Multi-agent Case-based Reasoning Inference Engine Proposal

for Reusable Robotics

Francisco Gindre, D. Kotlirevsky, S. Scaine, N. Jafelle Fraga and M. Daniela López De Luise

AIGroup, Universidad de Palermo, Mario Bravo 1050 8

floor, Ciudad Autónoma de Buenos Aires, Argentina

Keywords: Case-based Reasoning, Reusable, Service Robotics, SaaS, Intelligent Agents.

Abstract: Different articles and surveys on the matter identify a remarkable success of robotics on static environments

that has not replicated on dynamic areas of application. This paper presents the proposal of a research

project that identifies the main hurdles of this issue and present feasible solutions for the matter.

1 INTRODUCTION

The application of robotics as a mean to simplify or

outperform different tasks performed by humans has

been a matter of research since the 1960’s. An

exhaustive survey on “The Evolution of Robotics

Research” (Armada et al., 2007), reveals many

aspects of this discipline that have been in sight of

many research teams throughout the last five

decades.

The success of applied robotics is widely known

in the manufacturing industry (Armada et al., 2007),

providing improvements in key areas as

productivity, quality assurance and safety. This is a

key aspect for today’s trends in robotics application

as remarked on the mentioned article: “Currently,

the creation of new needs and markets outside the

traditional manufacturing robotic market (i.e.,

cleaning, demining, construction, ship building,

agriculture) and the aging world we live in is

demanding field and service robots to attend to the

new market and to human social needs” (Armada et

al., 2007). Several authors address many of the

hurdles that stand between the actual state of the art

of service robotics and the demands expressed in

Garcia Armada’s survey. The following subsections

identify three major issues to overcome in the field

of reusable robotics.

1.1 The Dynamic Environment Issue

The manufacturing field presents one significant

advantage over other possible applications of

robotics: a structured, controlled and limited

environment. Uncertainty is considered an undesired

deviation and therefore is handled by human

operators (Armada et al., 2007). Robots working on

dynamic environments require the ability of

autonomously dealing with uncertainty. Many Soft

Computing techniques have been applied to meet

this requirement such as Bayesian and Neural nets,

Expert Systems (Russell and P., 2003); (Amit,

2000); Concept Learning (López De Luise et al.,

2010), Genetic Algorithms (Gindre et al., 2010) and

Case-Based Reasoning (CBR) (Kolodner, 1992) as

problem solving tools.

1.2 Hardware Variability: Reuse Vs.

Re-work

The cited survey on robotics evolution reveals a

variety of fields in which robotics is being applied

and enumerates various examples of architectures

used for each one of them. This variety is recognized

as limitation in terms of “reusability against re-

work” due to incompatibilities between different

hardware architectures and programming languages

and paradigms. NASA’s CLARAty (Nesnas et al.,

2006) and the “Player/Stage” (Gerkey et al., 2003)

architectures are two approaches towards the

mitigation of the re-work issue by adding abstraction

layers and standardizing communications between

them (with socket APIs).

A study on robotics software reusability (Mallet

et al., 2007) identifies two different aspects of that

matter: Software Architecture and Control

Architecture. The former refers to the usage of

software engineering best practices that facilitate

486

Gindre F., Kotlirevsky D., Scaine S., Jafelle Fraga N. and Daniela López De Luise M..

Multi-agent Case-based Reasoning Inference Engine Proposal for Reusable Robotics.

DOI: 10.5220/0004252204860491

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 486-491

ISBN: 978-989-8565-39-6

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

reuse. The latter refers to the assembly of such

components into an executable piece of software that

runs on real hardware.

Furthermore, Mallet’s research enumerates

certain factors that help reuse like: “Avoid all-in-one

approaches”, “Avoid static design” and usage of

frameworks as a mean to software abstraction. It

also introduces various examples on OpenRobots

(LAAS-CNRS, 2012) projects that attempt to reduce

re-work and improve reuse specific domains of

application.

1.3 Per-agent Knowledge Issue

Machine learning techniques as CBR or Concept

Learning combine present and past contexts as

feedback to address new situations that must be dealt

with. When having many units (agents) deployed on

the same or a similar environment, local knowledge

bases are generated as a result of using such

algorithms. This derives in the local availability of

new information of the past experiences of each

agent, which generates a “Per-Agent” information

island (Cengeloglu et al., 1994). Such information

schema is a known Anti-Pattern (ISRD Group,

2005) in terms of Software Engineering can be turn

into a reusable software Service-Client pattern by

adding a service-like repository that allows the

information to be shared among the deployed agents

(Patterson and Fox, 2012).

Figure 1a: The CBR Cycle (Kolodner, 1992).

This paper presents the architecture and

prototype design of the SIR project (for its name in

Spanish “Sistema Inteligente Robotizado”). A Case-

Based Reasoning inference engine that helps

intelligent robot agents to cope with dynamic

environments and their challenges. Designed with a

component-based architecture to improve reusability

and reduce re-work by adding abstraction layers and

using broadly used communication protocols to

decouple software from hardware variability and

avoid per-agent information islands.

2 THE SIR_CORE CBR-ENGINE

PROPOSAL

The SIR project aims to develop a system that

provides a centralized inference engine to multiple

collaborating service robots regardless their

hardware or software capabilities. This centralized

inference system (SIR-CORE) aims to deliver a

common framework that allows these robots to have

the capability of acting upon unknown situations by

reusing their own or their collaborators’ past

experiences. The following subsections explain the

selected approaches to address the difficulties shown

in sections 1.1, 1.2 and 1.3.

2.1 CBR for Dynamic Environments

Expert or Knowledge Based Systems (KBS) have

been widely used to capture and represent the

knowledge of an expert of a specific domain and

translate it into a representation that can be of use to

a computer system that emulates the assertions of

the expert. An review of the state of the art of CBR

(Watson and Marir, 1994) recollects that the

efficiency and accuracy of the KBS is tightly

coupled to how effective and exhaustive was the

knowledge elicitation process and how well it is

implemented and represented as a piece of software.

It also identifies the following drawbacks of KBS:

 Knowledge elicitation is a difficult process, often

being referred to as the knowledge elicitation

bottleneck;

 Implementing KBS is a difficult process

requiring special skills and often taking many man

years;

 once implemented model-based KBS are often

slow and are unable to access or manage large

volumes of information; and

 they are difficult to maintain

These setbacks represent the opposite requirements

for a reusable inference engine. CBR presents itself

as a technique to overcome the mentioned setbacks

of KBS (Kolodner, 1992).

Firstly, CBR eliminates the knowledge elicitation

bottleneck by reducing it to the act of gathering the

different cases that are presented to a given CBR

agent in a particular area of application. Its

elicitation process means, providing base cases plus

the appropriate means to sense the environment and

gather useful information from it to conform a case

to the case-based reasoner.

Even though CBR systems are initialized with a

set of known cases provided by experts on the area

of application, the elicitation process continues

Multi-agentCase-basedReasoningInferenceEngineProposalforReusableRobotics

487

within the CBR cycle itself (figure 1). When a

dynamic environment presents an unknown context

to a given CBR agent, this context can be elicited as

a new case that the agent will attempt to solve with

solution of the most similar case present on it’s own

case knowledge base.

Figure 1b: The CBR Cycle (Aamodt and Plaza, 1994).

Secondly, a single case or a subset of similar

cases can be selected without traversing a set of IF-

THEN rules. Cases can be selected by a variety of

methods depending the data that they encapsulate.

This property enforces reusability, since the case

retrieval can be generalized to the use of a metric

that measures the grade of similitude between 2 or

more cases. However, there is an underlying setback

in it. Querying a vast CBR knowledge base is not

effortless. Kolodner identifies it as the indexing

problem: “is the problem of retrieving applicable

cases at appropriate times” (Kolodner, 1992). This

problem is accentuated when the number of cases in

the KB increases significantly since retrieval

performance decays.

Efforts to mitigate this problem include case

labelling (indexing), two-step ranking that perform

an O(n) search (Walker, 2007). A Hierarchical

organization of cases with induction and decision

Trees achieved a good case retrieval performance.

Nevertheless, according to (Walker, 2007) new case

elicitations tended to render the classifier useless.

In terms retrieving cases (Blanzieri and Ricci,

1999) and (Walker, 2007) conclude that Nearest

Neighbour Classifiers show superior results opposed

to the mentioned techniques. Cunningham’s paper

(Cunningham, 2009) shows different techniques to

measure the distance between two or more cases.

Other works show the usage of relational databases

as Case KB’s by adapting the SQL language to cope

with its limitations to query non-relational data

(Portinale and Montani, 2008) and the use of data

mining techniques as a knowledge discovery aid for

CBR KB (Aamodt et al., 1998).

The evidence shows that the indexing problem

constitutes a scalability limitation for Case Based

Reasoning. Since are non-relational data, RDBMS

need to be modified in order to be viable Case

knowledge bases. The SIR project proposes the use

of NoSQL document databases to handle large

collections of non-relational information (Padhy et

al., 2011).

Thirdly, “A case-based reasoner solves new

problems by adapting solutions that were used to

solve old problems” (Reisbeck and Shank, 1989).

This means that “maintenance” of the knowledge

base occurs on the CBR cycle intrinsically, as a

desired effect instead of as an action to fix the KB

itself. Figures 1a and 1b describe how the Adapt-

Criticize-Evaluate sub-cycle fits in the CBR cycle

proposed by Janet Kolodner. When a CBR agent

senses the environment a new case is created. This

new case will be matched with the existing cases.

Given there is feasible match, the known resolution

will be applied to the sensed case. The application of

this solution will be criticized and evaluated for its

results and adapted if necessary. The result will be

stored in the knowledge base as a new elicited case.

In conclusion, because of the mentioned

contrasts, the CBR cycle can be applied to many

problems (Watson and Marir, 1994). This wide

range of application is major requirement in the

design of the SIR project and one of the main

reasons of using CBR instead of other KBS.

Figure 2: RESTful communications over HTTP.

2.2 RESTfulness for Reuse

Software abstractions are a mean to achieve software

reuse. Both CLARAty and Player/Stage focus on

abstracting communication layer between

components to separate behavior from

implementation. In these terms the SIR project

focuses in designing a communication layer over a

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

488

protocol (TCP/IP) that relies on a standard structure

to represent the data rather than sending arbitrary

bytes transferred over sockets. SIR_CORE is a Web

Service (Patterson and Fox, 2012) that relies on

REST (Fielding, 2000) over HTTP (Berners Lee et

al., 1996) to exchange data and requests with the

deployed agents. Service robot agents use the

SIR_CORE CBR engine “as a Service” by

requesting feedback from a given case sending a

GET request to a RESTful URI provided by

SIR_CORE to all the agents. RESTfulness provides

a standardized communication between clients and

servers that avoids the All-in-One approach and

provides the necessary interfaces to for each party to

implement them as needed in any hardware platform

provided that the data interchange adopts the

standard protocols and formats. HTTP supports

many data interchange formats such as JSON

(json.org, 2006) that is a widely used “lightweight,

text-based, Language-independent data interchange

format” or XML (W3 Consortium, 2012).

This makes possible that human operator request

information to the SIR_CORE from a Web

application as well.

Figure 3: Proposed service robotic agent architecture.

2.3 Shared Knowledge Base

Robot Swarms are presented as a collection of

homogenous robots with a simple hardware

architecture collaborating towards a common (and

collaborative) objective (Mohan and Ponnambalam ,

2009). Opposed to that, autonomous standalone

robots are have complex architectures (that can be

heterogeneous as CLARAty) that implement

different mechanisms, like concept learning to

“survive” in the environment they are introduced

into while they attempt to accomplish a given task.

Deploying various robots of the same kind,

configured with the same software and KB on an

equal dynamic environment, can result in diverse

knowledge bases depending on the situations these

different agents went through.

In order to avoid information islands,

autonomous robots would have to synchronize their

knowledge bases either with a central entity or

between peers. Although there are frameworks that

address knowledge exchange between peers

(Cengeloglu et al., 1994), that information would be

shared among those agents who are in range to

establish a communication and that are mutually

compatible. The SIR project proposes a centralized

client/server or Software as Service (SaaS)

architecture that allows a set of heterogeneous

agents to share their knowledge by persisting the

sensed cases in the SIR_CORE server.

This design allows cases to flow transparently

among peers while also contributes to simplify the

hardware implementation of each client, since part

of the computing needs (the CBR) are outsourced to

the SIR_CORE service. Since the communication

protocols used are basically the Internet protocols,

there is virtually no limitation of distance between

these agents between themselves or the SIR_CORE

in order to share their knowledge base. The only

limitation is how much latency is acceptable for the

specific tasks the service robotic agents perform.

3 THE SIR_ROBOT MODULE

The SIR project is designed to let the SIR_CORE

deal with the computational complexity of the

inference engine, so that robotic agents can be as

simple as the problem allows them to be.

A manufacturer adopting the SIR architecture

will only be obliged to provide the necessary

hardware resources for a piece of software to create

and transmit RESTful requests through (wireless)

network that reaches a route to SIR_CORE and

parse the service’s response into executable

instructions for the robot agent.

4 PROJECT FEASIBILITY

4.1 Off-the-Shelf Hardware

Figure 3 depicts a retail off-the-shelf SIR_ROBOT

implementation. The proposed agent prototype

consists of a Lego MindStorms® NXT 2.0 kit (Lego,

2012) with the Open Source LeJOS firmware

(Bagnall, 2011) which runs a Java Virtual Machine

and an Android 2.3 compatible mobile phone

(Google, 2012).

Multi-agentCase-basedReasoningInferenceEngineProposalforReusableRobotics

489

Since the Lego kit is not capable of reaching

SIR_CORE through HTTP an additional level of

abstraction was added which makes this

implementation of SIR_ROBOT a 2-tiered

application. The uppermost tier (android phone)

provides the means of complying with mandatory

requirement of SIR: reaching SIR_CORE over

HTTP RESTful requests. Additionally, this tier

parses the response and communicates the actions

over serialized objects via Bluetooth 2.0 to the

LeJOS firmware and vice versa. A Linksys 100Mbit

router is being use as access point between the

android terminal and the SIR_CORE.

4.2 Wireless Network Capabilities

RESTFul requests add an additional abstraction

layer over hardware and network communications to

maximize software reuse and tolerance to hardware

variability at the expense of additional latency. The

feasibility of this architecture in terms of network

resources consumption is being estimated on current

Wi-Fi standards like IEEE 802.11n (IEEE, 2009)

and projected to state of the art Gigabit Wireless

estimations (Kneckt et al., 2011). For cases where

response times are critical, it is being considered to

enhance the SIR_ROBOT with an additional tier that

handles certain self-preservation behaviours that do

not depend of RESTful requests to the server.

(López De Luise et al., 2010).

Figure 4: The SIR Cycle.

5 CONCLUSIONS

The objective of the SIR project is to design and

implement software architecture that allows n

service robotic agents that rely on diverse hardware

implementations and capabilities to interact with an

inference SaaS platform (SIR_CORE) to perform

different tasks.

This paper presents the rationale upon which the

SIR project was designed. Along with its strengths

and weaknesses and the proposals for enhancing the

former and mitigating the latter. Additionally, Figure

4 show how the SIR architecture fits n robotic agents

into the CBR cycle with minimum hardware

requirements by relying in State of The Art SaaS and

WebServices communication protocols and existing

an network infrastructure.

6 FUTURE WORK

Activities planned for Q3, Q4 2012 and Q1 2013

include the implementation of the SIR_CORE basic

architecture, assembly of two distinct SIR_ROBOT

implementations to follow basic cases from a unique

SIR_CORE Service, latency and stress tests using

802.11b, g, n and public 3G cellular networks.

REFERENCES

Aamodt, A. and Plaza, E. (1994) 'Case-Based Reasoning:

Foundational Issues, Methodological Variations, and

System Approaches', AI Communications, 39-59.

Aamodt, A., Sandtorv, H. A. and Winnem, O. M. (1998)

'Combining Case Based Reasoning and Data Mining -

A way of revealing and reusing RAMS experience ',

Safety and Reliability; Proceedings of ESREL ’98 ,

Balkena, Rotterdam, 1345-1351.

Amit, K. (2000) Artificial Intelligence and Soft Computing

Behavioral and cognitive modeling of the human

brain, 1

edition, CRC Press.

Bagnall, B. (2011) Intelligence Unleashed: Creating

LEGO NXT Robots with Java, 1

edition, Variant

Press.

Berners Lee, T., Fielding, R. and Frystyk, H. (1996) RFC-

1945 Hypertext Transfer Protocol -- HTTP/1.0, May,

[Online], Available: http://tools.ietf.org/html/rfc1945

[14 may 2012].

Blanzieri, E. and Ricci, F. (1999) 'Probability Based

Metrics for Nearest Neighbor. Classification and Case-

Based Reasoning', ICCBR '99 Proceedings of the

Third International Conference on Case-Based

Reasoning and Development, London, 14-28.

Cengeloglu, Y., Khajenoori, S. and Linton, D. (1994) 'A

Framework for Dynamic Knowledge Exchange

Among Intellingent Agents', Symposium Control of

the Physical World by Intelligent Agents, New

Orleans, LA, USA, 11-18.

Cunningham, P. (2009) 'A Taxonomy of Similarity

Mechanisms for Case-Based Reasoning', Knowledge

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

490

and Data Engineering, IEEE Transactions on, vol. 21,

no. 11, Nov, pp. 1532-1543.

Fielding, R.T. (2000) Architectural Styles and the Design

of Network-based Software Architectures, Irvine:

University Of California, Irvine.

García Armada, E., Jiménez, M. A., González de Santos,

P. and Armada, M. (2007) 'The Evolution of Robotics

Research: From Industrial Robotics to Field and

Service Robotics', IEEE Robotics and Automation

Magazine, vol. 14, no. 1, March, pp. 90-103.

Gerkey, B. P., Vaughan, R. T. and Howard, A. (2003) 'The

Player/Stage Project: Tools for Multi-Robot and

Distributed Sensor Systems', In Proceedings of the

11th International Conference on Advanced Robotics,

317-323.

Gindre, F., Trejo Pizzo, D. A., Barrera, G. M. and López

De Luise, M. D. (2010) 'A Criterion-based Genetic

Algorithm Solution to the Jigsaw Puzzle NP-Complete

Problem', Lecture Notes in Engineering and Computer

Science , 367-372.

Goodrich, M. and Tamassia, R. (2005) Data Structures

and Algorithms in Java, 4

edition, Wiley.

Google (2012) Android Developers, 14 may, [Online],

Available: http://developer.android.com/index.html

[14 may 2012].

IEEE (2009) IEEE Standards Association - 802.11n,

[Online], Available: http://standards.ieee.org/findstds/

standard/802.11n-2009.html [14 may 2012].

ISRD Group (2005) Introduction to Database

Management Systems, Tata McGraw-Hill Education.

json.org (2006) ietf.org, july, [Online], Available:

http://www.ietf.org/rfc/rfc4627.txt?number=4627 [14

may 2012].

Kneckt, J., Alanen, O., Zheng, C., Huovinen, T. and

Nihtila, T. (2011) 'IEEE 802.11ac: Enhancements for

very high throughput WLANs', International

Symposium on Personal Indoor and Mobile Radio

Communications (PIMRC), Helsinki, Finland, 849-

853.

Kolodner, J. (1992) 'An introduction to case-based

reasoning', Artificial Intelligence Review, vol. 6, no. 1,

march, pp. 3-34.

López De Luise, M. D., Barrera, G. and Iglesias, J. J.

(2010) 'An autonomous robot prototype using Concept

Learning model', Lecture Notes in Engineering and

Computer Science , Buenos Aires, 547-550.

LAAS-CNRS (2012) Open Robots, 13 august, [Online],

Available: http://www.openrobots.org/ [13 august

2012].

Lego (2012) Lego, 14 may, [Online], Available:

http://mindstorms.lego.com/en-us/Default.aspx [14

may 2012].

Mallet, A., Kanehiro, F., Fleury, S. and Herrb, M. (2007)

'Reusable Robotics Software Collection', Proceedings

of Second Int. Workshop on Software Development

and Integration in Robotics, Roma.

Mohan, Y. and Ponnambalam , S., (2009) 'An Extensive

Review of Research in Swarm Robotics', Nature &

Biologically Inspired Computing, 2009. NaBIC 2009.

World Congress on, Coimbatore, India, 140-145.

Nesnas, I. D., Simmons, R., Gaines, D., Kunz, C., Diaz-

Calderon, A., Estlin, T., Madison, R., Guineau, J. and

McHenry, M. (2006) 'CLARAty: Challenges and

Steps Toward Reusable Robotic Software',

International Journal of Advanced Robotic Systems,

vol. 3, no. 1, March, pp. 23-30.

Padhy, Rabi, Prasad, Patra, Ranjan and Satapathy , ,

(2011) 'RDBMS to NoSQL: Reviewing Some Next-

Generation Non-Relational Database's', International

Journal Of Advanced Engineering Sciences And

Technologies , vol. 11, no. 1, September, pp. 15-30.

Patterson, D. and Fox, A. (2012) Engineering Long-

Lasting Software: An Agile Approach Using SaaS and

Cloud Computing, Alpha Edition edition, Strawberry

Canyon LLC.

Portinale, L. and Montani, S. (2008) 'A Fuzzy Logic

Approach to Case Matching and Retrieval Suitable to

SQL Implementation', Tools with Artificial

Intelligence, 2008. ICTAI '08. 20th IEEE International

Conference on, Dayton, 241-245.

Reisbeck, C. K. and Shank, R. C. (1989) Inside Case-

Based Reasoning, 1

edition, Hillsdale: Lawrence

Erlbaum Associates.

Russell, R. and P, N. (2003) Artificial Intelligence: A

Modern Approach, 2

edition, Upper Saddle River:

Prentice Hall.

W3 Consortium (2012) W3C, 24 january, [Online],

Available: http://www.w3.org/XML/ [14 may 2012].

Walker, D. (2007) Similarity Determination And Case

Retrieval In An Intelligent Decision Support System

For Diabetes Management, Russ College of

Engineering and Technology of Ohio University.

Watson, I. and Marir, F. (1994) 'Case-Based Reasoning: A

Review', The Knowledge Engineering Review, vol. 9,

no. 4, October, pp. 355-381.

Multi-agentCase-basedReasoningInferenceEngineProposalforReusableRobotics

491