• collision avoidance
• formation flight
• uav

The algorithm showcased in this work aims to provide collision avoidance capabilities
for single agents and swarms of UAVs in a simple and efficient way, also
accounting for minimum deviation and rerouting to intended trajectory and time
spent within a formation maximization, all while keeping safe distance from potential
threats and computational power requirements low. The algorithm also looks at
providing an interface compatible with existing autopilot systems to keep it suitable
for a real world implementation.
The work is focused on utilizing and enhancing existing methodologies. The case
of a single agent having to avoid multiple obstacles will be addressed first and then
later extended to formation flight. The system developed is capable of avoiding both
static obstacles and dynamic ones, like other aerial vehicles. A protection sphere
gets generated around the obstacles and an avoiding command is found and issued
to the autopilot.
The algorithm includes the capability of recognizing which obstacles are more
dangerous than others in order to solve crowded scenarios in which avoiding all of them with the same maneuver would be unlikely, by transforming the problem into
a sequential one, in which obstacles are accounted for in smaller groups. Threat level
of the obstacles gets accounted for through the use of collision times, that incorporate
both speed and distance into the formulation, effectively representing how much an
obstacle is dangerous.
Geometric methodologies are used to command the agent through the danger to
keep complexity and computation times low but still find an optimal solution to the
problem, minimizing deviation from the original path through airspeed and heading
changes.
The main innovation lies into the extension of the same methodologies implemented
for the single agent through combination with continuum deformation of
the swarm’s shape in order to minimize occurrences of formation breaking to avoid
incoming obstacles.
Monte Carlo simulations are being provided to validate the algorithm and showcase
its capabilities, both in the single agent against multiple obstacles case and the
multiple agents, meaning a swarm of drones, against one obstacle.