On the Problem of Task Planning in Multi-robot Systems

Zhi Yan, Nicolas Jouandeau and Arab Ali Cherif

Advanced Computing Laboratory of Saint-Denis (LIASD), Paris 8 University

2 Rue de la Libert´e, 93526 Saint-Denis, France

Keywords:

Task planning, Multi-robot Systems, Coordination.

Abstract:

Multi-robot task planning (MRTP) is one of the fundamental problems for multi-robot systems. An important

question facing this research topic is, which robot should execute which task so as the expected overall sys-

tem performance can be maximized? Many approaches have been proposed for such a purpose. This paper

investigates the existing works in the ﬁeld. The approaches have been surveyed and some representatives are

compared with detailed results. A brief discussion and further research perspectives are also given at the end

of the paper.

1 INTRODUCTION

The task of coming up with a sequence of actions that

will achieve a goal is called planning (Russell and

Norvig, 2002). In multi-robot systems (MRS), plan-

ning can be used to coordinate robots to accomplish

the team mission. Unfortunately, optimal planning for

MRS is typically an N P -hard problem. The chal-

lenge then is to have tractable planning that produces

good solutions (Dias et al., 2005). Multi-robot plan-

ning is usually divided into two aspects: task planning

and motion planning. Task planning involves task de-

composition and task allocation. So far, research on

task planning of MRS is concentrated on task alloca-

tion problem, and relatively little on task decompo-

sition problem. In fact, the task decomposition is an

important research topic because the effect of task al-

location could be directly inﬂuenced by it.

2 MULTI-ROBOT TASK

DECOMPOSITION

The problem of multi-robot task decomposition

(MRTD) mainly refers to that the total mission to

be completed is decomposed into several single sub-

tasks which can be completed independently by a

robot separately, according to the characteristic, re-

quirement and resource allocation of the total mis-

sion itself (Chen et al., 2010). (Stone and Veloso,

1999) achieved collaboration between agents through

the introduction of formations which decomposes the

task space deﬁning a set of roles with associated be-

haviors. (Botelho and Alami, 1999) presented a de-

centralized system to describe and perform task plan-

ning, decomposition and allocation in multi-robot en-

vironments, called M+ protocol. (Zlot and Stentz,

2005) focused on complex tasks that can be decom-

posed into multiple inter-related subtasks. They ad-

dressed the task decomposition problem by general-

izing tasks to task trees within a peer-to-peer trading

market. (Tang and Parker, 2005) considered that, in

typical approaches to multi-robot team working, the

decomposition of the team task into subtasks is de-

ﬁned by the human designer in advance of the robot

team performance, and this pattern also outlines the

available multi-robot task solutions in advance of the

mission. So they described a methodology for au-

tomatically synthesizing task solutions for heteroge-

neous multi-robot teams.

3 MULTI-ROBOT TASK

ALLOCATION

The problem of multi-robot task allocation (MRTA)

can be considered as an instance of the well-known

optimal assignment problem (OAP). In domains

where the group dynamics have a signiﬁcant effect

on group performance, MRTA is known to be N P -

hard (Gerkey and Matari´c, 2004) (Dahl et al., 2009).

This is the most common case in MRS. The task al-

location in MRS is generally undertaken to assign

robots to tasks (or tasks to robots) so as to maximize

423

Yan Z., Jouandeau N. and Ali Cherif A..

On the Problem of Task Planning in Multi-robot Systems.

DOI: 10.5220/0004017204230426

In Proceedings of the 9th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2012), pages 423-426

ISBN: 978-989-8565-22-8

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

the expected overall system performance. However,

because MRTA becomes a dynamic decision prob-

lem that varies in time with environmental changes,

the static assignment method is no longer applicable.

Thus an alternative solution is to iteratively solve the

static assignment problem over time.

(Gerkey and Matari´c, 2004) gave a formal analysis

and domain-independent taxonomy of MRTA prob-

lems, in which the MRTA problems have been classi-

ﬁed into seven categories according to ability of robot

to perform tasks, number of robots required for a task,

and manner of the task assignment. They also ana-

lyzed and compared some iterated assignment archi-

tectures: ALLIANCE (Parker, 1994a), BLE (Werger

and Matari´c, 2000), and M+ (Botelho and Alami,

1999), and some online assignment architectures:

MURDOCH (auction-based) (Gerkey and Matari´c,

2002), ﬁrst-price auctions (market-based) (Stentz and

Dias, 1999), and dynamic role assignment (Chaimow-

icz et al., 2002), for MRTA, respectively.

The contract net protocol (CNP) has been devel-

oped by (Smith, 1980), which aims to achieve task

assignment with distributed control by a negotiation

process in multi-agent systems. So far, most meth-

ods of MRTA are based on the CNP model. (Botelho

and Alami, 1999) presented M+ system, a scheme for

multi-robot cooperation through negotiated task allo-

cation and achievement, which is the ﬁrst CNP-based

approach to MRTA. (Stentz and Dias, 1999) presented

the ideas of free market architecture for coordinating

a group of robots to achievea given objective (market-

based approach). This architecture deﬁnes revenue

and cost functions across the possible plans for ex-

ecuting a speciﬁed task. The task is accomplished by

dividing it into subtasks and allowing the robots to

bid and negotiate to carry out these subtasks. The ob-

jective is achieved by individual robots cooperating

and competing with each other to further their own

self-interests. (Zlot et al., 2002) applied these ideas to

multi-robot mapping and exploration problem. This

work borrows the market architecture which seeks to

maximize beneﬁt while minimizing cost, thus aiming

to maximize utility. The beneﬁt is information gained

by visiting a goal point, the cost is the estimated dis-

tance traveled to reach the goal (by using D* algo-

rithm (Stentz, 1994)), then the utility is the difference

between the beneﬁt and the cost. The market-based

coordination architecture has been applied to a Mars

exploration scenario (combined with D* algorithm

for robot motion planning). (Gerkey and Matari´c,

2002) presented the ﬁrst online assignment architec-

ture MURDOCH, which uses a ﬁrst-price auction to

assign each task (auction-based approach). The auc-

tion proceeds in ﬁve steps: task announcement, met-

ric evaluation, bid submission, close of auction, and

progress monitoring/contract renewal. The MUR-

DOCH system has been tested in two different do-

mains: a long-term loosely coupled task domain and a

short-term tightly coupled box-pushing task. The ma-

jor differences between auction-based approach and

market-based approach are that, 1) auction-based ap-

proach uses the bid based on the estimated cost, but

market-based approach takes into account both cost

and beneﬁt, 2) auction-based approach does not allow

task reassignment, but market-based approach allows

later reassignment.

In addition, (Dahl et al., 2009) presented an algo-

rithm for task allocation in groups of homogeneous

robots, which is based on vacancy chains, a resource

distribution strategy common in human and animal

societies. This algorithm uses local task selection, re-

inforcement learning for estimation of task utility, and

reward structures based on the vacancy chain frame-

work. (Hanna, 2005) proposed an approach which

allows robots to take into account the uncertainty of

task execution. They decomposed the MRTA prob-

lem into two stages. In the ﬁrst stage, each robot se-

lects its own tasks based on the expected beneﬁt using

Markov decision process (MDP). In the second stage,

an auction-based mechanism is applied to assign tasks

to robots. (Michael et al., 2008) proposed a dis-

tributed market-based coordination algorithm where

agents are able to bid for task assignment with the as-

sumption that agents have knowledge of all tasks as

well as the maximum number of agents that can be

assigned to every individual task. Each auction is per-

formed among neighboring groups of agents and re-

quires only local communication. They veriﬁed their

algorithm in multi-robot formation control problem.

(Shiroma and Campos, 2009) proposed a framework

called CoMutaR, which is designed to both tackle task

allocation and coordination problems in MRS. This

framework enables the single robot to perform mul-

tiple tasks concurrently by periodically checking and

updating task-related information during implemen-

tation. It has been tested and evaluated in simulation

in object transportation, area surveillance, and multi-

robot box pushing problem. (Wawerla and Vaughan,

2010) presented two task allocation strategies for a

multi-robot transportation system. One is based on

a centralized planner that uses domain knowledge to

solve the assignment problem in linear time. The

other enables individual robots to make individual

task allocation decisions using only locally obtainable

information and single value communication. (Yan

et al., 2011) developed a lightweight and robust de-

centralized approach based on trade rules for coordi-

nated multi-robot exploration (trade-based approach).

ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics

424

Table 1: A Comparison of Three CNP-based Online Multi-robot Task Allocation Methods.

Name Negotiation Task Task allocation ability

(Communication) allocation per iteration

model algorithm (trading round)

market-based publish/subscribe greedy algorithm (ﬁrst-price auction) single task

auction-based publish/subscribe greedy algorithm (ﬁrst-price auction) single task

trade-based apply/allocate greedy algorithm (ﬁrst-price auction) multiple tasks

Name Utility Task Communication Computation

consideration reassignment complexity complexity

market-based cost and beneﬁt allowed O(1)/bidder, O(n)/auctioneer O(n)

auction-based cost not allowed O(1)/bidder, O(n)/auctioneer O(n)

trade-based cost allowed O(1)/buyer, O(n)/seller O(n)

This approach is designed to simulate the relation-

ship between buyers and sellers in a business sys-

tem, to achieve dynamic task allocation by using a

mechanism of unsolicited bid. A comparison be-

tween market-based, auction-based and trade-based

approaches is given in Table 1.

Eventually, the task allocation for heterogeneous

and homogeneous systems may be different. In het-

erogeneous systems, task allocation may be deter-

mined by individual capabilities. (Parker, 1994b) in-

troduced the concept of task coverage, which mea-

sures the ability of a given team member to achieve

a given task. This parameter can be used as an index

to organize a robot team to perform a mission from

the available pool of heterogeneous robots. Task cov-

erage reaches the maximum value in homogeneous

teams, and decreases as teams become more hetero-

geneous. But in homogeneous systems, agents may

need to differentiate into different roles at design time

or dynamically at run time (Yan et al., 2011).

4 CONCLUSIONS

In this paper, we have brieﬂy discussed the existing

works on the problem of multi-robot task planning

(MRTP). Two important aspects, multi-robot task de-

composition (MRTD) and multi-robot task allocation

(MRTA), were identiﬁed. Three online MRTA meth-

ods based on contract net protocol, market-based,

auction-based and trade-based, were compared with

detailed results. Because there is no common bench-

marking platform and planning problems, thus the

purpose of the comparison is not to point out which

method is the best, but give an overview to im-

prove our understanding. Different strategies might

be suited to different planning scenarios.

We believe that the following are some promising

avenues for future research:

• The task decomposition can directly affect the ef-

ﬁciency of task allocation. A more complete the-

ory should be achieved.

• Environments where robots work in are usually

dynamic. Due to the unpredictability and uncer-

tainty of the environment, the revenue and cost

functions for task allocation will be difﬁcult to de-

ﬁne. For the purpose of maximizing system per-

formance, robust deﬁnitions and metrics for vari-

ous scenarios should be developed.

• The motion of the robot is also an important fac-

tor inﬂuencing task achievement. A future chal-

lenge is to combine both task and motion plan-

ning reasonably, in order to make the multi-robot

systems can be situated to more complex tasks in

more complex environments.

REFERENCES

Botelho, S. C. and Alami, R. (1999). M+: A scheme for

multi-robot cooperation through negotiated task allo-

cation and achievement. In Proceedings of the 1999

IEEE International Conference on Robotics and Au-

tomation (ICRA’99), pages 1234–1239, Detroit, MI,

USA.

Chaimowicz, L., Campos, M. F. M., and Kumar, V. (2002).

Dynamic role assignment for cooperative robots. In

Proceedings of the 2002 IEEE International Confer-

ence on Robotics and Automation (ICRA’02), pages

293–298, Washington, DC, USA.

Chen, J., Yang, Y., and Wei, L. (2010). Research on the ap-

proach of task decomposition in soccer robot system.

In Proceedings of the 1st International Conference on

Digital Manufacturing and Automation (ICDMA’10),

pages 284–289, Changsha, China.

Dahl, T. S., Matari´c, M. J., and Sukhatme, G. S. (2009).

Multi-robot task allocation through vacancy chain

scheduling. Robotics and Autonomous Systems, 57(6–

7):674–687.

Dias, M. B., Zlot, R., Kalra, N., and Stentz, A. (2005).

Market-based multirobot coordination: A survey

OntheProblemofTaskPlanninginMulti-robotSystems

425

and analysis. Technical Report CMU-RI-TR-05-13,

Carnegie Mellon University, Pittsburgh, PA, USA.

Gerkey, B. P. and Matari´c, M. J. (2002). Sold!: Auction

methods for multirobot coordination. IEEE Transac-

tions on Robotics and Automation, 18(5):758–768.

Gerkey, B. P. and Matari´c, M. J. (2004). A formal analysis

and taxonomy of task allocation in multi-robot sys-

tems. The International Journal of Robotics Research,

23(9):939–954.

Hanna, H. (2005). Decentralized approach for multi-robot

task allocation problem with uncertain task execu-

tion. In Proceedings of the 2005 IEEE/RSJ Interna-

tional Conference on Intelligent Robots and Systems

(IROS’05), pages 535–540, Alberta, Canada.

Michael, N., Zavlanos, M. M., Kumar, V., and Pappas, G. J.

(2008). Distributed multi-robot task assignment and

formation control. In Proceedings of the 2008 IEEE

International Conference on Robotics and Automation

(ICRA’08), pages 128–133, Pasadena, CA, USA.

Parker, L. E. (1994a). ALLIANCE: An architecture for fault

tolerant, cooperative control of heterogeneous mobile

robots. In Proceedings of the 1994 IEEE/RSJ Interna-

tional Conference on Intelligent Robots and Systems

(IROS’94), pages 776–783, Munich, Germany.

Parker, L. E. (1994b). Heterogeneous Multi-Robot Cooper-

ation. PhD thesis, Massachusetts Institute of Technol-

ogy.

Russell, S. J. and Norvig, P. (2002). Artiﬁcial Intelligence:

A Modern Approach (2nd Edition). Prentice Hall.

Shiroma, P. M. and Campos, M. F. M. (2009). CoMutaR: A

framework for multi-robot coordination and task allo-

cation. In Proceedings of the 2009 IEEE/RSJ Interna-

tional Conference on Intelligent Robots and Systems

(IROS’09), pages 4817–4824, St. Louis, MO, USA.

Smith, R. G. (1980). The contract net protocol: High-

level communication and control in a distributed

problem solver. IEEE Transactions on Computers,

29(12):1104–1113.

Stentz, A. (1994). Optimal and efﬁcient path planning for

partially-known environments. In Proceedings of the

1994 IEEE International Conference on Robotics and

Automation (ICRA’94), pages 3310–3317, San Diego,

CA, USA.

Stentz, A. and Dias, M. B. (1999). A free market architec-

ture for coordinating multiple robots. Technical Re-

port CMU-RI-TR-99-42, Carnegie Mellon University,

Pittsburgh, PA, USA.

Stone, P. and Veloso, M. (1999). Task decomposition, dy-

namic role assignment, and low-bandwidth commu-

nication for real-time strategic teamwork. Artiﬁcial

Intelligence, 110(2):241–273.

Tang, F. and Parker, L. E. (2005). Asymtre: Automated syn-

thesis of multi-robot task solutions through software

reconﬁguration. In Proceedings of the 2005 IEEE In-

ternational Conference on Robotics and Automation

(ICRA’05), pages 1501–1508, Barcelona, Spain.

Wawerla, J. and Vaughan, R. T. (2010). A fast and frugal

method for team-task allocation in a multi-robot trans-

portation system. In Proceedings of the 2010 IEEE In-

ternational Conference on Robotics and Automation

(ICRA’10), pages 1432–1437, Anchorage, AK, USA.

Werger, B. B. and Matari´c, M. J. (2000). Broadcast of local

eligibility for multi-target observation. In Proceedings

of the 5th International Symposium on Distributed Au-

tonomous Robotic Systems (DARS’00), pages 347–

356, Knoxville, TN, USA.

Yan, Z., Jouandeau, N., and Ali Cherif, A. (2011). Multi-

robot decentralized exploration using a trade-based

approach. In Proceedings of the 8th International

Conference on Informatics in Control, Automation

and Robotics (ICINCO’11), pages 99–105, Noordwi-

jkerhout, The Netherlands.

Zlot, R. and Stentz, A. (2005). Complex task allocation

for multiple robots. In Proceedings of the 2005 IEEE

International Conference on Robotics and Automation

(ICRA’05), pages 1515–1522, Barcelona, Spain.

Zlot, R., Stentz, A. T., Dias, M. B., and Thayer, S. (2002).

Multi-robot exploration controlled by a market econ-

omy. In Proceedings of the 2002 IEEE International

Conference on Robotics and Automation (ICRA’02),

pages 3016–2023, Washington, DC, USA.

ICINCO2012-9thInternationalConferenceonInformaticsinControl,AutomationandRobotics

426