An Agent-based System for Truck Dispatching in Open-pit Mines

Gabriel Icarte Ahumada

1,2 a

, Eduardo Riveros

and Otthein Herzog

3,4,5 b

International Graduate School for Dynamics in Logistics (IGS), University of Bremen, Bremen, Germany

Faculty of Engineering and Architecture, University Arturo Prat – UNAP, Iquique, Chile

Center for Computing and Communication Technologies - TZI, University of Bremen, Germany

Jacobs University Bremen, Germany

Tongji University Shangai, P.R. China

Keywords:

Truck Dispatching, Open-pit Mine, Multiagent Systems, Scheduling.

Abstract:

An important logistic process in open-pit mines is material handling due to its high operational costs. In this

process shovels extract and load materials that must be transported by trucks to different destinations at the

mine. Several centralized systems have been developed to support this process. The methods applied for these

systems are based on mathematical programming, heuristic processes or simulation modelling. The main

disadvantages in these systems are performing calculations in a timely manner, addressing the dynamics of

a mine, and not being able to provide a precise dispatching solution. In this paper, we describe a distributed

approach based on Multiagent Systems (MAS). In this approach, the real-world equipment items such as

shovels and trucks are represented by intelligent agents. To meet the target in the production plan at minimal

cost, the agents must interact with each other. For this interaction, a Contract Net Protocol with a conﬁrmation

stage was implemented. To evaluate the MAS, an agent-based simulation with data from a Chilean open-pit

mine was used. The results show that the MAS provide more precise solutions than the current centralized

systems in a practical calculation timeframe. In addition, the MAS decreases the truck costs by 20% on

average.

1 INTRODUCTION

An important logistic process in open-pit mines is ma-

terial handling, since it can account for up to 50% of

the operational cost (Alarie and Gamache, 2002). In

this process, the most important equipment items are

shovels and trucks. These equipment items must work

together to extract and to transport all the material re-

quired in the operational plan at minimum cost. If the

extracted material is waste, it must be transported to a

waste dump, and if it is ore, it must be transported to

a crusher or a stockpile. Figure 1 shows all the activ-

ities that a truck must perform to transport materials

from a loading point to an unloading point. This is

called the truck cycle. This cycle is performed and

repeated by each truck until the shift ends.

To achieve efﬁcient material handling, an impor-

tant decision must be made each time that a truck ﬁ-

nalizes an unloading activity: Where does this truck

https://orcid.org/0000-0002-1997-0053

https://orcid.org/0000-0003-4781-2551

have to go now? The answer to this question is not

easy due to the number of the involved variables in the

process and the dynamics of the environment where

the equipment items operate.

In the last decades, different centralized systems

have been implemented to support truck dispatching

in open-pit mines based on mathematical program-

ming, heuristic processes or simulation. The strengths

of these methods are their maturity and their well-

known implementation. However, the weaknesses

can be observed in addressing the dynamics of a mine

(Bastos et al., 2011), not being able to provide a

precise solution (Patterson et al., 2017), using esti-

mated information (Chang et al., 2015; Costa et al.,

2005; Krzyzanowska, 2007; Newman et al., 2010),

and obtaining a dispatching solution in a timely man-

ner when the model is too complex.

One common strategy applied in some centralized

systems is based on the multistage approach (Alarie

and Gamache, 2002). This approach uses a guide-

line that is computed in the upper stage. Afterwards,

this guideline is used by the lower stage as a refer-

Icarte Ahumada, G., Riveros, E. and Herzog, O.

An Agent-based System for Truck Dispatching in Open-pit Mines.

DOI: 10.5220/0008961800730081

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 73-81

ISBN: 978-989-758-395-7; ISSN: 2184-433X

ence to make real-time dispatching decisions. For in-

stance, most centralized systems determine the num-

ber of trips to transport the required material in the

production plan from a loading point to an unloading

point. They do this before the machines work. Then,

when the trucks are working, and when one of them

ﬁnalize an unloading activity, the centralized system

selects and provides a new destination. To do this,

the centralized system considers some criteria, such

as the number of trips done between a loading and un-

loading point, the distance to a loading point or to the

loading point with the least production, among others.

This is also called an allocation model.

Figure 1: The truck cycle.

Despite the use of these systems, the trucks and

shovels do not operate efﬁciently, since queues of

trucks form in front of shovels and crushers, and there

are also shovel idle times. Therefore, the question re-

mains as how to improve the efﬁciency in the material

handling process.

Alternatively, a better solution that would allow

the equipment items to operate more efﬁciently would

be to set up schedules for each equipment item. The

schedules would contain all the activities that the

equipment items must perform, pointing out the start

times, end times, etc.

In this paper, we present a distributed solution

based on Multiagent Systems (MAS) that provide

more precise solutions than current systems. In or-

der to demonstrate the validity of the proposed MAS,

a comparison against a centralized dispatching algo-

rithm and actual data from a Chilean open-pit mine

was performed.

The remainder of this paper is structured as fol-

lows: Section 2 presents related work. Section 3

presents the distributed solution based on MAS. The

dispatching algorithm used for the evaluation of the

MAS is described in section 4. Section 5 presents the

evaluation and discussion of the proposed MAS in a

case study. Finally, the conclusions and outlook are

presented in section 6.

2 RELATED WORK

There are several articles that deal with truck dis-

patching in open-pit mines. These articles show dif-

ferent approaches that try to achieve two goals: im-

prove productivity and reduce operating costs (Alarie

and Gamache, 2002). These approaches use a cen-

tralized strategy based on methods from operations

research, simulation modelling or heuristic proce-

dures. Operations research methods are most com-

monly used. For instance, (Ahangaran et al., 2012)

proposed a real-time dispatching model considering

trucks with different capacities. The model uses two

methods: ﬂow networks and integer programming.

(Ercelebi and Bascetin, 2009) employed a closed

queuing network theory for the allocation of trucks

and linear programming for the purpose of truck dis-

patching to shovels.

Simulation modelling is also a method often em-

ployed. For instance, (Hashemi and Sattarvand, 2015)

used simulation modelling in their work taking into

account the match factor indicator, which reﬂects the

relationship between shovel and truck productivity.

(Jaoua et al., 2012) presented a simulation framework

incorporating a trafﬁc simulator with a classic discrete

event simulation model of internal transport systems.

Another method used in recent years is related

to heuristic procedures. These methods have been

quite popular in the literature because they are easy

to implement and do not require much computation

when making dispatching decisions, which is impor-

tant when decisions must be taken in real-time. For

example, (Alexandre et al., 2015) presented a com-

parison of results obtained by different evolving algo-

rithms for the Multi-objective Open-Pit Mining Oper-

ational Planning Problem. (Mendes et al., 2016) used

a multi-objective genetic algorithm.

Most of these works provide a method that deter-

mines the next destination of the truck when the latter

ends an unloading activity. To do this, they take into

account a previously calculated guideline and the cur-

rent status of the mine. These methods are not able

to provide a precise description of the activities of

shovels and trucks. Therefore, they cannot guaran-

tee a good synchronization between the activities of

the equipment items (Patterson et al., 2017).

In order to address the problem mentioned, two

previous papers proposed using schedules for all the

activities that the shovels and trucks must perform

(Chang et al., 2015; Patterson et al., 2017). To do this,

they used a dispatching algorithm and a metaheuris-

tic method to generate the schedules. Their results

showed that these algorithms generated schedules for

different size instances with good results and perfor-

mance in practical frametimes.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

3 MULTIAGENT SYSTEM FOR

TRUCK DISPATCHING IN

OPEN-PIT MINES

To the best of our knowledge, there is no evidence

that a multiagent system has been applied to truck dis-

patching in open-pit mines. However several articles

show the use and applicability of MAS in the trans-

port domain. (Chen and Cheng, 2010) performed a

literature survey on MAS applied for the transport do-

main. The research results clearly demonstrate the po-

tential of using agent technology to improve the per-

formance of trafﬁc and transportation systems.

As was mention before, an aspect that is not ad-

dressed very well for the centralized approaches is the

dynamics of a mine. Equally signiﬁcant, the dynamic

of the environment plays an important role in MAS,

since changes in the environment affect the agents.

Several studies have considered this issue. For ex-

ample, (Fischer et al., 1996) considered trafﬁc con-

gestion and its effect on the delivery process. (Gath,

2015; Mes et al., 2007) considered the dynamic en-

vironment by the arrival of orders for delivery / col-

lection of products at any time. (Chen and Cheng,

2010) pointed out that the purpose of introducing mo-

bile agents into trafﬁc and transportation systems is to

increase the ﬂexibility and the ability of the system to

deal with uncertainty in a dynamic environment.

Despite the fact that MAS has not been applied to

truck dispatching in open-pit mines, we believe that

MAS is a suitable alternative to be applied in the con-

text of an open-pit mine. This is because of the evi-

dence of the application of MAS in the transport do-

main and the addressing of the dynamics of the envi-

ronment.

The objective of the developed MAS is to accom-

plish the targets of the production plan at the min-

imum cost. To do this, the agents interact with each

other to create schedules for each equipment item that

they represent. Here, an agent represents only one

equipment item from the real-world. Table 1 shows

the implemented agents, their objectives and proper-

ties.

Our multiagent system for truck dispatching in

open-pit mines offers several advantages over current

centralized systems:

• Efﬁcient dispatching solution. Using schedules

organizes the machine activities more efﬁciently

than an allocation model. In addition, the use of

speciﬁc data from equipment items generates the

schedules more precisely.

• Robustness. Due to the fact that there is not a cen-

tral node, if any node faults, the system continues

working.

• Flexibility. An agent can change its behaviour

to adapt to the new conditions in its environment

and, in this way, to achieve its own objectives or

the objectives of the entire system.

3.1 Interaction in the Scheduling MAS

In order to create the schedules, the agents must inter-

act with each other. To do this, a Contract Net Pro-

tocol (CNP) (Smith, 1980) is implemented. In the

context of the MAS for truck dispatching, the CNP

works as follows: a shovelAgent initiates a negotia-

tion with the truckAgents sending them a call for pro-

posals (CFP). The CFP points out the time when the

shovel is available to load a truck and the idle time

from the last loading. When a truckAgent receives

a CFP, it must decide whether or not to send a pro-

posal. To do this, it asks the unloadingAgent about

the prospective waiting time and, with the informa-

tion of the CFP, it determines whether the offer ﬁts in

its schedule. If yes, it prepares a proposal, sends it

to the shovelAgent and waits for the answer. If not, it

sends a rejection to the shovelAgent.

When the shovelAgent receives all the proposals,

or when the deadline is expired, it looks for the best

proposal. Then, the shovelAgent sends an acceptance

message to the truckAgent that proposed the best pro-

posal and sends a rejection message to the other truck-

Agents that sent a proposal. After this, the truckAgent

that receives the acceptance message and the shovelA-

gent that initiated the negotiation add to their sched-

ules all the activities that must be performed and the

negotiation is then ﬁnished. The shovelAgent repeats

this protocol until the target of the production plan is

met or until it reaches the end of the shift. In the case

that a shovelAgent ends a negotiation without a win-

ner, it starts a new negotiation adding one minute to

the loading time offered in the last negotiation.

Given that the agents work in parallel, several

CNP negotiations are performed concurrently. There-

fore, a truckAgent may receive several CFPs. If the

truckAgent is taking part in a previous negotiation and

it is still waiting for the answer from a shovelAgent,

the other received CFPs will be rejected. In this con-

text, it could happened that the truckAgent rejects a

CFP that is a better option than the CFP answered pre-

viously. This problem is also called ”the eager bidder

problem” (Schillo et al., 2002). To tackle this prob-

lem, a conﬁrmation stage was included in the original

CNP: when the shovelAgent ﬁnalizes the evaluation

of the proposals, it sends a conﬁrmation message to

the truckAgent with the best proposal. The truckA-

gent that receives the conﬁrmation message accepts

An Agent-based System for Truck Dispatching in Open-pit Mines

Table 1: Agent description.

Agent Real-world

representation

Objective Properties

truckAgent Trucks To create a schedule of the

activities of the truck at minimum

cost

Capacity, loaded velocity, empty

velocity, spotting time and

unloading time, layout of the mine.

shovelAgent Shovels Front

loader

To create a schedule of the

activities of the equipment that it

represents considering its target in

the production plan

Capacity, dig velocity, load

velocity and the destination of

extracted material

unloadingPointAgent Crushers

Stockpiles

Waste dumps

Its objective is to create a schedule

of the activities of the equipment

that it represents

Number of trucks unloading

simultaneously

the conﬁrmation if it has not received a better CFP

from another shovelAgent. Otherwise, it rejects the

conﬁrmation message. If the shovelAgent receives a

rejection of the conﬁrmation message, it sends a new

conﬁrmation message to the truckAgent that sent the

second-best proposal. The shovelAgent will continue

sending conﬁrmation messages until it receives an ac-

ceptance of the conﬁrmation or until there are no more

proposals. In this way, the truckAgent might decom-

mit a previous proposal sent. Figure 2 depicts the in-

teraction between the agents using the CNP with a

conﬁrmation stage. Table 2 shows a schedule exam-

ple for a truck created by the MAS using this protocol.

Figure 2: The interaction between the agents using the CNP

with the conﬁrmation stage.

3.2 Decision Making

The decision-making process of an agent is an im-

portant characteristic. A bad design could cause the

agent to make a wrong decision or take too much time

to make it. This could also affect the performance of

the whole system. In the MAS for truck dispatching

in open-pit mines, the shovelAgents and truckAgents

make the main decisions in the system.

A shovelAgent, after receiving all the proposals

(or when the deadline is expired), must evaluate all

the proposals using a utility function. This function

promotes the proposals that propose to start the

loading on time and with the least time to perform all

the activities.More formally:

O f f er = number ∈ {0} ∪ N. It represents the time

offered by a shovel to start a loading

P = set of received proposals p

. . . .p

= (arrivalTime, cost)|arrivalTime ∈

{0} ∪ N, cost ∈ {0} ∪ N

The decision is a multicriteria problem as

Decision = arg

min

∈ P){U(p

), ..,U(p

)}

U = U(arrivalTime)

+U (cost)

U(cost) = cost

δ = (arrivalTime–o f f er) ∈ Z

U(arrivalTime) =











δ, if δ < 0 (The truck arrives earlier)

0, if δ = 0 (The truck arrives just on time)

2 ∗ δ, if δ > 0 (The truck arrives later)

Where U(arrivalTime)

is the normalized value of

U(arrivalTime) and U(cost)

is the normalized value

of U(cost). The normalization formula applied is

x − min(x)

max(x) − min(x)

A truckAgent must decide whether the offer re-

ceived can be performed by the truck. To do this, the

truckAgent looks for a free time slot considering the

loading time offered by the shovelAgent. If it ﬁnds

one, the truckAgent estimates the total time that the

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

Table 2: Example of schedule created for a truck.

Assignment Destination Start

Time of

the Trip

Arrival

Time

Start

Time of

the

Spotting

Start Time

of the

Loading or

Unloading

End Time of

the

Assignment

0 Shovel.01 00:47:01 01:20:23 01:20:23 01:21:36 01:23:12

1 WasteDump.02 01:23:12 01:32:33 01:32:33 01:32:33 01:33:23

2 Shovel.04 02:10:39 02:18:47 02:18:47 02:20:00 02:21:12

3 WasteDump.03 02:21:12 02:26:38 02:26:38 02:26:38 02:27:28

4 Shovel.04 02:27:28 02:31:37 02:31:37 02:32:50 02:34:02

5 WasteDump.03 02:34:02 02:39:28 02:39:55 02:39:55 02:40:45

truck spends to perform all the activities. Then, it de-

termines whether the offer is suitable for the free time

slot. If there is not a free time slot, the truckAgent

sends a rejection message to the shovelAgent.

Another decision by a truckAgent is made during

the conﬁrmation stage. Here, the truckAgent must de-

cide whether it accepts the conﬁrmation message or

rejects it. In the original CNP, we mentioned that a

shovelAgent would add one minute if the negotiation

ends without a winner. In the conﬁrmation stage, it

was mention that a truckAgent would reject a con-

ﬁrmation message if it has received a better CFP (a

better CFP is that one that the truck can perform at

a lower cost). If this situation is repeated continu-

ally (negotiation of a shovelAgent without a winner),

it will generate idle time in the schedule of the shovel.

To tackle this situation, the truckAgents consider

the shovel idle time since the last loading. If this

time is less than one minute, the truckAgent can reject

the conﬁrmation message. Nevertheless, if the shovel

idle time is higher or equal to one minute, the truck-

Agent must accept it. Thus, the truckAgent chooses

to achieve the target of the production plan instead of

decreasing its own cost.

4 A DISPATCHING ALGORITHM

In order to evaluate the schedules generated by the

MAS, we developed a dispatching algorithm called

Disp-ALG. Our algorithm uses the following inputs:

• Travel Times Matrix. These are the travel times

that the trucks need to travel from one point to

another point at the mine.

• Production plan. This points out the amount of

material that must be extracted by the shovels and

transported to a destination.

• Equipment data. This is information about avail-

able shovels and trucks, such as capacity and ve-

locity.

Table 3: Dispatching algorithm.

Name Description

S Set of shovels (index s)

T Set of trucks (index t)

J Set of destination of the material extracted by

shovel s (index j)

Loading time of the shovel s

Unloading time at the destination j

Travel time from destination j to shovel s

Travel time from shovel s to destination j

The algorithm output is a schedule of the activ-

ities that the equipment items must perform. The

notation of sets, indices, parameters and variables

used throughout the algorithm is shown in Table 3.

More formally:

truckActivity = {”emptyTrip” ∨ ”loading” ∨

”loadedTrip” ∨ ”unloading”}

plan = {(s, j, x)|s ∈ S, j ∈ J, x ∈ N

shovelSchs = {(s, startTime, endTime)|s ∈

S, startTime ∈ N, endTime ∈ N,t ∈ T }

truckSchs = {(t, truckActivity, startTime, endTime,

f rom,to)|t ∈ T, activity ∈ truckActivity, startTime ∈

N, endTime ∈ N, f rom ∈ (S ∪ J), to ∈ (S ∪ J)}

The Main algorithm (algorithm 1) consists of 3

steps: the ﬁrst one looks for a shovel task that is de-

ﬁned as (s, j, tll). The second step looks for the best

truck t to perform the shovel task. If the second step is

successful, the algorithm proceeds with the third step,

which consists of adding the activities and times to

the schedules of the truck t and shovel s; otherwise

the shovel task is discarded. These steps are repeated

while the operational targets of each shovel have not

been met or until the last activity of the shovels ex-

ceeds the end of the shift. To determine whether these

conditions are met, the algorithm invokes the boolean

function ISCONDITIONSMEET.

The function FINDJOB (algorithm 2) is a function

that returns a shovel task. To do this, the algorithm

selects a shovel s (SELECTSHOVEL) in the production

An Agent-based System for Truck Dispatching in Open-pit Mines

Algorithm 1: Main.

Input: S, T, Plan, H

Output: shovelSchds, truckSchds

1: shovelSchds ←

0;truckSchds ←

2: while not ISCONDITIONSMEET do

3: s, j, tll ← FINDJOB(S, shovelSchds, H)

4: if s not null then

5: T, act, act

←

FINDTRUCK(T, s, j,tll, truckSchds)

6: if t not null then

7: SCHEDULE(t,s,j,tll,shovelSchds,

truckSchds, act, act’)

8: end if

9: end if

10: end while

plan whose target has not been met. Then, it gets

the destination j of the material extracted by shovel

s (GETDESTINATION), its schedule (GETSCHEDULE)

and the end time of the last loading activity tll

(GETENDTIMELASTLOADING) of the shovel s. Fi-

nally, it returns s, j, and tll. The next time that

the function is invoked, the function selects another

shovel following a round-robin method.

Algorithm 2: FindJob.

Input: S, shovelSchds, H

Output: s, j, tll

1: tll ←

2: s ← SELECTSHOVEL(S, H)

3: if s not null then

4: return null

5: else

6: j ← GETDESTINATION(s)

7: tempShovelSchd ← GETSCHEDULE(s)

8: if tempShovelSchd is null then

9: return s, j, 0

10: else

11: tll ← GETENDTIMELASTLOAD-

ING(tempShovelSchd)

12: if tll > H then

13: return null

14: else

15: return s, j, tll

16: end if

17: end if

18: end if

After ﬁnding a shovel task, the function FIND-

TRUCK (algorithm 3) is invoked with the parameters

s, j, tll. The function returns the best truck t to per-

form the shovel task and its previous and next activi-

ties to tll (act and act

). In order to do this, the func-

tion gets the schedules of the trucks (GETSCHEDULE)

and looks for all the trucks that have a free time slot

to perform the shovel task. Then, for each of these

trucks, the algorithm calculates the time to perform

all the activities necessary to perform the shovel task.

The truck that performs all the activities in the least

amount of time is returned by the function. The pro-

cedure SCHEDULE (algorithm 4) updates the shovel

and truck schedules adding all the activities with their

start and end times.

Algorithm 3: FindTruck.

Input: T, s, j, tll,truckSchds

Output: t, act, act

1: t ←

2: for each t ∈ T do

3: tempTruckSchd ← GETSCHEDULE(t)

4: if tempTruckSchdisnull then

5: return t, null, null

6: else

7: for each act ∈ tempTruckSchd do

8: if act

activity

unload

∧ act

endtime

tll ∧ act

startTime

> tll then

9: totalTime ← c

+ c

10: if totalTime < (act

starTime

−

act

endtime

) then

11: if (tll +c

) < act

start

then

12: return t, act, act

13: end if

14: end if

15: end if

16: end for

17: end if

18: end for

19: return null

Algorithm 4: Schedule.

Input: t, s, j,tll, shovelSchds, truckSchds, act, act

Output:

1: empTrip ← t, ”emptyTrip”,tll −c

act

,tll, act

, s

2: loadAtShovel ← t, ”load”,tll, c

, s, s

3: tripToUnload ← t, ”loadedTrip”, tll + C

,tll +

, s, j

4: Unload ← t, ”unload”, tll + c

+ C

,tll + c

+ c

, j, j

5: truckSchds ← truckSchds ∪ empTrip ∪

loadAtShovel ∪tripToUnload ∪Unload

6: shovelLoad ← s, tll, tll + c

7: shovelSchds ← shovelSchds ∪ shovelLoad

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

5 EVALUATION AND

DISCUSSION

In order to validate the MAS, two experiments were

performed. In the ﬁrst one the aim was to make a

comparison between the MAS and the dispatching al-

gorithm presented in section 3 in terms of time cal-

culation to generate the schedules. The aim of the

second experiment was to compare the truck costs

obtained by the MAS and the dispatching algorithm

against the actual data. The experiments include sev-

eral simulations run in PlaSMA, which is described in

the next subsection.

5.1 Agent-based Simulation

Trialling a new method in real-world scenarios is the

perfect way to determine whether the method works

well or not. However, this is very hard to perform

because it implies risks to the involved entities. In

our case, to trialling the MAS in a real open-pit mine

would affect the performance of the entire activity of

the mine. In order to trial and assess the MAS, we

applied a multiagent-based simulation.

PlaSMA (Warden et al., 2007) is an agent-based

event-driven simulation platform created for simula-

tion and evaluation multiagent systems. It has a spe-

cial focus on simulating logistics processes. It is

based on the FIPA-compliant Java Agent DEvelop-

ment Framework (JADE) (Bellifemine et al., 2007).

The transport infrastructure within the simulator is

represented as a directed graph where nodes represent

crossroads or logistic points such as warehouses and

destinations. Edges represent different types of roads.

The platform supports graphs with up 300.000 edges

and 150.000 nodes, as well as the import of real-world

infrastructure from OpenStreetMap.

5.2 Experimental Setup

In our experiments we use actual data from a copper

open-pit mine in Chile. The modelled transport in-

frastructure contains 608 nodes and 1,272 edges. A

heterogeneous ﬂeet of trucks and shovels operate in

shifts of 12 hours. The agent properties such as ve-

locities and capacities are set based on the actual data.

Figure 3 shows a simulation deployed in PlaSMA.

All simulations have been run on a laptop com-

puter with an Intel Xeon 3 gigahertz CPU, 32 giga-

bytes of RAM and Windows 10.

Figure 3: Open-pit mine simulated in PlaSMA.

5.3 MAS vs Disp-ALG

Several scenarios of different sizes were used in our

ﬁrst experiment. The scenarios had the same charac-

teristics, i.e. the same layout of the mine and equip-

ment items with the same properties (velocity, capac-

ity, etc). The differences between the scenarios were

the size of the ﬂeet and the length of the shift (H).

Table 4 shows the scenarios and the times it took the

MAS and the Disp-ALG to generate the schedules.

Table 4: Comparison of the required time to generate the

schedules between MAS and Disp-ALG.

Scenario

H Shovels Trucks

MAS

(min)

Disp-

ALG

(min)

1 1 10 0.04 0.0025

3 3 25 0.43 0.0028

6 5 40 2.45 0.0075

9 7 60 4.52 0.0091

It observes that the MAS takes much more time

to generate the schedules in comparison to the Disp-

ALG. This is due to the overload generated by the

simulation platform where the MAS is deployed and

the communication effort among agents - aspects that

are not present in the Disp-ALG. Notwithstanding the

above, the time that the MAS takes to generate the

schedules is a practical timeframe for the mining in-

dustry.

5.4 MAS vs Actual Data

In the second experiment, we simulated ﬁve shifts.

These simulations were set using the actual data. The

MAS and the Disp-ALG generated schedules for the

equipment items to transport the same amount of

transported material as displayed in the actual data.

Table 5 shows the amount of transported material and

the truck costs to transport those materials through

An Agent-based System for Truck Dispatching in Open-pit Mines

Table 5: Comparison of transported material and truck costs between Disp-ALG, MAS and actual data.

Disp-ALG MAS Actual Data

Scenario

Shovels

Trucks

Material

Transported

(tons)

Travel

Times

(hours)

Material

Transported

(tons)

Travel

Times

(hours)

Material

Transported

(tons)

Travel

Times

(hours)

Delta

Material

Transported

(%)

Delta

Travel

Time

(%)

1 11 99 218,891 357.39 351,659 597.44 350,117 821.89 +0.44% -27.31%

2 11 98 203,570 414.81 405,903 713.34 404,921 849.96 +0.24% -16.08%

3 12 93 285,677 446.94 389,404 656.58 386,973 783.49 +0.62% -16.19%

4 12 97 284,732 439.53 384,158 698.355 381,785 831.34 +0.62% -16.01%

5 10 102 259,492 408.85 363,531 587,07 362,020 820.66 +0.42% -28.46%

the three alternatives: actual data, MAS and the Disp-

ALG.

It observes that the MAS generates schedules in

which the trucks transport more material than the

Disp-ALG. This is because the matching between a

shovel task and a truck is more precise in the MAS.

The Disp-ALG follows a sequential creation of the

schedule: it looks for the best truck for a shovel task

and then continues with the following shovel task.

However, it can happen that this truck is more appro-

priate for another shovel task. The MAS avoids this

situation thanks to the concurrent negotiation mech-

anism in which the truck can decide between several

shovel tasks and select the most appropriate. For in-

stance, it selects the shovel task with the least cost for

the truck or it selects the shovel task of the shovel with

higher idle time.

Regarding truck costs, we focused on the truck

costs in the schedules generated by the MAS, since

the schedules generated by the Disp-ALG do not

manage to transport the amount of material trans-

ported in the actual data. The schedules generated

by the MAS allow the trucks to transport the same

amount of material as the actual data (even a little

bit more), decreasing the truck costs by 20% on av-

erage in comparison with the actual data. The main

reason for these savings is that the MAS travel times

of a truck are less than the travel times from the ac-

tual data, since the agents in the MAS use the shortest

path for their travel, whereas the truck operators in the

real world decide by themselves which path to follow.

Another reason is the use of speciﬁc data to allow for

more adapted calculations of the activity times of each

equipment item, and in this way, the agents can create

more appropriate and efﬁcient schedules.

6 CONCLUSIONS

Truck dispatching in an open-pit mine is an impor-

tant and complex process. Several centralized sys-

tems support this process. However, most of them

follow an allocation model to dispatch the trucks that

does not guarantee a correct synchronization between

the operations of the equipment items. In order to ad-

dress this situation, we developed a multiagent system

(MAS) with agents that represent equipment items

from the real world. The agents interact with each

other using a contract net protocol with a conﬁrma-

tion stage to generate schedules for their represented

equipment item. Schedules are a more precise way to

organize and synchronize the equipment item activi-

ties than those proposed by the allocation models.

In order to evaluate the MAS, we implemented

several scenarios in PlaSMA, which is a simulation

platform for multiagent systems. We compared the

MAS performance against a dispatching algorithm

(Disp-ALG) developed by us and against actual data

from an open-pit mine in Chile. The results show

that the MAS achieves more efﬁcient schedules than

a centralized approach such as the Disp-ALG due to

its concurrent negotiation model. The MAS schedules

achieve the production levels of the actual data, while

the Disp-ALG does not. However, the MAS takes

much more time to generate the schedules in compar-

ison with the Disp-ALG, although the time that the

MAS takes to generate the schedules is still within a

practical timeframe. Regarding the truck costs, the

MAS schedules achieve the same production levels

(even a little bit higher), decreasing the truck costs by

20% on average.

Our work demonstrates that an agent-based sys-

tem for truck dispatching in open-pit mines is a suit-

able alternative. Several characteristics of the agent

technology such as ﬂexibility, robustness, and auton-

omy allow the agents to generate a dispatch solution

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

that is more precise than the current approaches.

In our further research, we will develop a sim-

ulated open-pit mine. In the simulation, the simu-

lated trucks will perform the operations pointed out

in their schedules. In addition, the dynamic aspects

of the material handling process will be considered,

e.g., dealing with a major change in the mine such as

equipment failures or changes in the mine layout. In

these cases, the affected agents will have to react ap-

propriately, interacting with each other to update their

schedules.

REFERENCES

Ahangaran, D. K., Yasrebi, A. B., Wetherelt, A., and Foster,

P. (2012). Real –time dispatching modelling for trucks

with different capacities in open pit mines. Archives

of Mining Sciences, 57(1):39–52.

Alarie, S. and Gamache, M. (2002). Overview of Solu-

tion Strategies Used in Truck Dispatching Systems

for Open Pit Mines. International Journal of Surface

Mining, Reclamation and Environment, 16(1):59–76.

Alexandre, R. F., Campelo, F., Fonseca, C. M., and de Vas-

concelos, J. A. (2015). A Comparative Study of

Algorithms for Solving the Multiobjective Open-Pit

Mining Operational Planning Problems. In Inter-

national Conference on Evolutionary Multi-Criterion

Optimization, volume 9019, pages 433–447.

Bastos, G. S., Souza, L. E., Ramos, F. T., and Ribeiro, C.

H. C. (2011). A single-dependent agent approach for

stochastic time-dependent truck dispatching in open-

pit mining. In 14th International IEEE Conference

on Intelligent Transportation Systems (ITSC), pages

1057–1062.

Bellifemine, F., Caire, G., and Greenwood, D. (2007). De-

veloping Multi-Agent Systems with JADE.

Chang, Y., Ren, H., and Wang, S. (2015). Modelling and

optimizing an open-pit truck scheduling problem. Dis-

crete Dynamics in Nature and Society, 2015.

Chen, B. and Cheng, H. H. (2010). A review of the applica-

tions of agent technology in trafﬁc and transportation

systems.

Costa, F. P. D., Souza, M. J. F., and Pinto, L. R. (2005). Um

modelo de programac¸

ao matem

atica para alocac¸

est

atica de caminh

oes visando ao atendimento de

metas de produc¸

ao e qualidade. Rem: Revista Escola

de Minas, 58(1):77–81.

Ercelebi, S. G. and Bascetin, A. (2009). Optimization of

shovel-truck system for surface mining. Journal of the

Southern African Institute of Mining and Metallurgy,

109(7):433–439.

Fischer, K., M

uller, J. P., and Pischel, M. (1996). Coopera-

tive transportation scheduling: an application domain

for dai. Applied Artiﬁcial Intelligence, 10(1):1–34.

Gath, M. (2015). Optimizing Transport Logistics Processes

with Multiagent-based Planning and Control. PhD

thesis.

Hashemi, A. S. and Sattarvand, J. (2015). Proceedings

of the 12th International Symposium Continuous Sur-

face Mining - Aachen 2014. pages 213–214.

Jaoua, A., Riopel, D., and Gamache, M. (2012). A simula-

tion framework for real-time ﬂeet management in in-

ternal transport systems. Simulation Modelling Prac-

tice and Theory, 21(1):78–90.

Krzyzanowska, J. (2007). The impact of mixed ﬂeet hauling

on mining operations at Venetia mine. Journal of the

Southern African Institute of Mining and Metallurgy,

107(4):215–224.

Mendes, J. B., D’Angelo, M. F. S. V., Maia, N. A.,

and Veloso, R. R. (2016). A Hybrid Multiobjec-

tive Evolutionary Algorithm for Truck Dispatching in

Open-Pit-Mining. IEEE Latin America Transactions,

14(3):1329–1334.

Mes, M., van der Heijden, M., and van Harten, A. (2007).

Comparison of agent-based scheduling to look-ahead

heuristics for real-time transportation problems. Euro-

pean Journal of Operational Research, 181(1):59–75.

Newman, A. M., Rubio, E., Caro, R., Weintraub, A., and

Eurek, K. (2010). A review of operations research in

mine planning. Interfaces, 40(3):222–245.

Patterson, S. R., Kozan, E., and Hyland, P. (2017). Energy

efﬁcient scheduling of open-pit coal mine trucks. Eu-

ropean Journal of Operational Research, 262(2):759–

770.

Schillo, M., Kray, C., and Fischer, K. (2002). The eager bid-

der problem: a fundamental problem of DAI and se-

lected solutions. Proceedings of the 1st international

Joint Conference on Autonomous Agents and Multia-

gent Systems (AAMAS’02), (January):599–606.

Smith, R. G. (1980). The Contract Net Protocol:High-

level communication and control in a distributed prob-

lem solver. IEEE Tranactions on Computers, C-

29(12):1104–1113.

Warden, T., Porzel, R., Gehrke, J. D., Herzog, O., Langer,

H., and Malaka, R. (2007). Towards Ontology-Based

Multiagent Simulations : Plasma Approach.

An Agent-based System for Truck Dispatching in Open-pit Mines