IEEE Robotics & Automation Magazine - September 2022 - 128

function to control the generated velocity in different
directions. The framework is validated in simulation and
on a real dual-arm system. We show that we can precisely
toss objects within a workspace of .. .
02 04 m2
#
Moreover,
we show that the algorithm can adapt on-the-fly to changes
in object location.
Background on Bimanual Robot Systems
Swift robot manipulation of objects in an unstructured and
dynamic environment is crucial for industry. In logistics, for
instance, the booming of e-commerce and its related challenges
have increased the need to speed up the pace of pickand-place
operations. In current applications, robots usually
pick up and release the products with almost zero contact
velocity. One solution to speed up the process is to move from
this quasi-static approach
toward a dynamic one
where robots are allowed
to grab and release products
with nonzero contact
velocities. This can
be achieved by designing
robot controllers that
are aware of induced im -
pacts. Planning impact is
challenging due to noise
in perception and control.
In that regards, one
important aspect is to
generate motion robust
to imprecision as to when and how much impact is incurred.
Moreover, the motion should be robust throughout the task
from grabbing with impact to release, be it by placing, handing
over, or tossing of the object.
In this article, we consider the problem of grabbing and
releasing an object in one swipe with a dual-arm robotic system.
The desired manipulation task is motivated by the need
to perform fast pick-and-place or pick-and-toss operation in a
depalletizing context (see Figure 1). Such repetitive and physically
demanding work is usually performed by humans for
the lack of similarly fast, precise, and robust bimanual robot
systems. The bimanual tasks envisioned here extend the
complexity of the control problem as it requires, in addition to
controlling for impact, enforcing the coordination of the two
arms. A poorly coordinated system, where one arm reaches
the object before the other, would lead to uncontrolled impact.
Bimanual pick-and-toss requires precise coordination
Unlike classical approaches
that grab the object with
quasi-zero contact velocity,
the proposed approach is
able to grasp the object
while in motion.
before and after contact with the object. Coordination at contact
ensures that the object is not set off balance at pickup.
Once contact is established, the resulting interaction forces
need to be controlled to ensure a stable grasp and to induce
the desired velocity on the object for proper tossing. Controlling
robustly coordinated motions of multiarm systems opens
the door to a larger variety of tasks. Besides depalletizing, this
could include manipulations that are too complex or heavy
for a single robot and require two or more robotic arms.
Some applications could be the fast picking up of open trays
or cases, fast picking up of luggage from an airport's conveyor
belt, and so forth.
Dual-arm control has been extensively studied; see, for
instance, [1] for a review. Several methods have been proposed
to coordinate multiple robots' motion [2]-[4], to control
simultaneously robots' motion and forces [5]-[8], and to
optimize contact forces at run time, using two robot manipulators
[9], or a humanoid robot in [10], using quadratic programming
(QP). All previous cited works assume that the
object is already grasped by the robot and focus on the postcontact
manipulation phase. The free motion phase and its
transition toward the contact phase and after the contact
phase for placing and tossing were not considered. A first
approach to smoothly coordinate two robotic arms in freespace
motion and when making simultaneous contact based
on DS was proposed in our group [11]. This approach uses a
virtual object to constrain and coordinate the motion of the
robots. However, it did not control for force at contact. Our
group extended this further in [12] to propose a DS that
could generate both motion and forces for the dual-arm system.
Yet these two approaches still assumed quasi-static grab
with end-effector velocities vanishing as they approach the
grasping points on the object. Moreover, they ignored the
problem of tossing or placing the object once in hand.
Recently, an impact-aware controller formulated within a
(a)
(b)
Figure 1. Illustrations of a dual-arm manual and robotic pick-andplace
operations. (a) A human dual-arm grabbing and placing
objects in a palletizing task within a Vanderlande facility. (Source:
Photo courtesy of Vanderlande.) (b) A pair of two real and
simulated real KUKA LBR IIWA robots grabbing an open box (top)
and an object containing a small moving object inside (bottom).
128 * IEEE ROBOTICS & AUTOMATION MAGAZINE * SEPTEMBER 2022
QP framework was presented in [13] and was applied to dualarm
grabbing of a box with a contact velocity of
015 m/s.
.
This work offers a powerful approach to control impact with
nonzero contact velocities. The QP scheme, however, relies on
a planning of the grasping motion, making it less robust to
imprecise perception or dynamic changes of the object's pose.
In our work, instead, we use as motion generators autonomous
DSs for their fast and time-independent replanning
IEEE Robotics & Automation Magazine - September 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2022
