IEEE Robotics & Automation Magazine - June 2020 - 170

Smart Collaborative Systems for Enabling
Flexible and Ergonomic Work Practices

HRC Technology

HRC Productivity

Human Monitoring,
Shared Intelligence,
Feedback, ...

Socioeconomic
Targets

HRC Interaction
Quality

HRC
Standardization

Occupational Health and Safety
Guidelines and Productivity and
Living Standards

Figure 1. Several fundamental aspects, in addition to technology advancement, must
be taken into account to create flexible, productive, and ergonomic HRC systems.

workers are mostly based on heuristic
algorithms (e.g., using the ergonomics
assessment worksheet) and do not have
the required resolution for determining
the function of individual muscles. This
ultimately limits the design of effective
technologies personalized to individual
workers. Common monitoring techniques rely on simple measurements
(e.g., limb accelerometry or kinematics)
where the worker's ergonomics is determined based on whole-body postures.
However, even kinematically correct
postures may underlie negative muscle
compensatory strategies that could
increase mechanical tensions and loads
on musculoskeletal tissues. Although
complex and advanced biomechanical
analyses exist, these are bounded to
laboratory settings and are not viably
transferable to factory settings because
they often involve lengthy data acquisition and time-consuming offline analyses with quantitative results being
generated only weeks after the initial
subject's assessment. In factory settings, the accurate assessment of a
worker's musculoskeletal function
should be performed via noninvasive
sensing technologies that require short
preparation time as well as advanced
yet rapid musculoskeletal functionprobing techniques that can provide
instant data on human musculoskeletal
mechanics [8]-[10].
Computational musculoskeletal
models can provide advanced analysis
and an understanding of body function
during complex dynamic tasks [11].
Inverse dynamics models have been
proposed in which the contribution of
individual muscles to joint actuation is
170

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

JUNE 2020

resolved according to a priori defined
optimization criteria (e.g., minimize the
squared muscle-activation sum or metabolic cost of transport) and/or by
enforcing preselected muscle-reflexive
rules [12]. However, current approaches
are dissociated from in vivo body function and therefore limited in describing
workers' body function in real-world
scenarios; that is, it is not possible to
know a priori what the human body
really optimizes during a working task,
if anything at all. Moreover, even
though one model can be tuned to
reproduce experimental data [e.g., electromyography (EMG)] in one instance,
synergies between muscles, or even
between motor units, are highly variable
across tasks, training, and fatigue levels
[13] and are directly influenced by the
environment, e.g., assistive devices or
working settings. Therefore, an alternative solution is needed that can capture
workers' true muscle-activation patterns
and convert them into realistic estimates of musculoskeletal forces.
One way to do this is by developing a
new class of data-driven musculoskeletal
models. The idea is to combine multimodal body movement sensing with forward dynamics musculoskeletal
modeling, i.e., as opposed to state-of-theart inverse dynamics-based modeling
techniques [14]. Data-driven modeling
has recently been proposed for fusing 3D
body kinematics and muscle EMG
recordings and for simulating how muscles activate and how they contract and
generate mechanical force with multiple
skeletal degrees of freedom simultaneously, in both upper and lower extremities, without making any assumptions

about muscle recruitment strategies [15].
More recently, these techniques were
employed to connect robotic exoskeletons and bionic arms with the human
neuromuscular system, thereby restoring
lost motor function in neurologically
impaired individuals as well as amputee
subjects. The results showed that patients
could achieve the voluntary control of
wearable robots and that the performance of the established human-
machine interface did not deteriorate
across large time scales, i.e., days and
weeks [16].
Sensing and Feedback
Interfaces
To enable an effective interaction
between humans and robots, it is fundamental to ensure a correct information exchange between the natural and
the artificial side. This requires suitable
interfaces that monitor human behavior-to properly plan the execution of
the collaborative task-and strategies
that increase the mutual awareness of
the human-robot dyad. In the literature, different solutions have been proposed to sense human behavior in free
motion or during interactions with the
environment. Regarding kinematic
sensing, both vision-based and wearable
devices have been developed with special attention to the hand (e.g., a glovebased system or inertial measurementunit-based solutions; see [17] and [18]
and Figure  2). Recently, commercial
and research solutions for whole-body
kinematic tracking [see, e.g., Xsense
(https://www.xsens.com)] as well as for
accurate and wearable EMG acquisition (e.g. Trigno by Delsys, Inc. or
FreeEMG by BTS Spa) and ground
reaction forces (http://www.moticon.
de) were introduced. In parallel, the use
of inertial units and low-cost, wearable
EMG devices [see, e.g., the Myo Gesture
Control Armband (https://www.myo
.com/)] has been successfully applied to
implement body-machine interfaces for
the control of and interaction with
robotic and assistive systems (see, e.g.,
[19]). For warning feedback delivering
information to humans on cobot status,
wearable devices usually rely on vibrotactile stimuli, which are also applied for
https://www.xsens.com http://www.moticon.de http://www.moticon.de https://www.bynorth.com/ https://www.bynorth.com/
IEEE Robotics & Automation Magazine - June 2020

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2020
