IEEE Robotics & Automation Magazine - December 2018 - 78

transport the object were distributed to the attached multirotors, and each agent was controlled in a decentralized way. In
[2], multirotors were connected to an object with spherical
joints mounted on the object, allowing each multirotor to
rotate freely with respect to the hinge. Another cooperative
aerial manipulation system used cable suspension to transport an object [3]. This system was proven to be differentially
flat, which allowed dynamic decoupling of the cooperative
system into a set of individual dynamics.
A robotic arm with multiple degrees of freedom (DoF) has
also been used for constructing an aerial manipulator.
Because the movement of the arm as well as the manipulated
object causes considerable inertial effects, this type of aerial
manipulation requires more complicated control strategies.
Particularly for cooperative cases, the internal forces from the
object affect all the aerial manipulators, so special attention is
required. In [4], impedance control was used with the aid of
six-axis force/torque sensors attached at the end effector of a
robotic arm to compensate for the end effectors' internal and
external disturbances. In contrast to [4], an adaptive controller with an online estimation of the mass of an unknown
object was proposed in [5]. Based on the estimation, internal
forces exerted on the object were regulated without additional
sensors. In a hierarchical control framework proposed in [6],
the desired net force required to transport the object was initially computed and distributed to the desired contact force of
each aerial manipulator.
An effort was made to decouple the complex dynamics of
a cooperative system into a set of perturbed individual
dynamics. This approach is attractive because some advanced
control algorithms for a single aerial manipulator [7]-[9] can
be directly employed for cooperative tasks. However, exact
decoupling of the dynamics requires precise information
about the inertial properties of the object and the grasp mechanism between end effectors and object. Such information is
generally not known, and the object's reaction force was treated as a disturbance in an aerial manipulator [10]. The authors
designed a disturbance observer (DOB) to estimate and reject
the external disturbances. The DOB-based robust controller,
initially developed to control a single aerial manipulator proposed in [11], has been extended to a cooperative manipulation task in [10]. In [12], planning within the available
computation time was investigated for cooperative assembly
tasks using aerial vehicles.
Research Perspective in Aerial Cooperation
In the previous section, we discussed existing controllers that
allow aerial manipulators to follow the desired trajectories.
However, despite the importance of path planning in collaborative aerial manipulation, relevant research seems to be relatively incomplete.
Our objective with cooperative aerial manipulation planning is to minimize energy-related cost functions along the
trajectory and provide time-affordable computation of the
trajectory. Energy efficiency is an essential concern in aerial
manipulation due to the short flight time. In the same vein,
78

IEEE ROBOTICS & AUTOMATION MAGAZINE

december 2018

we should choose the best pose for the additional robotic
arm used as the manipulation mechanism to minimize the
power consumption without risking safety. To ensure the
safety of a cooperative aerial manipulation system, the trajectory should be computed quickly enough to respond to any
contingency. For instance, the system should be able to
swiftly change the pose and avoid obstacles encountered
along the initially planned path. From such a perspective,
this article discusses possible approaches that can incorporate energy efficiency, obstacle avoidance, and safe cooperation, among other issues.
The main contribution of the framework proposed in [13]
is that fast computation is possible for reacting to unexpected
situations, even though all of the dynamic and kinematic
properties of cooperative aerial transportation are considered
during planning. The major improvements compared with
our previous work [13] are as follows:
1) We compare sampling-based planning and our learningbased planning for the aerial cooperation scenario in terms
of computational time and performance index.
2) A robust controller [5], [11] that improves the flight
performance of cooperative aerial manipulation has
been integrated.
3) With the improved flight performance, we validate our
synthesis through several experimental scenarios, including a considerable change in heading angle.
Conventional Planning Approaches
In this section, we investigate the existing motion-planning
methods and how these approaches can be implemented in
aerial cooperation.
Optimization-Based Motion Planning
Optimization techniques in motion planning offer the explicit
guarantee of local (or global) optimality for the planned trajectory with respect to a cost function. Moreover, several constraints can be considered in a unified way, ensuring the
dynamic and kinematic feasibility of the planned trajectory
for a cooperative aerial manipulation system.
Nonlinear programming (NLP) can be used in straightforward, optimization-based motion planning. One easy way
to formulate NLP for the aerial manipulation system is to
assume a continuous-time trajectory of a specific representation such as a polynomial, B-spline, or Bézier curve. The trajectory optimization problem is converted into a static
nonlinear optimization problem involving the coefficients of
the corresponding representation [14]. Based on this conversion, the optimization search space is confined to a span of
bases defined in the representation space. However, a simple
NLP is not appropriate for generating suitable trajectories
when the dynamics and constraints of the target system
become complicated. The main reason is that a solution satisfying complex constraints may not exist in the restricted
search space. Particularly for cooperative aerial manipulation, complex dynamics and constraints further reduce the
search space for planning.

IEEE Robotics & Automation Magazine - December 2018

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2018