Abstract:
An effective collision avoidance logic should prevent collision without excessive
alerting. This requirement would be even more stringent for an
automatic collision avoidance logic, which is probably required by Unmanned
Aerial Vehicles to mitigate the impact of delayed or lost link issues.
In order to improve the safety performance and reduce the frequency
of false alarms, this thesis proposes a novel collision avoidance logic based
on the three-layer architecture and a real-time trajectory planning method.
The aim of this thesis is to develop a real-time trajectory planning algorithm
for the proposed collision avoidance logic and to determine the integrated
logic’s feasibility, merits and limitations for practical applications.
To develop the trajectory planning algorithm, an optimal control problem
is formulated and an inverse-dynamic direct method along with a two
stage, derivative-free pattern search method is used as the solution approach.
The developed algorithm is able to take into account the flyability
of three dimensional manoeuvres, the robustness to the intruder state uncertainty
and the field-of-regard restriction of surveillance sensors. The
testing results show that the standalone executable of the algorithm is able
to provide a flyable avoidance trajectory with a maximum computation
time less than 0.5 seconds.
To evaluate the performance of the proposed logic, an evaluation framework
for Monte Carlo simulations and a baseline approach for comparison
are constructed. Based on five Monte Carlo simulation experiments, it is
found that the proposed logic should be feasible as 1) it is able to achieve
an update rate of 2Hz, 2) its safety performance is comparable with a reference
requirement from another initial feasibility study, and 3) despite a
0.5 seconds computation latency, it outperforms the baseline approach in
terms of safety performance and robustness to sensor and feedback error.
