedom to make their

own choices. As an answer to this problem Alonso

(2004a, 2004b) proposed a “right-based”

coordination mechanism that allows agents to reason

and make decisions, but that implies enforcement of

rules at the same time. A middle-way between

freedom and norms, the idea of “right” has been

modelled formally as a system of axioms in (Alonso,

2004b).

In the next section the notion of “right” in MAS

is explained in some detail. We shall then present a

series of experiments we carried out with a traffic

simulator to test the relative efficiency of “right-

based” agents with regards to RCT agents and

normative agents. We shall finish with an analysis of

the results and some conclusions.

2 RIGHTS

Roughly stated, a right is considered as a set of

restrictions on the agents' activities that allow them

enough freedom, but at the same time constrain their

behaviour.

In their work on commitments Norman et al.,

(1998) introduced a “right” operator to help in

governing agent interaction and in the creation of

inter-agent agreements. According to this model, if

an agent wants to achieve its goals it will need to

seek “permissions” to perform all the necessary

actions. The notion of right we test is stronger in that

if an agent has the right to execute a set of actions

then

 It is permitted to perform any action in the set

(under certain constraints or obligations);

 The rest of the group is not allowed to execute

any action inhibiting the agent from exercising

its right, and

333

Alonso E. and Kristofersson-Izmajlow P..

Right-based Coordination for Environments with Changing Complexity - Results in Trafﬁc Simulation.

DOI: 10.5220/0004174703330336

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 333-336

ISBN: 978-989-8565-38-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)



 The group is obliged to prevent this inhibitory

action (and, eventually, to sanction the offender).

This third meta-right contextualizes the “right to

claim” (Castelfranchi, 1995) and plays a crucial role

in any social interaction: it means that any agent has

the right to ask for help if its counterpart in the

interaction (a short term deal or a long term

established pattern of behaviour) doesn't abide by

the terms of the “contract”.

It is not the purpose of this position paper to

evaluate thoroughly the validity of such theoretical

gains in co-ordination; rather, we have developed a

MAS simulator and tested the efficiency of right-

based agents against RCT agents and normative

agents in various settings of increasing complexity.

3 SIMULATION

We used as a core the VisSim traffic simulator at

http://www.vissim.de/index.php?id=1801, and

adapted it to include the agent architectures,

information provided by the system to the agents,

data saving, statistics and interaction between the

agents.

3.1 System Features

The system allows agents to perceive their

environment forward, backwards, and to the sides

back and forth. It gives full information about the

distance to other agents as long as the other agent is

on the same stretch of the road. It also gives their

speeds and direction. In the system the agent can

only see one agent ahead, meaning that if we have

three agent-cars driving in a row in front of us, we

will only see the closest one. The agents can change

their speed and position on the road (lane) in order

to go past other agents. Each car’s initial speed is set

randomly.

The system allows building and redefining roads

and junctions, defining the number of lanes in each

direction and the type of junction and the traffic-

light rules. It also enables defining the rate of new

incoming agents, where new agents enter the system

every n time steps (one car every n time units, 1/n),

and are removed from the system when crashed

(after 10 time units) or when they reach the end of

the lane. The entry per time unit is connected to each

lane.

3.2 Experiment Parameters

All the experiment results are based on 100 time

steps, where the data for each 10 steps is averaged.

The basic scenario upon which complexity builds is

a junction that cars approach from the four cardinal

directions. Lights regulating the traffic may be red,

yellow or green.

In total we have run 8 experiment scenarios, 4

for a single lane and 4 for double lanes. In each case,

the scenarios differed according to how often a new

car entered into the system: every 50, 100, 300 and

500 time steps, respectively (1)-(4) in the Results

tables. That is, we used two parameters to increase

the complexity of the scenario, namely, the number

of lanes in each direction and the rate of incoming

agents.

The efficiency of the three agent architectures

(RCT, right-based, and normative) in the different

scenarios was assessed against the following values:

(A) Number of cars that entered the junction;

(B) Number of cars that exited the junction;

(D) Average speed in the junction;

(E) Average time spent in the junction.

Obviously, the rates of entries and exits are not

informative in themselves, rather they relate directly

to the speed averages, the time spent by the cars in

the system and the number of crashes. The time

spent in the system depends in turn on the other two

factors –on which we focus the analysis of results in

section 5.

Before presenting the results, we describe how

the agents were represented –taking into account that

a utility function that rewards speed and punishes

crashes underlies the RCT agents’ architecture, as it

does the right-based architecture when rights allow

it.

3.3 RCT Architecture

The free agent architecture is based exclusively on

its perceptions of what is in front of the agent. The

agent will always try to find the best possible way to

get to its selected target exit from a junction. In

pursuing this goal the agents are free to do whatever

they want.

3.4 Normative Architecture

The normative architecture uses traffic lights to

manage the flow. What the agent does depends on

the light in the junction. Only one light will be green

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

334



at any time. The norms used were (assuming the

English traffic code):

 When the light is green go;

 When the light is yellow stop;

 When the light is red stop;

 If more than one lane, you can only go straight

ahead and left from the left lane;

 If more than one lane, you can only go straight

ahead and right from the right lane;

 If while being in the junction someone gets in

front to you, stop.

3.5 Right-based Architecture

The purpose of this architecture is to create a

coordination structure that depending on the

situation can be either very strict or a very lax one.

The rights used were:

 The right to not being obstructed while in

junction;

 The right to enter the junction if the agent’s light

is green;

 The right to enter the junction if junction is

empty;

 The right to do drive wherever the agent wants.

4 RESULTS

We are presenting the results as a 4x4 matrix

according to how long it takes for a new vehicle to

enter the system (1-4 above) and the different

assessment criteria (A)-(E).

Table 1: Single lane junction entry for a normative agent.

(1) (2) (3) (4)

(A) 799 1331 3236 3236

(B) 799 1331 3235 3235

(D) 5 9 22 22

(E) 12 13 13 13

Table 2: Single lane junction entry for a right-based agent.

(1) (2) (3) (4)

(A) 798 1332 3293 3297

(B) 798 1331 3293 3296

(D) 6 9 22 22

(E) 12 13 13 13

Table 3: Single lane junction entry for a RCT agent.

(1) (2) (3) (4)

(A) 737 1154 2890 3975

(B) 707 1099 2591 3309

(D) 4 8 18 19

(E) 13 13 13 13

We repeat the experiments with double lane roads,

for each agent architecture, RCT agents, normative

agents and right-based agents.

Table 4: Double lane junction entry for a normative agent.

(1) (2) (3) (4)

(A) 1598 2663 6405 6045

(B) 1585 2622 6371 6371

(D) 15 23 23 23

(E) 23 23 23 23

Table 5: Double lane junction entry for a right-based

agent.

(1) (2) (3) (4)

(A) 1596 2659 6564 6557

(B) 1592 2634 6525 6508

(D) 15 15 23 22

(E) 23 23 23 23

Table 6: Double lane junction entry for a RCT agent.

(1) (2) (3) (4)

(A) 1565 2615 7548 12408

(B) 1346 2185 5053 4064

(D) 12 17 20 3

(E) 24 23 24 34

5 CONCLUSIONS

In this position paper we have introduced the

concept of right and right-based as an alternative to

traditional approaches to MAS co-ordination. The

core of the paper focuses on a set of experiments

that test the hypothesis that right-based agents may

prove efficient in scenarios, a traffic network in our

study, of increasing difficulty. Needless to say, the

results we report are very preliminary both in terms

of the complexity of the environment and,

consequently, in the analyses of the results. Having

said that, there is some useful data that we can

Right-basedCoordinationforEnvironmentswithChangingComplexity-ResultsinTrafficSimulation

335



speculate on.

All the experiments show the same trends. As the

complexity of the environment increases, the worse

the RCT architecture performs. At the same time the

better the normative agents behave. In the

experiments one can clearly see that the norms

create a very well defined environment. An

environment which is easy to predict, as no agent

will behave in any other way then what the rules

prescribe. The problem with the free rider

architecture is that the free choice it has (to do

whatever it chooses) creates the complexity of the

environment. Thus the more agents with free choice

turn up the more complex the situation becomes.

The norms remove that problem but in doing so they

create a static environment where agents are bound

by the norms. What we gain in efficiency we lose in

autonomy. In the right-based case one can see a

different situation. Rights do not bind the agent in

the same way. Rights affect what is happening in the

system but only when the situation becomes too

complex to be handled without rights. Rights do not

have to be obeyed at any time, which is the case

with norms. In summary: RCT agents cannot cope

with complex situations. In such domains, agents

with rights behave the same way as “enslaved”

agents –yet they preserve their autonomy. Though

tempting, we are not extrapolating this preliminary

conclusion to real-life social scenarios.

REFERENCES

Alonso, E., 2004a. Rights for and Argumentation in Open-

Agent Systems. Artificial Intelligence Review 21, 3-

24.

Alonso, E., 2004b. A Formal Theory of Rights and

Argumentation in Open Normative Multi-Agent

Systems. In W. Zhang and V. Sorge, (Eds.),

Distributed Constraint Problem Solving and

Reasoning in Multi-Agent Systems, Frontiers in

Artificial Intelligence and Applications 112, pp. 153-

167, Amsterdam, The Netherlands: IOS Press.

Castelfranchi, C., 1995. Commitments: From individual

intentions to groups and organizations. In V. Lesser

and L. Gasser (Eds.), Proc. of the First International

Conference on Multi-Agent Systems (ICMAS-95), pp.

186–196, Cambridge, MA: The MIT Press.

Norman T., Sierra C., and Jennings N., 1998. Rights and

commitments in multi-agent agreements. In Y.

Demazeau (Ed.), Proc. of the Third International

Conference on Multi-Agent Systems (ICMAS-98), pp.

222–229, Washington, DC: IEEE Computer Society

Press.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

336