Cooperative Vehicle Target Tracking 22.5 Effective Coordination Schemes 511
Part B | 22.5
22.5 Effective Coordination Schemes
An important aspect of a cooperative target tracking
system is measuring its overall performance [22.70]. In
particular, connecting MoPs (and their probability density functions (PDFs)) with MoEs requires integration
of specific performances into overall effectiveness in
achieving objectives and goals (Fig. 22.10).
Simulations or standard reinforcement learning
methods [22.17] are not feasible for large parameter
spaces. The expert choice of attributes linking MoPs
and MoEs is a viable method, but is not rigorous enough
to test the effects of different coordination methods.
Complex nonlinear systems often show critical behavior: small changes in parameter settings lead to completely different behavior of the system. Renormalization methods have been used to analyze the criticality
of systems with many degrees of freedom [22.69, 71].
Renormalization can mathematically bridge between
bottom-up and top-down perspectives. Essentially, gradients in the system are analyzed from the critical point
in parameter space. The parameter space is, in fact, of
infinite dimension if beliefs have to be handled. This
means that the parameters are functionals, and it is
necessary to find independent subprocesses in order to
handle the overall complexity. Equality of processes has
been investigated and found to be challenging [22.72].
A sensitivity analysis to initial conditions is the key to
allow a qualitative evaluation of the effectiveness of
a coordination scheme. A quantitative evaluation becomes possible by high-fidelity simulation for a limited
number of parameter settings in small domains previously identified by the qualitative evaluation.
We divide the design process into two parts:
Creation of benchmark problems to calibrate and
standardize the mapping between MoEs and MoPs
is the understanding phase.
Constructing an efficient cooperative team by inserting coordination into layers of originally independently operating surveillance assets (and
thereby creating efficient and suffciently effective
functionalities) is the trading methodology.
22.5.1 The Understanding Phase
Benchmark problems are simulations characterized by
an unbiased initial condition for each team and positioned at a critical point in the parameter space. The
likelihood of one team prevailing over the other is initially equal. Benchmark problems are simplifications of
the real world that distil key aspects of a system to clarify and calibrate the underlying dynamics. Often the
simplifications allow analytic solutions against which
more comprehensive (but opaque) numerical models
can be tested for fidelity.
In benchmark problems for the evaluation of the
effectiveness of coordination methods, the critical design aspect is the transition from one to many agents
as a means to achieving the objectives. For example,
in multistatic surveillance, cooperative target tracking
enables the detection and tracking of targets hidden
in clutter by increasing the effective antenna aperture
and thus narrowing the focus and increasing the signalto-noise ratio. In the detection of anomalous behavior,
cooperative sensing synchronized with a global observation network enables the construction of a time
dependent, multidimensional vector whose magnitude
and direction can reveal a target’s hidden process. These
two scenarios share the common features of multisensor
data fusion and a smart adversary.
Approaches to generating benchmark problems
vary:
The heuristic of looking for independent subprocesses or functionalities. Numerical simulations
governed by analytical considerations on the importance of these functionalities gives as a by-product
the formerly unknown functional connections between components in the entire process. This functional relationship represents the link between the
a-priori MoPs and the posterior MoEs.
A game scenario constructed by experts. Idealistically, the game is set up such that each side has an
equal chance of winning at the start. Alternatively,
the initial conditions can be deliberately biased to
test specific hypotheses. Fundamental to target detection and tracking problems is the process of
minimizing uncertainty. Teams that best succeed in
such minimization (through cooperative behavior)
will have a better chance to win the game.
22.5.2 The Trading Methodology
In the understanding phase explicitly separated (independent) agents or functionalities are generated. The
trading methodology starts with uncoordinated independent layers of these agents. From this starting point
with a specific position in the effectiveness-efficiency
plane and a guaranteed minimal effectiveness the risk
harming the effectiveness of the system by inserting
dependencies is traded with the predicted gain of efficiency by inserting these dependencies. For example,
using the same acoustic sources in the layered defence
Part B | 22.5
22.5 Effective Coordination Schemes
An important aspect of a cooperative target tracking
system is measuring its overall performance [22.70]. In
particular, connecting MoPs (and their probability density functions (PDFs)) with MoEs requires integration
of specific performances into overall effectiveness in
achieving objectives and goals (Fig. 22.10).
Simulations or standard reinforcement learning
methods [22.17] are not feasible for large parameter
spaces. The expert choice of attributes linking MoPs
and MoEs is a viable method, but is not rigorous enough
to test the effects of different coordination methods.
Complex nonlinear systems often show critical behavior: small changes in parameter settings lead to completely different behavior of the system. Renormalization methods have been used to analyze the criticality
of systems with many degrees of freedom [22.69, 71].
Renormalization can mathematically bridge between
bottom-up and top-down perspectives. Essentially, gradients in the system are analyzed from the critical point
in parameter space. The parameter space is, in fact, of
infinite dimension if beliefs have to be handled. This
means that the parameters are functionals, and it is
necessary to find independent subprocesses in order to
handle the overall complexity. Equality of processes has
been investigated and found to be challenging [22.72].
A sensitivity analysis to initial conditions is the key to
allow a qualitative evaluation of the effectiveness of
a coordination scheme. A quantitative evaluation becomes possible by high-fidelity simulation for a limited
number of parameter settings in small domains previously identified by the qualitative evaluation.
We divide the design process into two parts:
Creation of benchmark problems to calibrate and
standardize the mapping between MoEs and MoPs
is the understanding phase.
Constructing an efficient cooperative team by inserting coordination into layers of originally independently operating surveillance assets (and
thereby creating efficient and suffciently effective
functionalities) is the trading methodology.
22.5.1 The Understanding Phase
Benchmark problems are simulations characterized by
an unbiased initial condition for each team and positioned at a critical point in the parameter space. The
likelihood of one team prevailing over the other is initially equal. Benchmark problems are simplifications of
the real world that distil key aspects of a system to clarify and calibrate the underlying dynamics. Often the
simplifications allow analytic solutions against which
more comprehensive (but opaque) numerical models
can be tested for fidelity.
In benchmark problems for the evaluation of the
effectiveness of coordination methods, the critical design aspect is the transition from one to many agents
as a means to achieving the objectives. For example,
in multistatic surveillance, cooperative target tracking
enables the detection and tracking of targets hidden
in clutter by increasing the effective antenna aperture
and thus narrowing the focus and increasing the signalto-noise ratio. In the detection of anomalous behavior,
cooperative sensing synchronized with a global observation network enables the construction of a time
dependent, multidimensional vector whose magnitude
and direction can reveal a target’s hidden process. These
two scenarios share the common features of multisensor
data fusion and a smart adversary.
Approaches to generating benchmark problems
vary:
The heuristic of looking for independent subprocesses or functionalities. Numerical simulations
governed by analytical considerations on the importance of these functionalities gives as a by-product
the formerly unknown functional connections between components in the entire process. This functional relationship represents the link between the
a-priori MoPs and the posterior MoEs.
A game scenario constructed by experts. Idealistically, the game is set up such that each side has an
equal chance of winning at the start. Alternatively,
the initial conditions can be deliberately biased to
test specific hypotheses. Fundamental to target detection and tracking problems is the process of
minimizing uncertainty. Teams that best succeed in
such minimization (through cooperative behavior)
will have a better chance to win the game.
22.5.2 The Trading Methodology
In the understanding phase explicitly separated (independent) agents or functionalities are generated. The
trading methodology starts with uncoordinated independent layers of these agents. From this starting point
with a specific position in the effectiveness-efficiency
plane and a guaranteed minimal effectiveness the risk
harming the effectiveness of the system by inserting
dependencies is traded with the predicted gain of efficiency by inserting these dependencies. For example,
using the same acoustic sources in the layered defence