曝光臺 注意防騙 網曝天貓店富美金盛家居專營店坑蒙拐騙欺詐消費者

to
model
collision
avoidance
problems when the aircraft are equipped with (nearly) perfect sensors and discusses one possible solution method. It discusses an extension to MDPs, called partially observable MDPs (POMDPs), that incorporates an observation model to account for state uncertainty. The section discusses the QMDP method for approximating the solution to a POMDP that leverages the solution of the underlying MDP.
Section
3
discusses
how
to
model
a
simpli.ed,
two-dimensional
collision
avoidance
problem
as
an MDP. In this simpli.ed collision avoidance problem there is a single intruder who is not equipped with a collision avoidance system. Neither aircraft maneuvers horizontally. The own aircraft is equipped with perfect sensors and the pilot responds deterministically to advisories issued by the system. The optimal solution to the problem is obtained using dynamic programming (DP). The section presents an example encounter and various performance assessment results comparing the DP-derived logic to TCAS in head-on encounters.
Section
4
quanti.es
the
robustness
of
the
DP
logic
of
Section
3
to
modeling
errors.
It
presents
the results of evaluating the logic on a three-dimensional white-noise model as well as a higher-.delity encounter model with a signi.cantly di.erent model structure. It introduces robust DP as a technique for making solutions less sensitive to particular choices of dynamic models. A state-based robustness analysis is performed to estimate the value of performance metrics starting from arbitrary states in the state space.
Section
5
extends
the
solution
method
of
Section
3
to
encounters
in
three
dimensions
and
al-
lows both aircraft to maneuver horizontally. To help mitigate the increased computational demands for
a
model
of
higher
dimensionality,
Section
5
introduces
an
approach
to
solving
the
problem
when
only some of the state variables are controllable. The approach involves decomposing the full prob-lem into controlled and uncontrolled subproblems, both of which are solved separately using DP algorithms.
Section
6
discusses
how
to
optimize
the
DP
logic
of
Section
5
using
probabilistic
pilot
response
models in which the future response of the pilot to advisories is uncertain. The section presents two di.erent pilot response models. Belief state .ltering is used to update the belief regarding the compliance of the pilot to advisories.
Section
7
discusses
how
to
optimize
the
logic
in
the
presence
of
state
uncertainty
due
to
sensor noise. The own aircraft is modeled as being equipped with a TCAS-like sensor that receives measurements of intruder range, bearing, and altitude. The performance of the QMDP method is compared against TCAS in simulation. The results of a robustness analysis are also shown.
Section
8
discusses
how
to
optimize
the
logic
when
the
intruder
is
equipped
with
the
same
collision avoidance system as the own aircraft. The advisories issued by the systems must be coordinated so that the aircraft do not accidentally maneuver into each other. When the state is fully observable, the problem can be modeled as a multi-agent MDP, which can be solved e.ciently using DP. With the introduction of sensor noise, the problem is transformed into a decentralized POMDP, which can be solved approximately using a number of strategies. Several coordination strategies are evaluated in simulation.
　
中國航空網 www.k6050.com
航空翻譯 www.aviation.cn
本文鏈接地址：Robust Airborne Collision Avoidance through Dynamic Programm(10)