IEEE Robotics & Automation Magazine - June 2012 - 21

But apart from appearance, versatility, and dexterity,
another aspect is more crucial: safety. As Isaac Asimov
stated in his first law in 1942 [36]: "A robot may not injure
a human being or, through inaction, allow a human being
to come to harm." In addition to applying sophisticated
strategies to prevent dangerous situations in advance, the
robot must also be capable of feeling contact forces so as to
react properly if a situation with physical human-robot
interaction occurs [2].
Whether in movies or in the press coverage, mainly
humanoid robots are shown when robotic systems are
addressed. From an engineering point of view, it is a big
challenge to coordinate such a large number of DoF simultaneously. In addition to humanoid robots such as Honda's Asimo [3], Robonaut 2 [4], and the HRP-2 robot [5], a
variety of wheeled systems has been developed: Rollin' Justin [6], Armar-III [7], Twendy-One [8], PR2 [9], to name
just a few examples. But regardless of the specific structure
of the system, the requirement of handling several objectives simultaneously is a common property. These range
from features such as precise task execution, collision
avoidance, and the compliance with physical constraints to
higher-level objectives (realizing desired postures or maintaining the manipulability).
Based on the operational space formulation [10],
many different methods have been developed for planning and the reactive control of such systems [11]-[13].
Multiple tasks are performed simultaneously on a biped
humanoid robot in a whole-body control framework
including issues such as the control of the center of mass,
obstacle avoidance, and posture control [14]. Brock and
Khatib [15] introduced the elastic strips framework that
allows executing previously planned motions in a
dynamic environment. They reactively adapt to changes
in the environment, e.g., when an obstacle is approaching
the manipulator. The majority of these control strategies
rests upon the design of artificial repulsive/attractive
potential fields [16]. However, having a large number of
DoF raises the question of a proper redundancy resolution. Especially when potential field-based strategies are
applied, the problem of local minima in the case of competing objectives is crucial. An early technique by Siciliano and Slotine [17] utilizes the null space projection to
derive joint velocities that execute a low-priority task
without disturbing any task with higher priority. Sentis
and Khatib proceeded similarly to realize a hierarchy of
behavioral primitives [11]. Another example for a
consistent installation of a hierarchy can be found in [18],
wherein a measure is imposed that indicates the feasibility of a task operating in the null space of a higher-priority task. That coefficient may then lead to a transition
changing the priority order in real time. To integrate unilateral constraints into such a hierarchy, Mansard et al.
proposed a control law based on a specific inverse operator so as to smooth the activation/deactivation process of
subtasks [19].

As this article is about reactive, dynamic mobile
manipulation, we have to define that term in the first
place. In this context, reactivity represents the ability to
locally react on unpredictable, unmodeled dynamics and
environments [16]. The word dynamic expresses the
motion characteristics of the mobile manipulator.
Motions are not executed slowly, but they are fast enough
such that dynamic effects have to be considered due to
their significant influence. In the literature, mobile
manipulation is mostly treated as a static problem to be
solved in the high-dimensional configuration space [20].
Dynamic effects are taken into account quite scarcely
[11], [14]. In this work, we incorporate the dynamics of
the system. Moreover, we do not consider physical constraints on the planning level [21] but handle them reactively by utilizing the redundant DoF.
This article integrates the newest results of the robotics
community on reactive, dynamic mobile manipulation
control in a consistent framework and gives solutions to
several open questions. The proposed framework allows to
demonstrate the methodologies on a highly complex
robotic system (see Figure 1) with torque control interface
at a high level of reliability and performance. The implementation in a 1-ms cycle comprises the simultaneous
consideration of nine reactive tasks that are integrated into
a hierarchy with two basic levels. A further subdivision of
these two levels is performed to specify the robot behavior
in greater detail. A newly developed passivity-based algorithm for reactive avoidance of self-collisions [22] is
presented and integrated into the whole-body control

Figure 1. Mobile humanoid Rollin' Justin of the German
Aerospace Center (DLR) with 51 actuated DoF [26].

JUNE 2012

IEEE ROBOTICS & AUTOMATION MAGAZINE

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