Using BDI agents to improve driving modelling in a commuter scenario

By :-Prakhar Sharma(111271)

Sumit Gupta(111318)

Based on research paper by:

• Rosaldo J.F. Rossetti • Rafael H. Bordini • Ana L.C. Bazzan • Sergio Bampi • Ronghui Liu• Dirck Van Vliet

• All of them from a university in brazil.

Abstract

• Now a days multi-agent systems are widely used in the field of research.• Applied to varities of areas including social sciences, engineering, and

mathematical and physical theories.• Today we are going to address the complex task of modelling drivers

behaviour through this technique.• Extension to the DRACULA MODEL.

• This work is divided into two main parts:-

• 1. DRACULA framework and the extension proposed to support our agent-based traffic model.• 2. The reasoning mechanism of drivers modelled by means of a Beliefs,

Desires, and Intentions (BDI) architecture.

Introduction

• Traffic is increasing at a rapid rate.• Physical modification to the road infrastructure is no longer the best

alternative as very expensive to implement.• Why not improve the efficiency of the traffic control system?• There are some solutions but they cannot be said as effective ones• So we need alternative means to cope with the traffic and

transportation problem

Introduction

• First approach was to directly influence users behavioral patterns with the help of intelligent transportation systems (ITS).• Integrating all factors, both dynamic and static, which can somehow

affect the traffic flow, is central to ITS. • More recently, new-generation trafc network models have also

emerged from scratch• Multi-agent system (MAS) is a sub-field of the distributed artificial

intelligence (DAI), which has received an increasing interest in the last few years

ITS from a multi-agent system perspective

• The concept of a multi-agent system is a modelling approach devised to represent systems whose entities, coined agents, exhibit intelligence, autonomy, and some degree of interaction, both with one another and with the environment• An agent must be capable of perceiving facts through sensors and

acting upon the environment through effectors.• There are two kinds agent structures : 1. the reactive agents 2. the cognitive agents

• Reactive approach is totally based on a simple approach of mapping perceptions to actions• Cognitive approach is fully endowed with the reasoning capabilities.

• An agent structure sometimes has the characteristics of both types to some types.

Example of applied MASOn a normal working day, agent A is about to leave home for another home-work journey. Albeit A is familiar with the streets to his work, he decides to check the traffic conditions before starting the trip in order to avoid a possible traffic jam during the morning peak-hour. Thus, agent A logs onto the Internet and accesses a traffic information system website. Knowing where to get to and estimating the time he will need to perform the journey, A can plan his trip. Thus, he selects a course of actions that will result in reaching his place of work. He chooses a route to follow and a time to leave so that he can arrive by a desired arrival time. Once he has planned his trip, he can execute it. While A has not found any obstacle within the journey, he can keep executing his original plan. However, he has just found that certain road on his route is interrupted. As A is not able to drive through that road anymore, he has to reconsider his options and find another alternative route to get to his destination. Therefore, A abandons his original plan and starts executing a new one. In another point in time, A just realises that he has not got enough petrol to complete his trip. Albeit he knows he will probably be late, he priorities changing his plans once again in order to stop somewhere to get more petrol. After doing so, he is finally able to arrive at work.

Analysis of the above example

• autonomy--agent A could identify what his objectives were and which actions he needed to carry out to yield the expecting results; • social ability--as in the above example, agent A could ask for some help in

order to ease the execution of his actions, for instance, by contacting a service provider such as a traveler information centre.• reactivity--responding to traffic signals and breaking in order to avoid

colliding with others are some well known examples of reactive behavior. Albeit perceiving an interruption and adopt- ing another route could be modelled as reactions from the driver, deciding whether to divert may also imply more sophisticated inference mechanisms.

• adaptability--agent A is adaptable in the sense that he may reconsider his options and adopt another strategy in order to accomplish his goals, in the case the original plan becomes inadequate; • pro-activity--agent A also must be able to prioritize the execution of

an action to the detriment of his original plans, for instance, arriving later at work after adopting another route that is more convenient owing to some other reasons. • Therefore for capturing these features into the driver representation

we adopt a Beliefs, Desires, and Intentions(BDI) architecture.

An extension to the DRACULA framework

DRACULA framework

An agent based demand model

The specification language: AgentSpeak(L)

• AgentSpeak(L) is a language devised to bridge the gap between formal modelling and practical implementation as far as BDI agents are concerned.• It basically reduces the task of modelling intentional agents to

identifying base beliefs, goals and plans.

Note that both beliefs and goals are predicate symbols (as are actions, as seen later). A predicate symbol is a goal if it is preceded by the operators ‘!’ or ‘?’ (in def 2). Thus, an agent a may have the achievement goal of being in location Y in a future state, expressed by !location(a, Y ). It may also check what its present position is, by a test goal such as ?location(a, X),given its set of base beliefs (updated through perception of the environment)

Agents go through repeated cycles where they observe the environment and, based on their observations and goals, they execute certain actions that may change the state of the environment. This may influence the agents beliefs as well, which need to be revised.

A BDI driver modelThe main goal of this work is to present a framework such that it can be used to model and implement commuter scenarios. In our model, we shall consider that there exists a population of potential drivers who are able to perform journeys throughout the road network. Thus, drivers from the population, who have decided to make a trip on a certain day, constitute the demand for travel on that day. In our multi agent traffic model we consider two agents, namely the driver agent and the information system agent. So, our multi-agent traffic system is given by the tuple <N,A,ITS> where N is a set of road links, A is a set of driver agents, and ITS is a set of intelligent transportation systems.

As already mentioned, a driver agent is represented as a reasoning and intentional entity in the system. Its behavior is represented in terms of mental attitudes such as beliefs, desires, and intentions, which represent informational, motivational, and intentional states, respectively. Then, a driver cognitive structure is given by the tuple <E,B,P,I,A,SE,SO,S> as presented in Definition 6 for BDI agents.

A journey performed throughout the network starts at an origin and ends at destination, usually represented as traffic zones. Hence all the base belief predicates defined are as follows:-• Route(zoneorg,zonedest)• Taveltime([link1,link2,…..linkn]) where adjacent(linki,linki+1)• State(l,s)• Today(day)• TimeNow(time)• Purpose(day,p)• Destination(p,zonedest)• Location(l) where !Location(l) represents that the driver is willing to reach link l, whereas ?Location(l) check what is present location l with respect to destination.

In our commuter scenario, when planning a journey, the driver needs to make decisions about • the purpose for the trip• the destination it intends to reach • the desired arrival time• what time it needs to depart• route to its destination.

Thus, a trip is given by the tuple <p,zoneorg,zonedest,tarv,tdpt,r>.

Example of possible clauses in a set of base beliefs

The commuter scenarios

• Scenario 1In the first scenario, drivers are supposed to make every decision with respect to the trip before starting the journey. Afterwards, diversions are not allowed. We assume that no information is provided to individuals, neither before nor during the trip. Therefore, drivers make their decisions based solely on their set of base beliefs.

Plan 1

• Starting the day. Perceiving that a new day starts is an external stimulus, which is represented by the event today(day). The set of base beliefs is updated, then. When the driver perceives the beginning of another day it becomes committed to an activity purpose for that day.

01. +today(Day) : true02. <- ?purpose(Day, P);

03. ?destination(P, ZDst);

04. !location(ZDst).

• With respect to the route selection, we use the achievement goals• !anyRoute(zoneorg,zonedst,[links])• !bestRoute(zoneorg,zonedst;[links])• !usualRoute(zoneorg,zonedst,[links])• to represent the desire of a driver to choose any route, the best

route, or the usual route, respectively, among the route(zoneorg,zonedst,t,[links]) clauses in the set of its base beliefs.• The belief predicate tripRoute(zoneorg,zonedst,[links],strategy) is

used to represent the particular route chosen by the driver. Where strategy can be any, best, usual, system. Thus, whenever a route is selected, the set of base beliefs is updated and the belief tripRoute is added to represent the driver’s selection.

Plan 2a

• Changing location by selecting any route. In this case the driver opts for the strategy of getting any route, despite the cost.

01. +!location(ZDst) : location(ZOrg) & (not(ZDst==ZOrg))02. <- !anyRoute(ZOrg,ZDst,R);03. !chooseDepartureTime(R).

Plan 2b

• Changing location by selecting the usual route. The strategy adopted in this plan is to choose the usual route.

01. +!location(ZDst) : location(ZOrg) & (not(ZDst==ZOrg))02. <- !usualRoute(ZOrg,ZDst,R);03. !chooseDepartureTime(R).

Plan 2c

• Changing location by selecting the best route. In this plan, the driver attempts to find the best route among the route(zoneorg,zonedst,t,[links]) clauses the set of its base beliefs.

01. +!location(ZDst) : location(ZOrg)&(not(ZDst==ZOrg))02. <- !bestRoute(ZOrg,ZDst,R);03. !chooseDepartureTime(R).

Plan 2d

Getting to the intended location (end of the recursion). This plan identifies two situations.• The driver decides to stay at home so, the intended location is already

the present location.• The driver has moved throughout the chosen route and finally arrived

at the destination.

+!location(ZDst) : location(ZDst)<-true.

Plan 3

• Choosing any route. To confirm the selection, an update to the set of base beliefs occurs in line 2, and the trip route is added to the base beliefs.

01. +!anyRoute(ZOrg,ZDst,R) : route(ZOrg,ZDst,T,R)02. <- +tripRoute(ZOrg,ZDst,R,any).

Plan 4

• Choosing the usual route. In our commuter world model, we consider that the driver already has in its base beliefs a belief of having a preferred route, which is the usual route. In this case, the driver will keep such a route.

01. +!usualRoute(ZOrg,ZDst,R) : tripRoute(ZOrg,ZDst,R,usual)02. <-true.

Plan 5a

• The belief predicate expectedTravelTime(zoneorg,zonedst,time) is used to represent the expected travel time from zoneorg to zonedst. No route is considered, though. In other words, a driver may have an estimation of the necessary time to reach a certain destination without considering any route, at first instant. Choosing the best route .

01. +!bestRoute(ZOrg,ZDst,R) : route(ZOrg,ZDst,T,R)02. & expectedTravelTime(ZOrg,ZDst,TExptd) & (T<TExptd)03. <- +tripRoute(ZOrg,ZDst,R,best)04. +expectedTravelTime(ZOrg,ZDst,T);05. !bestRoute(ZOrg,ZDst,R).

Plan 5b

• Keeping the instantiation for the best route. This plan finally keeps the instantiations of the appropriate parameters for bestRoute, when the recursion in Plan 5a finishes (when all attempts to satisfy the context part of Plan 5a fail, the best route then instantiated is the one to be chosen).

01.+!bestRoute(ZOrg,ZDst,R) : true<-true.

Plan 6a

• Choosing departure time in the case of having any route. Once drivers have chosen the route, they can set a departure time for the trip, given by the addition of the belief tripDepartureTime(time,[links]), in line 6.

01. +!chooseDepartureTime(R): tripRoute(ZOrg,ZDst,R,any)02. <- ?today(Day);03. ?purpose(P, Day);04. ?arrivalTime(P,TArv);05. ?route(ZOrg,ZDst,T,R);06. +tripDepartureTime(TArv-T;R).

Plan 6b

• The same is considered in the case of having the best route.01.+!chooseDepartureTime(R) : tripRoute(ZOrg,ZDst,R,best)02. <- ?today(Day);03. ?purpose(P, Day);04. ?arrivalTime(P;TArv);05. ?route(ZOrg;ZDst;T;R);06. +tripDepartureTime(TArv-T;R).

Plan 7

• This is a plan for effectively moving along the chosen route. Once a driver is committed to achieve certain destination and perceives that the desired departure time holds at that moment, it can execute its journey, then.

01. +timeNow(T) : tripDepartureTime(T,R)02. <- moveAlong(R).

Plan 8

• When the driver reaches its destination, it compares the present time with the desired arrival time. The belief perceivedArrivalCostis updated to reflect the experienced delay in future journeys.

01. +location(ZDst) : true02. <- ?today(Day);03. ?purpose(Day, P);04. ?arrivalTime(P;TArv);05. ?timeNow(T);06. +perceivedArrivalCost(T-TArv,ZDst)

Scenario 2

For the second scenario we shall consider that drivers are allowed to ask for information before starting a trip, which could help their decision-making process. No route diversion is possible, though. In this way, we shall make some considerations with regard to communication in our multi-agent traffic system. In our approach, inter-agent communication is accomplished through message passing. Some trip planner applications on the Internet are examples of such information sources.

Plan 9

• In this case, the driver opts to ask for advice. This plan is similar to Plans 2a, 2b, 2c. The driver is aimed at getting a route suggested by the system, though. In line 4, the term sysAg identifies the information system to which the message is addressed and belief predicate sysRoute provides the route suggested by sysAg.

01. +!location(ZDst) : location(ZOrg)&(not(ZDst=ZOrg))02. & preTripInformationSystem(user) & acceptanceWillingness(V)03. &(V<=20)04. <- request(sysAg,sysRoute(ZOrg,ZDst,T,R)).

Plan 10a

• An event occurs indicating that the request has been responded (or a broadcast has been sent). The decision on whether to accept it is made as soon as the message arrives. The driver accepts the suggestion and tripRoute is instantiated. In line 2, the strategy term of the predicate tripRoute is system, indicating the route was suggested by an external source. In line 3, note that only aftertripRoute has been instantiated the driver can proceed with the selection of a departure time.

01.+sysRoute(ZOrg,ZDst,T,R) : acceptanceWillingness(V)&(V<=20)02. <- +tripRoute(ZOrg,ZDst,R,system);03. !chooseDepartureTime(R).

Plan 10b

• In this case, the event occurs indicating the request has been responded (or a broadcast has been sent) but the driver ignores it and opts for the usual route instead. Other strategies could be used as alternatives to choosing the usual route, in line 2, such as any route.

01. +sysRoute(ZOrg,ZDst,T,R) : acceptanceWillingness(V)&(V>20)02. <- !usualRoute(ZOrg,ZDst,Q);03. !chooseDepartureTime(R).

Plan 11

• In the case of adopting the suggested route, setting the departure time follows the same approach as in Plans 6a and 6b.

01.+!chooseDepartureTime(R) : tripRoute(ZOrg,ZDst,R,system)02. & sysRoute(ZOrg,ZDst,T,R)03. <- ?today(Day);04. ?purpose(P, Day);05. ?arrivalTime(P,TArv);06. +tripDepartureTime(TArv-T,R).

Requesting link state• Plan 12When the driver is not very confident of its choice, it may evaluate the state of links within the route and possibly use such information to decide on another route. The following plan illustrates this situation after an anyRoute has been chosen. In this situation, the driver decides tore-evaluate the state of a certain link believed to be jammed as perceived from previous journeys. • Plan 13The answer from the information system is perceived. The drivers acceptance results in an attempt at assuming the usual route instead. After adjusting the route, the departure time needs to be re-evaluated.

Scenario 3In the third scenario, drivers can receive information from external sources both before starting a trip and during the course of a journey. VMS(variable message signs) and DRGS(dynamic route guidance systems) are examples of external sources that can be used during a trip.Non-interactive information sources.Non-interactive information sources are those aimed at reaching most users. Contrary to what we have seen so far, as drivers need to make a request to interactive sources in order to get information, communication happens in one direction only. Thus, requests are not necessary to receive information. Good examples of this kind of sources are the mass information systems, such as radio and TV. Traffic signs and, most recently, VMS, also have the ability to reach most drivers travelling throughout the network.