|
.
Taxonomy Multi-agent systems Verification
and Validation
_____________________________________________________________________________________
Overview
The
demand for autonomy leads to consideration of increasingly complex systems
with ever more demanding performance specifications, and to mathematical
representations beyond time-driven continuous linear/ nonlinear systems, to
event-driven and to hybrid systems. Moreover, this autonomy quest pushes
forward interdisciplinary research in areas at the intersection of control,
computer science, networking, driven by application needs in physics,
chemistry, biology, finance.
_____________________________________________________________________________________
Taxonomy
Operating autonomy
(a)
the operator closes the numerical loop via control laws (e.g.
heading, slope, speed, altitude...)
(b)
operator monitored execution (e.g. waypoints are defined; the
operator checks whether the waypoints are correctly reached and deals with
failures), i.e. the whole decision loop is acted by the operator.
Decisional autonomy
(a)
includes the abilities of producing the task plan and supervising
its execution, while being reactive to events from the lower levels (e.g. waypoints
are recalculated autonomously if forbidden areas appear in the course of
the mission; the operator is highly involved within the decision loop, i.e.
inputs area features, checks whether recalculated waypoints are OK and
deals with other failures);
(b)
autonomous
situation assessment and replanning are performed
(the operator may close the decision loop when needed and when possible,
i.e. when communications are available).
_____________________________________________________________________________________
Autonomy in the context of multi-agent systems
Agents are
active, persistent software components, which perceive, reason, act,
communicate, and are capable of autonomous action in order to meet
their design objectives.
Absolute
autonomy
An absolutely
autonomous agent is one which has a freedom to choose any actions it likes.
Absolute autonomy is not a desirable feature of agent systems, because
useful agents ordinarily have some function, which constrains them.
Social
autonomy
Coordination with others reduces the
autonomy of individual agents. For example, by choosing to queue at a
bus-stop, an individual gives up some portion of its autonomy (its freedom
to get on the bus first), in order to coordinate with others attempting to
board the bus. In this way, the good of the whole is maximised at the
expense of the individual. Social autonomy is the case where an agent
attempts to coordinate with others where appropriate, but displays autonomy
in its choice of commitments to others (e.g. in making the decision to join
the queue).
Interface
autonomy.
To perform
useful functions, agent autonomy is typically constrained by an API
(application programming interface). Interface autonomy describes the level
of autonomy hidden behind the API of an agent - what the agent can choose
to do subject to obeying the API. It is therefore autonomy with respect to
internal design.
Execution
autonomy
The extent to
which an agent has control of its own actions is its level of execution
autonomy. This flavour of autonomy arises because an agent which is
controlled to some extent by a user or other process may appear to be
autonomous to other agents, but is clearly less independent than an
uncontrolled agent. An example of the constraint of execution autonomy is
an e-commerce agent which requests verification from a user before
proceeding to complete a transaction.
Design
autonomy
The extent to
which an agent design is constrained by outside factors is described by
design autonomy. For example, communication with other agents may require
an ability to represent beliefs, or for communications to be implemented in
a specific language. The design is therefore constrained by these
requirements.
_____________________________________________________________________________________
Verification and Validation of Autonomous Systems
Autonomy models have an abstract view of the system
they control. The nature of this abstraction, and the corresponding modeling paradigms, can vary according to the needs of
each application.
The abstraction method used has a critical influence
on the complexity and tractability of the verification task:
·
Discrete models describe
the system in terms of transitions between states, where a state describes
a stable configuration of the system for some duration of time. Discrete
models are usually based on some form of automata.
·
Real-time models can
specify time durations between events, whereas un-timed discrete models
only address the order in which transitions occur.
·
Continuous models represent
the continuous change in the system with respect to time, using some form
of differential equations.
·
Hybrid models mix
continuous changes and discrete transitions.
|
