ARGO - A Customized Jason Architecture for Programming Embedded Robotic Agents

Pantoja, Stabile, Lazarin and Sichman 2016 EMAS@AAMAS 2016, Singapore, 09/05/16 ARGO 1

ARGO: A Customized Jason Architecture for Programming Embedded Robotic Agents

1. Instituto de Matemática e Estatística (IME), Universidade de São Paulo (USP), Brazil2. Escola Politécnica (EP), Universidade de São Paulo (USP), Brazil

3. Centro Federal de Educação Tecnológica (CEFET/RJ), Brazil4. Universidade Federal Fluminense (UFF), Brazil

Laboratório de Técnicas Inteligentes - LTI

Carlos Eduardo Pantoja 3,4

Márcio Fernando Stabile Junior 1Nilson Mori Lazarin 3

Jaime Simão Sichman 2,1

III Workshop on Engineering Multi-Agent SystemsEMAS@AAMAS 2016

Singapore 09/05/2016


ARGO

The Argo by Lorenzo Costa

Argo was the ship that Jason

and the Argonauts

sailed in the search of the

golden fleece in Greek

mythology.


Outline

1. Introduction

2. Building Blocks: Jason / Perception Filters / Javino

3. ARGO

4. Case Study

5. Obtained Results

6. Conclusions and Further Work


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Motivation

MAS A robot is a physical entity, composed by customized hardware, sensors and actuators

How can we program and control a robot including reactive and goal-directed behaviours? .


BDI model

[http://www.inf.ufrgs.br/prosoft/bdi4jade]


Jason

[Bordini et al. 2007]


Motivation

Programming robotic agents using Jason has revealed to be a difficult task• Bottlenecks can occur

» high cost of processing perceptions» large intention stack is generated

• Integration with hardware is not implemented• Hence, the robot may not succeed !


Motivation

Javino [Lazarin and Pantoja 2015]• middleware for communication between Java and

microcontrolers (Arduino)• However, using several sensors may compromise the

robot execution time

Perception filters [Stabile Jr and Sichman 2015] • filters are able to improve Jason agent's performance

in a significant way


Motivation

Instead of taking into account all perceptions ....

MAS


Motivation

One can filter perceptions!

MAS


Objectives

ARGO provides a customized Jason architecture for programming embedded robotic agents• Javino + Perception filters

Layered robot architecture Experiments using a ground vehicle platform in a

real-time collision scenario Evaluations of filters impact


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Jason [1]

• AgentSpeak Interpreter [2]

[1] [Bordini et al. 2007] [2] [Rao 1996]


Jason [1]

• AgentSpeak Interpreter [2]

Most time-consumingprocesses

[1] [Bordini et al. 2007] [2] [Rao 1996]


Profiling

86% of total processing time


Profiling

99% of total processing time


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Perception filters

[van Oijen and Dignum 2011] • Integrating agents (2APL, Jadex and Jason) to

computer games;• Middleware responsible for perception filtering;• Interest Subscription Manager.

[Bordeux et al. 1999] • Extend AGENTlib with a perception mechanism;• Perception filter types:

» Range filter;» Field of view filter;» Type detector filter.


Perception filters

Example of Jason perceptions• List of annotated literals

temperature(right,36)temperature(back,38)light(left,143)distance(front,227)distance(right,30)


Perception filters

Example of our perception filter specification<PerceptionFilter> <filter>

<predicate>temperature</predicate> </filter> <filter>

<predicate>light</predicate> </filter> <filter>

<predicate>distance</predicate><parameter operator="NE" id="0">front</parameter>

</filter></PerceptionFilter>

distance(front,227)[source(percept)]


Perception filters

Example of filter change internal action• Name of file passed as parameter

+!carry_to(R)<− ! take (object, R); .change_filter(search); −object (r1); !!search(slots).


Perception filters



Perception filters



Perception filters

Changes in Agent class

public void buf(List<Literal> percepts) { if (percepts == null) { return; } int adds = 0; int dels = 0; long startTime = qProfiling == null ? 0 : System.nanoTime();

filter(percepts);

Iterator<Literal> perceptsInBB = getBB().getPercepts(); while (perceptsInBB.hasNext()) { ...


Perception filters

Changes in Agent class

private static void filter(List<Literal> percept) { if(currentObjective==null){ return; } Iterator<Literal> it = percept.iterator(); while (it.hasNext()) { if (remove(it.next())) { it.remove(); } }}


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Javino

Javino is a protocol for exchanging messages:• between low-level hardware and a high-level

programming language• double-side library for communication• provides error detection


Operation modes

Listen mode• only from hardware to software

AGENTsend a message in

every loopget when it

wants


Operation modes

Request mode• from software to hardware;• the hardware answers.

AGENTrequest a message

answer with a message


Operation modes

Send mode• from software to hardware;• the hardware executes an action.

AGENTsend a

messageexecute a low-level command


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



ARGO is:• a customized architecture for Jason• employs both Javino middleware and perception

filters » Javino provides a bridge between the intelligent agent

and the robots sensors and actuators» Perception filters act blocking specific perceptions in

runtime ARGO aims to be a practical architecture for

programming automated embedded agents using BDI agents

ARGO


It directly controls the actuators at runtime It receives perceptions from the sensors

automatically within a pre-defined time interval It enables changing filters at runtime It enables changing accessed device at runtime ARGO agents may communicate with others

Jason Agents It enables to decide when to perceive the real

world at runtime

ARGO overview


ARGO overview


Overview of Robot’s Architecture


Receiving percepts

Sensors capture raw data from the real world and

send them to the microcontroller employed.


Receiving percepts

In the firmware layer, raw data is transformed into

perceptions based on the AOPL chosen.


Receiving percepts

Javino is responsible for sending the percepts to the reasoning layer using serial

communication


Agent’s reasoning

The agent is able to reason with percepts coming

directly from real world and the MAS can be

embedded in single-board computers (Raspberry,

etc.) or a computer with USB interface


Executing an action

Agent deliberates and if an action has to be executed, an action

message using Javino is sent.


Executing an action

Javino sends the action message to the

microcontroller connected in the USB port described

in the message.


Executing an action

All possible actuator’s functions are programmed to be executed in response to serial messages coming

from Javino.


Executing an action

The actuator is activated.


Jason’s reasoning cycle with filters


ARGO’s reasoning cycle


Customized architecture



Customized architecture created to differentiate Argo

agents from common Jason’s

agents



Javino instance for each Argo agent.



Returns the ARGO agent’s Javino

instance.



The serial port from which the agent is receiving perceptions and executing

actions.



Defines if the agent has to perceive or not

the real world.



A time interval, in milliseconds, for the

next real world sensing



Function responsible for returning the perceptions

from the real world if:

i) the perceptions is not blocked;

ii) the time limit was reached;

iii) the agent is an ARGO agent



Responsible for filtering perceptions, as stated

before.



Changes in TransitionSystem class

public boolean reasoningCycle() {…ag.buf(this.realWorldPerceptions());…}



New realWorldPerceptions function

public List<Literal> realWorldPerceptions() {long perceiving = System.nanoTime();List<Literal> percepts = new ArrayList<Literal>();

if(((perceiving - lastPerceived) < this.limit) || this.blocked)return null;

lastPerceived = perceiving;

if (this.agArch.getArgo().requestData(this.agArch.getPort(), "getPercepts")) {String rwPercepts = this.agArch.getArgo().getData();String perception[] = rwPercepts.split(";");

for (int cont = 0; cont <= perception.length - 1; cont++) {percepts.add(Literal.parseLiteral(perception[cont]));

}return percepts;

}


Internal Actions

ARGO Internal Actions:• .limit(x)

» defines the sensing interval in milliseconds• .port(y)

» defines which serial port should be used by the agent• .percepts(open|block)

» decides whether or not to perceive the real world • .act(w)

» sends to the hardware an action to be executed by a microcontroller • .change_filter(filterName)

» defines the filter to constrain perceptions in runtime


Limitations Limit of 127 serial ports

• Due to limitation of USB Connection to one port at a time

• Avoids competition• It can be changed at runtime

Only ARGO agents can control devices• Common Jason agents do not have access to Javino

ARGO agents must be atomic• Cannot create more than one instance of the same agent


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Case study

• The robot configuration: • 4 distance sensors• 4 light sensors• 4 temperature sensors• 1 Arduino board • 1 Arduino 4WD chassis

• Initial distance of 2m from the wall• The robot moves at constant speed• The robot should stop before

achieving a specified desired distance from the wall


Evaluating the experiment

Experimental design guidelines defined by [Jain 1991] Essential terms: Response variable

• Processing time» from the moment the robot perceives the wall until it stops

• Final distance » from the position the robot stops to the wall

Primary Factors• Desired distance • Perception interval • Filter


Evaluating the experiment Essential terms:

• Levels» Values that a factor can assume

Factor LevelsDesired distance 40 cm 80 cm 120 cm

Perception Interval 25 ms 35 ms 50 ms

Filter No Filter Front Side Front Distance

• Replications» Three times for each experiment (81 experiments)


Evaluating the experiment

Desired distance

Initial distance

2 m


Filters

Front side filter<PerceptionFilter> <filter>

<predicate>temperature</predicate><parameter operator="NE" id="0">front</parameter>

</filter> <filter>

<predicate>light</predicate><parameter operator="NE" id="0">front</parameter>

</filter> <filter>




Filters

Front distance filter <PerceptionFilter> <filter>

<predicate>temperature</predicate> </filter> <filter>

<predicate>light</predicate> </filter> <filter>




Evaluating the experiment Agent code:



Set serial port COM8. Arduino device.



Set an interval of 25ms for perceiving

the real-world



Open the selected port to start receiving

percepts



Activates frontSide filter



Send a message to the microcontroller to move ahead



Keep moving ahead while

the perceived distance is

greater than the distance

limit



Stop when it perceives the

wall



Some additional

plans


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Perception Interval 20









Desired Distance 40 Desired Distance 80 Desired Distance 120

0

20

40

60

80

100

120

No filter Front Side

Front Distance

Fina

l Dis

tanc

e

Experiments

In all experiments, the robot

collided with the wall!!!












0

20

40

60

80

100

120


Front Distance

Fina

l Dis

tanc

e

Experiments

In some experiments, the robot

didn’t collided with the wall!!!

But it stopped closer to

wall compared to

the front distance

filter












0

20

40

60

80

100

120


Front Distance

Fina

l Dis

tanc

e

Experiments

In quite all the

experiments, the robot

didn’t collided with the wall!!!


Experiments

The use of the filter was important for obtaining a better response time

Factor Variation attributedDistance Limit (L) 1,415%Perception Interval (I) 0,165%Filter (F) 88,965%Interaction between L and I 0,525%

Interaction between L and F 3,715%

Interaction between I and F 0,265%

Interaction between L and I and F 1,725%

error 3,285%


Outline

1. Introduction


3. ARGO

4. Case Study

5. Obtained Results



Conclusions

The main contribution of ARGO is to offer an open architecture that enables Jason agents to integrate with hardware and to use perception filters• Reduction processing

It allows an agent to decide in runtime:• when to start or to stop perceiving• the interval between each perception• which filters to use


Further work

Different filtering methods Extending ARGO for multi-robot systems Testing ARGO in different domains Provide other hardware-side libraries

• PIC16F, Intel and STM32.


References

[Bordini et al. 2007] Bordini, R.H., Hubner, J.F., Wooldridge, M. Programming Multi-Agent Systems in AgentSpeak Using Jason. John Wiley & Sons Ltd., 2007.[Lazarin and Pantoja 2015] Lazarin, N.M., Pantoja, C.E. A Robotic-Agent Platform For Embedding Software Agents Using Raspberry Pi and Arduino Boards. In: Proc. 9th Software Agents, Environments and Applications School (WESAAC 2015), Niterói, RJ, Brazil, 2015.[Rao 1996] Rao, A.S. AgentSpeak(L): BDI Agents Speak Out in a Logical Computable Language. In: de Velde, W.V., Perram, J.W. (eds.) Proc. of the 7th European Workshop on Modelling Autonomous Agents in a Multi-Agent World (MAAMAW 1996). Lecture Notes in Artificial Intelligence, vol. 1038, pp. 42-55. Springer-Verlag, Secaucus. USA, 1996.[Stabile Jr. and Sichman 2015] Stabile Jr., M.F., Sichman, J.S. Evaluating Perception Filters In BDI Jason Agents. In: Proc. 4th Brazilian Conference on Intelligent Systems (BRACIS 2015), Natal, RN, Brazil, 2015.


Acknowledgements


END

THANKS FOR YOUR [email protected]

[email protected]@cefet-rj.br

[email protected]