Aerospace and Electronic Systems Magazine September 2016 - 31

weather condition, which may not be realistic in worst case collision with uncertainties. The large range requirement also indicates
its insensitivity to GPS position errors (5 m, or 1.3%). Therefore,
collaborative SAA is a more feasible choice. Indeed, one can observe that the emergence of the small UAS industry today stimulates the rapid growth of small-size ADS-B technology. Examples
are XPS-TR from Sagetech [24] (100 g and 12 W max power) and
PING2020 from uAvioni [28]. In this article, we conservatively
assume that the vehicles of interest are equipped with collaborative SAA. Therefore, the collision avoidance algorithm must be
capable of handling communication delays.
Given the collision detection mechanism, what remains is the
collision avoidance algorithm, which constitutes the contribution
of this article. The safety controller generates optimal control actions when collisions are possible. The algorithm should be able to
incorporate bounded uncertainties such as communication delay,
wind disturbance, sensor uncertainty, and vehicle dynamics. However, the uncertainties are vehicle and scenario specific, and their
numerical values are not the focus of this article. Our simulation
only shows a typical scenario, indicating our algorithm can handle
these types of uncertainties. We want to address the following scenario: a manned helicopter ambulance equipped with (ADS-B)Out is flying at a low altitude to pick up a patient, when a quadrotor
UAS with (ADS-B)-In is delivering packages autonomously. The
quadrotor can receive real-time traffic information of the helicopter. If collision is possible, the quadrotor would avoid the manned
helicopter. The avoidance maneuver is performed exclusively by
the quadrotor UAS.
In the given drone-helicopter scenario, the current right of way
procedure would require to stop all drone activities until the helicopter completes its mission. However, this procedure is highly
inefficient and costly because quadrotors consume a lot of energy
when hovering. In addition, the pilot error issue is not addressed.
In a safety critical system, the proposed safety controller provides
the last safety guarantee when everything in the current procedure
fails.

LITERATURE REVIEW
FORMULATION ASSUMPTIONS
Collision avoidance designs entail assumptions on cooperation
between aircraft. The most conservative strategy is to set up the
problem as a noncooperative pursuit evasion game [29], in which
SEPTEMBER 2016

a helicopter is chasing a quadrotor. It would be ideal to permit the
helicopter to do whatever it likes, but the difference in vehicle capability shows this to be infeasible. A typical maximum speed of
commercial helicopters is 160 mph, or 70 m/s. For a quadrotor, the
maximum cruise speed is about 65 km/h, or 30 m/s. If a helicopter
is deliberately chasing a quadrotor, the quadrotor would eventually
be caught, regardless of its avoidance efforts.
We focus instead on a semicooperative strategy, in which the
responsibility for avoidance is placed on the quadrotor. In other words the quadrotor gives way to the helicopter. This is also
Google's current proposal [4]. We capture this in two specific assumptions.
The quadrotor initiates horizontal avoidance maneuvers exclusively.
The helicopter maintains almost constant heading, and does
not increase speed once the drone initiates collision avoidance.
The first assumption preserves the "freedom of sky" for helicopter pilots. Horizontal maneuvers are preferred for two reasons.
First, the helicopter produces unpredictable strong vertical vortex
rings [30], seriously affecting the controllability of the quadrotor in vertical maneuvers. To account for the vortex uncertainty, a
much larger safety region is necessary. Second, losing altitude in
low-altitude flights increases the chance of hitting obstacles. Third,
vertical maneuvers consume more power than horizontal ones. The
second assumption enables us to take advantage of relative kinematics in modeling. It is conservative if we assume the helicopter
pilot reacts rationally, meaning that he could only facilitate the
avoidance maneuver when the quadrotor reacts.

COLLISION AVOIDANCE
Aircraft collision avoidance algorithms have a rich history in the
literature, and are generally referred to as the free-flight problem.
There are geometry and probability based approaches [31], [32],
as well as optimal control based methods [33]-[36]. Our goal is
to reduce the side effect of avoidance as much as possible, which
implies optimality.
The first attempt on addressing the collision avoidance problem in an optimal manner is presented in [35], [36]. The avoidance
problem is formulated as a path planning problem, in which the optimal path could be found by solving a mixed integer programming
problem. The forbidden zone is set by rectangular constraints for
simplicity. However, this is a centralized approach, which requires
avoidance actions from both vehicles. In addition, solving mixed

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine September 2016