Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
5th
International Conference on Informatics in Control, Automation and Robotics - ICINCO 2008, May 11 - 15, 2008, Funchal, Madeira,
Portugal.
Abstract: Mobile embedded systems belong among the typical applications of the distributed systems control in real
time. An example of a mobile control system is the hydrogen powered prototype car control system. The
proposal and realization of such distributed control system represents a demanding and complex task of real
time control for track optimizing with minimal fuel consumption. Design and realization of distributed
control system, mention above, is prepared to realize as a complex laboratory task. Control system software
using of multi-agent technology with dynamic mutual negotiation of mobile system parts. This task allows
in a form of control system for prototype race car modelling of distributed control system. The real
hardware and software model is also important motivation for extended study.
30 10
25
5
20
0
15 Time [s] 1900
Instant velocity [km/h] Average velocity [km/h] Fuel cell voltage [V]
10 Fuel cell current [A] Motor current [A]
5
Figure 5: Testing run 2.
0
720
Time [s]
Instant velocity [km/h] Average velocity [km/h]
Fuel cell current [A]
Fuel cell voltage [V]
Motor current [A] 2.3 Bio-telemetry system
Figure 4: Testing run 1. The vehicle is also equipped by bio-telemetry
system that makes possible to monitor biological
functions of the pilot. The embedded portable For demand of design, a classification of agents
telemetry system of biological parameters serves for was implemented (Figure 6). The classification is
reading and wireless data transfer of measured focused in external behavior of agent, his
values of selected biological parameters to far communication with neighborhood and at reactions
computer. at external stimulations (Figure 7). We divide here
The telemetric system can be used for real time the agents in proactive and reactive agents.
monitoring of the basic life functions of race driver.
The driver has to be very concentrated and the
reactions of driver have to be very quick. The
telemetry system provides better control of physical
and psychical condition of driver during race. It is
possible to analyze effect of a stress situations, high
speed and high psychical stress on the race driver
during the race and crisis situation, more precisely
during high physical and psychical stress.
For biophysical monitoring were chosen these
parameters:
• Electrocardiography – ECG
• Pulse frequency
• Oxygen saturation – SpO2 Figure 6: Schema of classification of the agents.
• Body temperature
• Outside temperature
• Respiration frequency
3 MULTI-AGENT CONCEPT OF
CONTROL SYSTEM
The track passage optimization task of the laboratory
car with minimal energy consumption in real time is
quite complex.
Range of inputs and outputs of the control
system, communication flows and safety of Figure 7: Schema of classification of the agents
operation require the adaptability at occurred (Nwana classification) according to primary
situations and environment changes – strategy attributes.
control.
Multi-agent systems (MAS) increase flexibility, Considering internal agent behavior, we divide
efficiency and provide learning capability of new agents at process and deliberative agents.
reactions and behavior. In comparison with classic Proactive agent is an entity generating
technologies the learning is a „new feature“. requirements and communicating with other agents
Contemporary, a lot of methodologies, in order to accomplish own targets. Reactive agent –
technologies, tools or standards at theoretical level an entity that executes actions immediately and
and level of concrete implementations are available explicitly at base of stimulus received from
(e.g. AUML, FIPA standard, ACL or KQML neighborhood. Process agent is an entity with pre-
languages, development environments JADE, defined internal behavior. The routes are defined.
ZEUS, JACK or Cougar). Deliberative agent is an entity that has available
Among basic expected properties of proposed knowledge about itself and his neighborhood. At the
MAS belong a strategic, targeted system behavior, basis of the knowledge, it creates plans for
robustness and adaptability at environment changes. accomplishment of own targets. The targets are
This can be provided by decentralization of defined.
control activities in the control system, by Strategic control includes number of
distribution of functions and by modularity based on components. The related subject should know what
fundamental elements – agents (Srovnal, V., result is preferred. It should be able to react at
Pavliska, A., 2002). intentions and actions of other participants of the
situation. It should be capable for cooperation • Layered control. Agent is described by number
searching and compromise where it is rational. layers of abstraction and complexity.
Basic subject of strategic control can be called • Layered knowledge base.
an agent. • Bottom-up activating
The agent is software and/or hardware entity
• Top-down execution.
created with the aim of autonomous solving of tasks
Agent is connected with environment through
with respect at environment in which they occur.
interface with sensors, actuators and communication
His functionality – internal action depends at
module. Control is allowed through layers at three
communication with neighborhood (e.g. with
levels: reactive layer, layer of local planning, and
sensors, actuators or other agents).
layer of cooperative planning. They are use
Functionality is therefore given by his tasks,
information from knowledge bases (“world” model,
targets, behavior and states.
“mental” model and “social” model), (Garani, G.
An intelligence of the agent represents often the
and Adam, G., 2006).
capability of conformity, development of
Reactive layer is responsible for adequate
interactions with neighborhood, quick learning, data
reactions at the stimulations from environment that
accumulation and self analysis.
require immediate reaction and execution of called
procedures from local planning layer. Fundamental
3.1 MAS structure description characterization of such layer is:
• Use of effective algorithm of compare with
The higher level of control system is represented by patterns of behavior. Serve to pick-out of the
a personal computer. In the PC the signal from
actual situations.
differentially GPS positioning system may be
entered, which represents the relative coordinate • Situation description for timely actual reactions
system of environment – allow the precise of the at received stimulus.
position of the race car on the circuit. At the output • Hard-wired links. Recognized situations are fix-
is connected GPRS communication modem which connected with targets for reactive behavior.
transmits commands for race car. Immediate execution of program actions.
The algorithm of agent’s cooperation was • Solution request of situations not–corresponding
proposed with the control agent on a higher level. with couples situation-action are transmitted in
The control agent determines the required behavior local planning layer.
of the whole control system as the response to the • Execution liability is coming from local
dynamic behavior of car and to the one‘s own global planning layer activate procedures of reactive
strategy in the task and knowledge about the last layer patterns of behavior.
situations, which are saved in the database. The Some situations can be not solved by execution
agent on a higher level controls the other agents of template action like an answer to stimulation from
(Srovnal, V., Horák, B. and Bernatik, R., 2004). environment only, but they require certain level of
The separate task is the transformation which deliberation. A function of plans creation for solving
converts the digital data position into the object of the targets performs the layer of local planning.
coordinates (car position on the circuit) which are Local planning layer have such fundamental data
saved in the database of the circuit. This database is structures:
common for all agents in the control system. Each • Targets – state sets. Sets are characterized by
agent sees actual the whole data and is capable of attributes that are fulfilled at reaching targets.
controlling its behavior in a qualified way. The basic
• Planning – planning from second principles.
characteristic of a control algorithm of a subordinate
agent is the independence on the number of decision Sets of plans are defined before in data structure
making agents for car on the circuit. – plans library. Mapping of target sets to plans
Agent system has a common goal, to control of library is existed. For each target is possible to
the car during race with optimizing - minimizing of assign the plan for its reaching.
fuel consumption and control of critical speed. For • Plans library – contain the plans for reaching of
successful assertion of one’s own race strategy the agent targets.
extraction and knowledge of changeable • Scheduling – secure the timely limited plans
environment and learning capabilities are very stratification. Be created the plan schedules like
important. the step sequences, to execute.
Main architecture of such hybrid agent system is
characterized via:
3.2 Cooperative planning layer a GPS station. In the event that this limit is
exceeded, the frame is cut out and the control quality
A basic control cycle of cooperative planning layer may be decreased.
is creation, interpretation, decision making and The main possibilities of algorithm adjustment
execution of local plans. are as follows:
In first phase the reports from nearby layers are • Dynamic control in the control and decision
processed. Reactive layer sends requests to solve module of a control agent.
new task or status of executed behavior templates. • The control and decision modules and
Schedules of active plans are actualized. communication protocol of the decision agents.
Subsequently the status from reactive layer executed • The strategy of planning in the control model of
procedures is checked. the action agent.
In case of successful procedures finalization the
• Learning of a race strategy and using the
plan is erased from accumulator. Reports from
extraction results for decision rules generation as
highest layer are related to creation or cancellation
of commitment for the plan execution at local base a part of the rules decision database of a decision
or plan evaluation. In case of plan execution request agent.
or his cancellation the accumulator of active plans is
actualized.
The plan availability is a result of difference of ACKNOWLEDGEMENT
his relative value for the agent and his costs for
execution. The plan value is derived from target The Grant Agency of Czech Academy of Science
value that is possible reach by plan. The plan costs supplied the results of the project No. p.
are determined by function that assigns for every 1ET101940418 with subvention.
plan a real number calculated at basis of his The paper was also supported by KONTAKT
fundamental action costs according to specific rules. CZ-GR Project No. 7-2006-32 – “Communication
Nets in Distributed Systems”.
3.3 Multi-agent planning problem
In gradual hierarchy of targets may be not-realizable REFERENCES
target without coordination with other agents
appears. Purpose of multi-agent planning is in Srovnal, V., Pavliska, A., 2002. Robot Control
“unlock” such target through build of common plan. Using UML and Multi-agent System. In:
Unlock such target mean not explicit his reach. Be Proceeding 6th World Multiconference SCI
sufficient, if can agent continue, after execution of 2002. Orlando 2002, pp.306-311
common plan, in tracking of own local plans. Srovnal, V., Horák, B., Bernatik, R., 2004. Strategy
extraction for mobile embedded control
3.4 Transformation of common planning systems apply the multi-agent technology.
to the individual plans Lecture notes in computer science, Vol. 3038.
Springer-Verlag, Berlin Heidelberg 2004, pp.
Common plans contains actions of whole agents 631-637.
without synchronizations, are not direct executable. Garani, G., Adam, G., 2006. Qualitative Modelling
Are important, to transform common plans in to of Manufacturing Machinery. In: Proceedings
executable plans for individual agents. of the 32nd IEEE Conference on Industrial
Electronics, Paris 2006, pp. 3591-3596
4 CONCLUSIONS
The algorithm of the control system should be
proposed in a way so that it would ensure the
requirements for the immediate response of control,
so that the system of race car would be controlled in
real-time. That is why, it is very important so that
the algorithm for critical speed and fuel consumption
would be optimized. The system response should be
shorter than the time between two data frames from