IEEE Robotics & Automation Magazine - June 2012 - 23

Trajectories
(TCP, Postures)

Collision Avoidance
Self-Collision Avoidance

Localization/
Planning/
Trajectory
Generation

Cartesian Impedance

Robot and Joint-Level Control

Mechanical End Stops
Base Singularity Avoidance

Redundancy
Resolution

tcmd

Torque
Control
t

Arm Singularity Avoidance
Torso Posture Control
Base Posture Control

Upper
Body

Admittance
Control

Velocity
Control

Mobile
Base

Xodo

Joint Damping

Figure 3. A controller architecture for dynamic whole-body motions with nine simultaneous tasks. The joint controllers of the robot
are fed by the redundancy resolution block that gets input from the planning layer.

and intuitive rules that are essential for a proper robotic
behavior. In this respect, we have drawn up four types of
basic requirements that should be met:
1) safety
2) physical constraints
3) task execution
4) posture primitives.
In our opinion, these four categories describe the
key aspects. Considering such a guideline for the selection of the involved tasks is not a novelty but an intuitive basis of many well-known whole-body control
approaches as [11].
Concerning the prioritization among these requirements, safety is usually located at the top end. In contrast, a posture primitive relates to a favored, though not
essential task as effort minimization [14] or a desired
posture [11]. Therefore, this aspect is at the lowest priority level and may be carried out if sufficient structural
redundancy is left. The placement of the remaining two
items in the list is more ambiguous. Although several
physical constraints are crucial to prevent severe damage of the manipulator (e.g., avoidance of hitting joint
limits), it might be reasonable to give higher priority to
the task execution in some cases. This applies, if the
manipulator is sufficiently redundant with respect to the
main task. Then the compliance with these physical constraints can be provided and the task execution does not
have to be interfered by those tasks that are often defined by
unilateral constraints. This leads us directly to the second
reason for an exchange of physical constraints and task execution within the hierarchy: The integration of unilateral,
physical constraints into the higher levels of a task hierarchy
causes additional problems in terms of discontinuities in the
control law [19]. A new solution to that problem for torque-

controlled robots is presented in the "Redundancy Resolution Concept" section.
Let us now return to the particular controller structure
depicted in Figure 3. The safety aspect is addressed by algorithms for collision avoidance with external objects and
self-collision avoidance. A more detailed discussion on this
topic is given in the "Ensuring Robot Safety" section. The
issue of physical constraints particularly depends on the
specific structure and characteristics of the considered system (see the "Complying with Physical Constraints"
section). In the case of Justin, physical limitations are
reached in singular configurations of the mobile base. The
design of a proper singularity avoidance is an appropriate
remedy. Apart from that, the existence of mechanical end
stops of the joints has to be taken into account. Task execution is realized by a Cartesian impedance at the TCPs,
which is described in the "Task Execution" section. The
last point in the list comprises additional posture behaviors
or posture primitives [11], respectively. The structural
redundancy of multi-DoF robots, such as Justin, can be utilized to realize specific head poses, desired torso orientations, or arm postures. We restrict to torso and base
postures as well as to nonsingular arm configurations in
the "Maintaining Manipulability and Realizing Desired
Impedances" section.
The total number of nine tasks is a particular choice
that we have made here. Actually, the number, selection,
and parameterization depend on many different aspects
such as the type of the main task, the structure of the environment, the desired dynamical behavior, and so forth. As
an example, we recall the mentioned physical limitations
of the mobile platform. They are only relevant for highly
dynamic motions with fast rotations. In the case of slow
tasks, they can be ignored. However, there are also
JUNE 2012

IEEE ROBOTICS & AUTOMATION MAGAZINE

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2012