ORIGINAL RESEARCH

published: 18 August 2021

doi: 10.3389/frobt.2021.679304

Modelling and Control of a 2-DOF

Robot Arm with Elastic Joints for Safe
Human-Robot Interaction
Hua Minh Tuan 1,2, Filippo Sanfilippo 3* and Nguyen Vinh Hao 1,2
1
Department of Control Engineering and Automation, Faculty of Electrical and Electronic Engineering, Ho Chi Minh City University
of Technology (HCMUT), Ho Chi Minh City, Vietnam, 2Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam,
3
Department of Engineering Sciences, University of Agder (UiA), Grimstad, Norway

Collaborative robots (or cobots) are robots that can safely work together or interact
with humans in a common space. They gradually become noticeable nowadays.
Compliant actuators are very relevant for the design of cobots. This type of actuation
scheme mitigates the damage caused by unexpected collision. Therefore, elastic
joints are considered to outperform rigid joints when operating in a dynamic
environment. However, most of the available elastic robots are relatively costly or
difficult to construct. To give researchers a solution that is inexpensive, easily
customisable, and fast to fabricate, a newly-designed low-cost, and open-source
design of an elastic joint is presented in this work. Based on the newly design elastic
Edited by: joint, a highly-compliant multi-purpose 2-DOF robot arm for safe human-robot
Yongping Pan, interaction is also introduced. The mechanical design of the robot and a position
National University of Singapore,
Singapore
control algorithm are presented. The mechanical prototype is 3D-printed. The control
Reviewed by:
algorithm is a two loops control scheme. In particular, the inner control loop is
Alejandro Suarez, designed as a model reference adaptive controller (MRAC) to deal with
Sevilla University, Spain
uncertainties in the system parameters, while the outer control loop utilises a fuzzy
Sunan Huang,
National University of Singapore, proportional-integral controller to reduce the effect of external disturbances on the
Singapore load. The control algorithm is first validated in simulation. Then the effectiveness of the
*Correspondence: controller is also proven by experiments on the mechanical prototype.
Filippo Sanfilippo
filippo.sanfi[email protected] Keywords: collaborative robot, robot arm, series elastic actuator, human-robot interaction, robotics

Specialty section:
This article was submitted to 1 INTRODUCTION
Robotic Control Systems,
a section of the journal Cobots are robots intended for direct collaborative work with a human operator
Frontiers in Robotics and AI
(Wannasuphoprasit et al., 1997; Peshkin et al., 2001). Cobots appear and support humans
Received: 01 April 2021 in many situations in our daily life, e.g., search and rescue missions (Govindarajan et al., 2016),
Accepted: 02 July 2021
surveillance and inspection (Donadio et al., 2018; Brito et al., 2020), medical support (Mokaram
Published: 18 August 2021
et al., 2017), etc. One aspect, which makes cobots different from traditional robots, is the
Citation: capability to mitigate the damage caused by the unpredictable collision in dynamic shared
Tuan HM, Sanfilippo F and Hao NV
working space. There are various methods to achieve this capability, such as using force/torque
(2021) Modelling and Control of a 2-
DOF Robot Arm with Elastic Joints for
sensors (Li et al., 2020), using elastic joints, or using a collision detection algorithm without
Safe Human-Robot Interaction. changing a robot’s physical structure and adding external sensors (Xiao et al., 2018). Using
Front. Robot. AI 8:679304. elastic joints is one of the most common methods because it is efficient and low-priced.
doi: 10.3389/frobt.2021.679304 Conventional actuators are designed by following the traditional principle of “the stiffer the

Frontiers in Robotics and AI | www.frontiersin.org 1 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 1 | The proposed 2-DOF robot arm with elastic joints for safe human-robot interaction. The real 3D-printed prototype and the 3D model of the joint are
shown, respectively.

better” (Pratt and Williamson, 1995; Vanderborght et al., translational springs are used to handle the low force
2013). Due to having high force bandwidth, stiff actuators operation and reduce output impedance, stiction, and external
are suitable for position and speed control, and trajectory shock load. When the translational springs are fully compressed,
tracking tasks in isolated environments. However, in the torsional spring has a high effective stiffness and enhances the
unknown environments, these stiff actuators can be system bandwidth. This design is quite compact. The dynamic
damaged by undesired collision. In addition, most low-cost modelling and analysis of the proposed actuator is also
electric motors have to operate at high speed to obtain high demonstrated. In (Yu et al. (2013a), this same type of elastic
torque density. Therefore, to achieve low-speed output for actuator is adopted to design a knee-ankle-foot robot.
position control tasks, gear reduction is used at the expense of Successively, a force control approach is presented for the
introducing friction, noise, backlash, and torque ripple (Pratt same actuator in Yu et al. (2013b). First, two dynamical
and Williamson, 1995). On the other hand, elastic joints, with actuator methods are introduced based on different force
compliant motion and shock load absorption characteristics, ranges. Second, for the low force range, an optimal control
are developed to tackle these challenges. Shock tolerance of with friction compensation and disturbance rejection, which is
the elastic joints in unstructured environments is enhanced augmented by a feedforward control, is presented. The proposed
thanks to the elastic components working as a low-pass shock optimal control strategy is further extended to high force ranges.
filter. Furthermore, energy storage capability, power output Third, to manage the transition between low and high force
and force sensing are also enhanced (Paine et al., 2015). In control, a switching control technique is provided. The controller
spite of numerous advantages, controlling the position and is further enhanced in Yu et al. (2015). Human interaction
velocity of elastic joints is more challenging than conventional compensation, friction compensation, and a disturbance
stiff actuators due to the oscillation of elastic components. In observer are the primary components of the improved
addition, disturbance forces can cause the elastic joint to controller. When operating in human-in-charge mode, such a
deviate from its original position. control system allows the robot to achieve low output impedance,
The importance of human-robot interaction cannot be while precise force tracking is obtained when operating in force
overstated when it comes to cobots, especially when direct control mode.
physical interaction with humans is considered, In line with these research trends, a newly-designed low-cost,
i.e., rehabilitation robots. In this perspective, a design of an open-source design of an elastic joint is proposed in this work to
elastic actuator with springs having different stiffness is give researchers a solution that is economical, highly adaptable,
introduced in Yu et al. (2013c). The proposed design consists and quick to manufacture. The design is based on our previously
of a servomotor, a ball screw, a torsional spring between the proposed layout for the modules of Serpens (Sanfilippo et al.,
motor and the ball screw, and a set of translational springs 2019a; Sanfilippo et al., 2019b), a low-cost, open-source and
between the ball screw nut and the external load. The soft highly compliant multi-purpose modular snake robot. Based

Frontiers in Robotics and AI | www.frontiersin.org 2 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 2 | Different designs of the elastic element (A) series elastic actuator SEA23-23 from Yobotics (Pratt et al., 1997, 2002) (B) series elastic actuator from
Serpens robot (Sanfilippo et al., 2019a) (C) a torsional elastic element of wearable robots for knee assistance (Carpino et al., 2012) (D) LOPES′ spring (Lagoda et al.,
2010) (E) NASA Valkyrie’s spring (Paine et al., 2015; Radford et al., 2015).

FIGURE 3 | Elastic joint system (A) without external force/torque (B) compressed/tensed by external action (C) compression of the springs on the real prototype.

Frontiers in Robotics and AI | www.frontiersin.org 3 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

the results of experiments on the physical prototype are shown.

Finally, conclusions and future works are discussed in Section 7.

2 RELATED RESEARCH WORK

2.1 Related Control Methods
FIGURE 4 | The proposed two-loop controller. The advantages of elastic joints are first outlined by Pratt and
Williamson’s research (Pratt and Williamson, 1995), where a
proportional-integral-derivative (PID) control scheme was
on the newly designed elastic joint, the design of a 2-DOF robot proposed. Similarly, proportional-derivative (PD) controllers
arm for safe human-robot interaction is also proposed in this with on-line gravity compensation are considered in De Luca
article. The robot components are 3D-printed by using Fused et al. (2005), Zollo et al. (2007). The global asymptotic stability of
Deposition Modelling (FDM) manufacturing technology, thus these control laws is demonstrated via Lyapunov’s argument and
making the rapid-prototyping process very economical and La Salle’s theorem. Another alternative approach is based on
quick. The real 3D-printed robot arm is shown in Figure 1, robust controllers with a disturbance observer (DOB) (Kim and
together with the actuator model. Moreover, a decentralised Chung, 2015). In Talole et al. (2010), a combination of a feedback
control structure is introduced. Separate controllers are linearisation-based controller for the trajectory tracking problem
implemented for each joint. The control algorithm is based on and an extended state observer for uncertainty and states
a two-feedback loops position control mechanism: an adaptive estimation is considered. Feedback linearisation and robust
control is designed in the inner loop to stabilise the system and integral of sign of error (RISE) methods for controlling
deal with uncertainties, while a fuzzy controller is considered for position are proposed in Yin et al. (2016). Specifically, the
the outer loop to eliminate the influence of external force/torque. dynamics of the actuator is first feedback linearised, then a
Then, the controller is validated in simulation with (MATLAB, RISE method is applied to adapt the system model to
2018) and tested in reality with the 3D-printed prototype. The uncertainties. A model reference adaptive control approach is
simulation and experiment results show that the decentralised implemented in Losey et al. (2016) to adapt to uncertainties in
controller could stabilise and precisely control the elastic joint in system parameters, while the adoption law is demonstrated using
normal and disturbed conditions. Lyapunov’s theory. In Pérez-Ibarra et al. (2017), a H∞ force
The paper is organised as follows. A review of the related controller is used to deliver precisely the force required by a PD
research work is given in Section 2. In Section 3, we focus on the position controller on the outer loop. In Kaya and Çetin (2017),
description of the proposed control algorithm. In Section 4, the an adaptive state feedback regulator and a conventional PI
physical prototype of the 2-DOF robot arm is presented. In controller is adopted to control a rotary series elastic actuator.
Section 5, related simulation results are outlined. In Section 6, The adaptive state feedback regulator define the transient

FIGURE 5 | Fuzzy membership functions (A) input membership functions (B) output membership functions.

Frontiers in Robotics and AI | www.frontiersin.org 4 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

TABLE 1 | Fuzzy inference rules. position output of a BLDC motor. The iterative learning control loop
is used to control the output trajectory of the actuator. To the best of
our knowledge, an adaptive control aiming at both stabilising the
system as well as dealing with uncertainties to eliminate the influence
of external force/torque has not been released yet.

2.2 Related Mechanical Design

Elastic elements play a very important role in elastic joints. They are
responsible for absorbing unexpected collision and storing elastic
energy. The elastic elements can be springs and can have different
shapes (as shown in Figure 2). Springs in elastic joints can be
classified into torsion spring and extension/compression spring. In
(Pratt et al., 1997, 2002; for; Human and Cognition, 2020), a pair of
extension/compression springs are used, as shown in Figure 2A. In
behaviour of the SEA, then, the tracking error is reduced by the PI (Sanfilippo et al., 2019a), two springs are bent to fit into a compact
controller. In Penzlin et al. (2019), a clutched parallel elastic actuator housing, as shown in Figure 2B. Beside conventional springs, there
(CPEA) with an iterative learning control loop and a cascade control are some custom designs of the elastic element. Some examples are
loop is proposed. The cascade control loop is used to control the shown in Figure 2.

FIGURE 6 | Mechanical design of the elastic joint (A) exploded view. Load side: (1) load side shaft, (2) cap, (3) 8 mm-shaft flange, (4) load side compliant joint. Motor
side: (5) motor side driving joint, (6) cover, (7) flange extension, (8) 8 mm-shaft flange, (9) motor side shaft (B) 3D view of the elastic joint (C) assembled view of the design
of the elastic joint (D) view from the motor side (E) view from the load side.

Frontiers in Robotics and AI | www.frontiersin.org 5 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 8 | Diagram of the elastic robot arm.

TABLE 2 | Denavit-Hartenberg parameters.

d Θ a α

Joint 1 l1 q1 0 π/2
Joint 2 0 q2 l2 −π/2

To amplify the torque and reduce the speed of a motor, gearing

systems are adopted. Besides the common spur gears, some special
types of gear are also utilised in elastic joints. Strain wave gearing, also
known as harmonic gearing, can be applied, e.g., (Negrello et al., 2015;
Wang et al., 2017; Sergi et al., 2012). This mechanism consists of three
basic components: a wave generator, a flex spline, and a circular
spline. The advantages of the harmonic gearing mechanism are no
backlash, high compactness, lightweight, high gear ratios in a small
volume, and coaxial input and output shafts. In the research of Meng
(Wang et al.,2015; Wang et al., 2017), a nonlinear series elastic joint
with variable stiffness is presented. There are two motors and a
harmonic gearing mechanism in this design. One motor is connected
to the wave generator, while the other motor is used to change the
stiffness preset, and the link is attached to the flex spline. Planetary
gearing system is also utilised in elastic joints, e.g., (Lee and Oh, 2016;
Plooij et al., 2016). The advantages of the planetary gearing
mechanism are compactness, high efficiency, low backlash, high
torque density (i.e., high torque-to-mass ratio), and coaxial input
and output shafts. In Lee and Oh (2016), good control performance is
achieved while the size is still compact. The motor shaft is connected
to the Sun gear, the torsional spring is connected to the ring gear, the
FIGURE 7 | Mechanical design of the 2-DOF robot arm with elastic
load is attached to the carrier and the planet gears combine these
joints: (1) Gripper, (2) elastic joints, (3) GA25 DC motors, (4) AMT20 encoders.
three parts. One side of the spring and the motor is attached to a fixed

Frontiers in Robotics and AI | www.frontiersin.org 6 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 9 | Software design.

TABLE 3 | System parameters.

Parameter Value Parameter Value

Gear ratio(N) 1 Spring stiffness (Ks ) 3

Load damping coefficient (Dl ) 0.006 Motor damping coefficient (Dm ) 0.06
Load inertia (Jl ) 0.065 Motor inertia (Jm ) 0.1
Spring damping coefficient (Ds ) 0.6 Sampling rate (T) 0.001

base. In Plooij et al. (2016), a BIdirectional Clutched Parallel Elastic external noise and noise caused by electromechanical systems
Actuator (BIC-PEA) is introduced. The mechanism contains a (e.g., vibrations of the mechanical parts, voltage spike in
planetary differential, a torsion spring, and two brakes. It is stated electronic components, etc). By denoting the motor angular
that BIC-PEA design can save up to 65% of the energy consumption position as θ, the load angular position as q, the rotor inertia
in some specific tasks. as Jm , the motor damping coefficient as Dm , the stiffness
coefficient of the spring as Ks , the spring damping coefficient
as Ds , the load inertia as Jl , and the load damping coefficient as Dl ,
3 CONTROL ALGORITHM the mathematical model of the elastic joint system is obtained:

dm + τ m − N −1 τ s  Jm θ€ + Dm θ,
_ (1)
In this section, a mathematical model of the considered elastic
−1 −1 _
joint system is introduced. A novel control algorithm was τ s  Ks
N θ − q + Ds N θ − q, _ (2)
previously designed by our research group and proposed in dl + τ s + τ ext  Jl q€ + Dl q.
_ (3)
(Hua et al., 2019). This control algorithm is summarised in
this section. For further details, the reader is referred to (Hua
et al., 2019). Eq. 1 shows the relationship on the motor-side between the
motor torque, the spring torque and the motor angular position.
3.1 Mathematical Model The spring torque, τ s , is obtained by Eq. 2. The interaction
The schematic diagrams of the elastic actuator are illustrated in between the spring torque, the external torque and the load
Figure 3. There are two gears with gear ratio N  Nl /Nm , where angular position is illustrated by Eq. 3.
Nl and Nm are the number of teeth for the load and the motor
gear, respectively. The torques shown in these diagrams are motor 3.2 Controller Design
torque (τ m ), spring reaction torque (τ s ) and external torque (τ ext ). The proposed control algorithm diagram is presented in Figure 4.
Figure 3A illustrates the system when there is no external force/ The objective of this work is to develop a controller that can track
torque, while Figure 3B illustrates the system affected by an the desired load angular position when considering external
external action. The compression of the springs on the real disturbances on the load and uncertainties in system
prototype is also shown on Figure 3C. parameters. To achieve this objective, two separate types of
The whole system is affected by disturbances from the motor controllers are utilised for the motor-side and the load-side,
(dm ) as well as from the load (dl ). These parameters represent respectively. For the load-side, we propose using a Fuzzy PI

Frontiers in Robotics and AI | www.frontiersin.org 7 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 10 | Load-side response of the FPIC-MRAC (A) step input (B) sine input.

Controller (FPIC) to reduce the effect of external disturbances on 3.2.1 Fuzzy PI Controller
the load. The considered external disturbances are the forces/ An FPIC is applied to the load-side to cope with the effect of
torques caused by undesired collisions when operating in external disturbances on the elastic joint system. A fuzzy
unknown environments. The output of the FPIC is used as the controller is not based on a strict mathematical model and is
desired angular position of the motor. For the motor-side, a widely used to solve problems under uncertain environments,
Model Reference Adaptive Controller (MRAC) is used to cope with high nonlinearities (Al-Odienat and Al-Lawama, 2008). A
with uncertainties in system parameters. The uncertain system fuzzy algorithm is a control method based on fuzzy logic, which
parameters could be the inertia of the load or the stiffness of the uses “fuzzy inference rules” instead of “equations”. These fuzzy
spring. The idea is inspired to a previous research of (Losey et al., inference rules may come from experience of a human expert in
2016). The advantage of combining the FPIC and the MRAC controlling a specific object, or in other cases, from the
controllers is the possibility of achieving independence with understanding of dynamics and behaviour of the target plant.
respect to imprecise system parameters. In previous works A fuzzy controller can be combined with a conventional PID
(Losey et al., 2016), the motor angular position is prioritised controller to obtain a fuzzy PI, a fuzzy PD or a fuzzy PID
more than the load position because the target is force control. controller. The fuzzy PI type controller is known to be
The contribution of the paper is that a robust position more practical than fuzzy PD types because it is difficult for
control algorithm for the load position is proposed. In the fuzzy PD to remove steady state errors (Rao et al., 2010). The
addition, a better spring model is given with the damping relationship between the input and the output in a conventional
coefficient added. PI controller is expressed in the following equation:

Frontiers in Robotics and AI | www.frontiersin.org 8 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 11 | Motor-side response of the FPIC-MRAC (A) step input (B) sine input.

t
and the change of error are similar, including the rules
u(t)  Kp e(t) + Ki ​ e(t)dt, (4)
0 base of the fuzzy model: Negative Big (NB), Negative Small
(NS), Zero (ZE), Positive Small (PS), and Positive Big (PB).
where, Kp and Ki are scaling the coefficient of the feedback error,
The membership functions for the change of the control signal
the change of feedback error and the change of control signal,
are in singleton over the output, given NB, NS, ZE, PS, and
respectively. In fuzzy control, the experience of the human
PB. On the basis of the input and output membership
operator is applied to establish fuzzy inference rules. However,
functions, 25 fuzzy inference rules are established as shown
the integral of the error is more difficult to be analysed than the
in Table 1.
change of the error. Therefore, the control signal u is
differentiated so that the change of the error can be
considered. Then, by differentiating Eq. 4, we obtain: 3.2.2 Model Reference Adaptive Controller
Although there are various control algorithms available from the
u(t) _ + Ki e(t).
_  Kp e(t) (5) past literature (Pratt and Williamson, 1995; Talole et al., 2010;
A fuzzy PI controller can be obtained by combining Eq. 5 with Losey et al., 2016; Penzlin et al., 2019), uncertainties in system
the fuzzy controller. Inputs of the FPIC are the feedback error (e) parameters can lead to instability in many cases. Adaptive
_ while the output of the FPIC is the
and the change of this error (e), controllers are developed to overcome this problem. An
_
change of control signal (u). MRAC is an important adaptive controller typology in which
The membership functions for the input and the output are the desired response is expressed by a reference model. The
shown in Figure 5A,B, in which c1 − c5 are parameters to be adaptation law modifies the system parameters based on the
adjusted. These parameters are chosen by trial and error in the difference between the output of the real system and the output of
range of (0, 1). The narrower the membership functions are, the reference model. In this article, Lyapunov’s stability theory is
the faster is the response at the expense of larger oscillations applied to design the adaptation law. Lyapunov’s stability
and overshoots. The membership functions for the error criterion states that a system, x_  f (x), with equilibrium point

Frontiers in Robotics and AI | www.frontiersin.org 9 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 12 | Response of the FPIC-MRAC of the robot arm with external disturbance (A) load-side response (B) motor-side response (C) motor control voltage.

at x  0, is stable if there is a function, V(x), that satisfies the Firstly, the control law is derived. The motor-side system
following conditions: equations can be rewritten by using Eqs. 1, 2 (ignoring the motor-
side external disturbance dm ) as:
0 1 0
V(x)  0 if x  0, (6) θ_ ⎡⎢⎢⎢ ⎤⎥⎥⎥ θ ⎢⎡ ⎥⎤
⎥⎥⎥⎡⎣ ⎤⎦ + ⎢⎢⎢⎢⎢ ⎥⎥⎥⎥⎥
τ m + Ks N −1 q + Ds N −1 q
θ_ m  ⎢⎣⎡ ⎥⎦⎤  ⎢⎢⎢⎣⎢ −K N −2 −Ds N − Dm ⎥⎦ θ_
−2
⎣1⎦ _
V(x) > 0 if x ≠ 0, (7) €θ s
(9)
Jm Jm Jm
_
V(x) ≤ 0 for all x ≠ 0. (8) −1
_
 Aθm + B
τ m + Ks N q + Ds N q. −1

Frontiers in Robotics and AI | www.frontiersin.org 10 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

TABLE 4 | Reference model and MRAC parameters. In Eq. 11, the state matrix Ar , input matrix Br and state vector
Parameter Value Parameter Value θref of the reference model can be obtained as:

Damping coefficient (ξ) 1 p1 7 0 1 0 θr

Ar  , Br  , θref  . (12)
Natural frequency (ωn ) 5 p2 0.06 −ω2n −2ξωn ω2n θ_ r
q1 3 p3 0.256
q2 5 Learning rate coefficient (γ) 0.995 A general control law for the system with state Eq. 9 is:
 s N −1 q − D
τ m  Mθmd − Lθm − K  s N −1 q.
_ (13)
In Eq. 9, the state matrix A, input matrix B and state vector θm
of the system can be obtained as: Parameters with hats (D  m , J m , K
 s, D
 s, D
 l , J l ) are estimated
parameters, while matrices M and L need to be determined. By
0 1 0
⎡⎢ ⎤⎥⎥ ⎢
⎡ ⎤⎥⎥ θ substituting τ m to system Eq. 9, we have:
A  ⎢⎢⎢⎣ −Ks N −2 ⎢
−Ds N −2 − Dm ⎥⎥⎦, B  ⎢
⎣ 1 ⎥⎥⎦, θm 
⎢ . (10)
θ_
Jm Jm Jm θ_ m  (A − BL)θm + BMθmd + BKs − K
 s N −1 q
 s N −1 q.
+ Ds − D _ (14)
The motor-side system equations have the form of a second-
order system, so the reference model is a second-order system If perfect estimation of parameters is obtained, we have Ks 
model with the desired signal θmd , natural frequency ωn , and Ks and D  s  Ds . As the columns of matrices A − Ar and Br are
damping coefficient ξ: linear combinations of the vector B, there exists optimal matrices
M * and L* of matrices M and L such that:
θ.._ r 0 1 θ 0
θ_ ref    _r  + θ
θr −ω2n −2ξωn θr ω2n md (11) A − Ar  BL* , (15)
 Ar θref + Br θmd . Br  BM . *
(16)

FIGURE 13 | Load-side response of the PID controller of the robot arm (A) in normal condition (B) with external disturbance.

Frontiers in Robotics and AI | www.frontiersin.org 11 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 14 | Load-side response and control voltage of the FPIC-MRAC with human-machine interaction (A) load-side response (B) control voltage (C) external
disturbance.

Eqs. 15, 16 are called compatible conditions. If these −1

L* 
BT B BT (A − Ar )
conditions are satisfied and we have perfect estimation of (18)
 
−Ks N −2 + Jm ω2n 
−Ds N −2 − Dm + 2Jm ξωn .
parameters, then the controller 13 can yield perfect tracking of
the reference model. From these equations, the controller optimal Based on the above Eqs. 17, 18 for the controller optimal
matrices M * and L* can be obtained as below: matrices, the matrices M and L can be approximated as:
−1
M * 
BT B BT Br  ω2n Jm , (17) −1
M 
BT B BT Br  ω2nJ m , (19)

Frontiers in Robotics and AI | www.frontiersin.org 12 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

FIGURE 15 | Experiment: angle response and control voltage of the FPIC-MRAC of joint 1 in normal condition, sine waveform input (A) angle response of joint 1 (B)
control voltage of joint 1.

−1
L 
BT B BT (A − Ar ) where, it should be noted that:
 s N −2 − D
 s N −2 + J m ω2n  − D  m + 2J m ξωn . (20)
 − K
Ψ  B −θTm θmd
−N −1 q
− N −1 q
_ ,
Secondly, the error equation is determined. The feedback LT L*T
⎡ M ⎥⎥⎥⎤
⎢
⎢
⎢ ⎢
⎡ M * ⎥⎥⎥⎤
⎢
error, which is the difference between the output of the
Φ⎢ ⎢
⎢
⎢ ⎥⎥⎥⎥, Φ*  ⎢
⎢
⎢
⎢
⎢
⎥⎥⎥. (23)
real system and the output of the reference model, is ⎢
⎣K s ⎥⎦ ⎣ Ks ⎥⎥⎦
⎢
determined as: s
D Ds

e  θm − θref . (21) Thirdly, the adaption law can be obtained by applying

Lyapunov’s stability theory. A Lyapunov function is
The derivative of error is determined as: introduced as:
1 1 T
e_  θ_ m − θ_ ref  (A − BL)θm + BMθmd + BKs − K
 s N −1 q + Ds − D
 s N −1 q
_ V  ceT Pe +
Φ − Φ* 
Φ − Φ* , (24)
2 2
− Ar θref + Br θmd 
where, P is a symmetric positive definite matrix and γ is a learning
 s N −1 q + Ds − D
 Ar e − Ar θm + (A − BL)θm + BMθmd + BKs − K  s N −1 q
_ rate. The function V is positive definite. The derivative of V is
− Br θmd  Ar e + Ψ
Φ − Φ* , (22) obtained as:

Frontiers in Robotics and AI | www.frontiersin.org 13 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

1 1 _ 8 mm outer diameter. When the motor rotates, the springs

V_  ce_T Pe + ceT Pe_ +
Φ − Φ*  Φ
T
2 2 are compressed. This will create an elastic force on the
(25)
1 compliant joint. The load side shaft is connected to the
T
 − ceT Qe +
Φ − Φ*  Φ_ + cΨT Pe,
2 joint through the flange. The shaft is kept in place by a
cap. In our work, two extension/compression springs are
where, ATr P + PAr  −Q and Q is a symmetric positive definite chosen because they form a symmetrical configuration. In
matrix. The existence of matrix Q is demonstrated by using the addition, the extension/compression type of spring is used
Kalman-Yakubovich lemma (Åström and Wittenmark, 2013). If instead of torsion spring because it is more durable and can
the adaption law is chosen to be: bear a larger force.
Based on the newly introduced actuator, the design of
_  −cΨT Pe,
Φ (26) the 2-DOF elastic robot arm is shown in Figure 7. The joints
of the robot arm are revolute joints. The motor and the elastic
then the derivative of Lyapunov function V_ is negative definite joints are connected by pulleys and belts. The transmission ratio
with all e ≠ 0, which means that the feedback error between the from the motor to the actuator is 4:1 (the pulley of the motor has
output of the real system and the reference model will go to zero 20 teeth and the pulley of the actuator has 80 teeth).
when time goes to infinity. The simplified model of the robot arm is shown in Figure 8. The
In this article, due to their symmetry, the matrices P and Q are Denavit-Hartenberg table is shown in Table 2. The angles are in
chosen as it follows: radian and the lengths are in meter. From this table, the corresponding
transformation matrices are formed, as shown in Eq. 30.
p1 p2 q1 0
P , Q  , (27)
p2 p3 0 q2 cos
q1  0 sin
q1  0
⎢
⎡
⎢
⎢ sin
q1  −cos
q1  0 ⎤⎥⎥⎥⎥⎥
T12  ⎢
⎢ 0
⎢
⎢ ⎥,
where, q1 and q2 will be tuned appropriately. From equation ⎢ 0
⎣ 1 0 l1 ⎥⎥⎦
ATr P + PAr  −Q, p1 , p2 and p3 can be obtained by using the 0 0 0 1
following formulas:
q1 2p2 + q2 cos
q2  0 −sin
q2  l2 cos
q2 
p2  , p3  , p1  2ξωn p2 + ω2n p3 . (28) ⎢
⎡
⎢
⎢ sin
q2  0 cos
q2  l2 sin
q2  ⎤⎥⎥⎥⎥⎥
2ωn2 4ξωn T23  ⎢
⎢
⎢
⎢
⎢ 0 ⎥⎥⎦. (30)
⎣ −1 0 0
In real applications, discrete system equations are utilized 0 0 0 1
instead of continuous ones. By approximately discretising the
reference system Eq. 9, we obtain: 4.2 Hardware Design
In this section, the hardware used in the robot arm is presented.
θr (k + 1) 1 T θ (k) 0 The motor used in the robot arm is the GA25 DC motor
  _r + θ .
θ_ r (k + 1) −Tω2n 1 − 2Tξωn θr (k) Tω2n md (XYTmotor, 2020). It is a 12 V DC motor with a 4 mm shaft
(29) and a two channels encoder attached. The gearbox transmission
ratio is chosen as 217:1. The encoder for the load side is the AMT-
20 encoder (devices, 2020) from CUI Devices. It has incremental
and absolute encoders in a compact package. The incremental
4 PHYSICAL PROTOTYPE encoder has three channels (A, B, and Z). The absolute encoder
has 12 bits of absolute position information with Serial Peripheral
4.1 Mechanical Design Interface (SPI) communication. The microcontroller used is the
In this section, the mechanical design of the elastic joint and STM32F446RE on the STM32 Nucleo board (Stmicroelectronics,
the robot arm are described. The mechanical components of 2020).
the elastic joint are shown in Figure 6A. A 3D view of the same
parts is illustrated in Figure 6B, while the assembled view of 4.3 Software Design and Architecture
the actuator is illustrated in Figure 6C–E. The mechanism can The software design is described in this section. The concept of
be separated into load side and motor side. The motor side modularity is applied to the control architecture. Each joint is
mechanism includes an 8 mm metal shaft, an 8 mm metal independent, being controlled by a self-reliant slave controller,
flange, a flange extension, a cover, and a driving joint which directly communicates with the master controller
(component 9, 8, 7, 6, 5 in Figure 6A, respectively). The running on a personal computer (PC). Each slave controller
load side mechanism includes an 8 mm metal shaft, a cap, an communicates with the PC through a Universal Serial
8 mm metal flange, and a compliant joint (component 1, 2, 3, 4 Bus (USB) port on the board. Figure 9 shows the diagram of
in Figure 6A, respectively). The shafts are connected to the the control architecture. Each STM32 board has timer
other parts by the flanges. The flange on the motor side is peripherals, which are used to capture the encoder signals.
connected to the flange extension and the flange extension is The timers are also configured to be Pulse width modulation
connected to the cover and the motor side driving joint. The (PWM) output to control the motors through H-bridges
motor side driving joint has space to place two springs with (nhatbon, 2020). The control/update rate of the controller

Frontiers in Robotics and AI | www.frontiersin.org 14 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

is 100 Hz (10 milliseconds). The presented framework is a 5.3 Response of the Load and Motor-Side
multi-layer architecture that includes the following layers: System With External Torque to the Step
High-level control layer: it considers the control of the
overall manipulator, it includes kinematics according to the
Desired Signal
In this section, the effect of the disturbance on the elastic joint and
corresponding DOFs. Low-level control layer: it considers
the efficiency of the proposed algorithm is demonstrated. The
the control of the single joints. This makes it possible to use
disturbance signal affecting the load side is a sine wave signal with
the proposed approach for robotic arms with different DOFs.
an amplitude of five radian and a cycle time of 0.5 s. This signal is
applied to each joint. This is equivalent to the vibrations often
occurring in elastic joints. The sine response of the load-side system
5 SIMULATION RESULTS and the motor-side system is shown in Figure 12A,B, respectively.
In this section, simulations are outlined with the aim of The motor control voltage is shown in Figure 12C. In the proposed
validating the proposed control algorithm for the two-loop controller, the FPIC is adopted on the load side to reduce
considered elastic joint. The simulations are conducted on the influence of external disturbances and the MRAC is adopted on
(MATLAB, 2018). This software is chosen because of its ability the motor side to deal with uncertainties. When there is an external
to quickly constructing simulation models and validating torque, the desired motor angular position is adjusted by the FPIC,
control schemes. The responses of the elastic joint with step so that the error between the desired load angular position and the
and sinusoidal wave inputs are presented. The effect of real load angular position goes to zero. As shown in Figure 12B,
external torque and disturbances on the whole elastic the motor angular position changes when the external torque
system is also illustrated. The system parameters used in appears and the influence of this external torque is eliminated.
the simulation are shown in Table 3. To show more evidently the effectiveness of the proposed control
algorithm, a standard PID controller is applied and its results are
compared. As shown in Figure 13A,B, the responses of the PID
5.1 Response of the Load-Side System With controller and the FPIC-MRAC controller are similar in normal
condition. However, as shown in Figure 13B and Figure 12A, the
Step and Sine Desired Signals Without response of the FPIC-MRAC controller is significantly better than
External Torque the PID controller in the presence of external disturbance.
The response of the load-side system in normal condition
is illustrated in this section. Step and sine responses are shown
in Figure 10. In these figures, blue dashed lines are desired inputs 5.4 Response of the Load and Motor-Side
and red solid lines are load angular positions. The unstable stage System With Human-Machine Interaction
at the beginning of the simulation is the learning phase of the The impact of human-machine interaction on the elastic joints is
MRAC. The results show that the proposed controller has presented in this section. The load side response and the control
relatively fast response and small error. voltage are shown in Figure 14. The simulated human-machine
interaction of joint 1 appears from the second 20 to 30 and from
5.2 Response of the Motor-Side System 40 to 45. The simulated human-machine interaction of joint 2
With Step and Sine Desired Signals Without appears from the second 10 to 20 and from 30 to 40. The collected
External Torque data outline that despite being affected by the disturbance, the
The response of the motor-side system in normal condition is controller can adjust the voltage to keep the system stable.
presented in this section. Step and sine responses are shown in
Figure 11A,B, respectively. The reference model and the MRAC
parameters are shown in Table 4tbl4. The system parameters are 6 EXPERIMENTS
modified based on the adaption law (Eq. 26) to reduce the error
between reference model and real motor response. The adaption In this section, experiments are conducted to verify the
law is designed using Lyapunov’s approach. This has an effectiveness of the control algorithm on the mechanical
advantage: arbitrary large values of the learning rate coefficient prototype of the robot arm. The experiments consist of joints’
γ can be used (Åström and Wittenmark, 2013). response and trajectory tracking. It should be noted that, in the
The damping coefficient ξ of the reference model is chosen simulation, the elastic joints are controlled directly by changing the
to be critically damped (ξ  1) because there are no voltage, as shown in Figure 12C. However, in the experiments, the
oscillations or overshoots in this configuration and the elastic joints are controlled by torque produced from the motor
system returns to equilibrium in minimum time. However, which is proportional to the current injected by the H-bridge. This
Figure 11A shows that there is a small overshoot in current is dependent on the applied voltage, which is controlled by
the reference model graph. This is caused by the integrator adjusting the duty cycle of the PWM pulse.
in the FPIC, not by the reference model in the MRAC,
and could be reduced by appropriately adjusting the Kp 6.1 Experiment in Normal Condition
coefficient. The rising time of the reference model is In this subsection, experiment results related to the control of the
calculated by using the approximation formula: 5.83392ωn . real 2-DOF robot arm with elastic joints in the normal condition

Frontiers in Robotics and AI | www.frontiersin.org 15 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

are presented. The normal condition is when the robot arm is not 6.4 Trajectory Tracking Experiment When
disturbed by the external torque. The joint angle response and There Is Unexpected Torque
control voltage are demonstrated. There are two input signal In this subsection, trajectory tracking experimental results of the
types: sinusoidal and square waveforms. The sinusoidal elastic robot arm when there is an unexpected torque are
waveform input has an amplitude of 0.5 rad and a cycle time demonstrated. A sequence of images from the experiment is
of 8 s. The square waveform input has an amplitude of 0.5 rad and shown in Supplement Figure S8. The initial position is when
a cycle time of 8 s. The data is collected within 50 s. q1  0, q2  3π
4 and the end position is when q1  π, q2  6 . As
7π
Figure 15; Supplement Figure S1 show the angle response shown in the third image of the sequence (Supplement Figure
and the control voltage of joint 1 and joint 2, respectively, in S8), the unexpected torque is produced by the collision with a
normal condition with the sinusoidal waveform input. The human hand. The desired trajectory and the response of the robot
figures demonstrate that the decentralised control algorithm arm are shown in Supplement Figure S9A. Obviously, there is a
works as expected. As shown in Figure 15A, the absolute value section where the end effector of the robot arm deviates from the
of the positive peak of the motor angle is larger than the desired trajectory. It corresponds to the period when the arm is
negative peak. This is due to a small manufacturing defect in affected by undesired torque. Supplement Figure S9B,C
the actuator: the spring on one side is stiffer than the spring on illustrate the responses of joints 1 and 2, respectively. As
the other side. Despite this fact, it is clear that the controller shown in these figures, when the collision occurred from
still tracks the desired load angle. second 7 to second 12, the robot arm tries to hold its position
The angle response and the control voltage of joint 1 and joint to mitigate the damage. After that, when the robot arm is released,
2 are shown in Supplement Figures S2, S3, respectively, in it gradually converges to the end position.
normal condition with the square waveform input. These
figures demonstrate that the decentralised control algorithm
also works well in this situation. The error decreases to
smaller than 10° after just a few seconds. 7 CONCLUSION
A newly designed elastic joint was presented in this work based on
6.2 Experiment When There Is Unexpected our previuous design, which was previously introduced in
Torque Sanfilippo et al. (2019a), Sanfilippo et al. (2019b). Based on the
In this subsection, the experiment results of the real 2-DOF robot developed elastic joint, a 2-DOF robot arm with elastic behaviour
arm with elastic joints in presence of disturbances are presented. for safe human-robot interaction was also presented. The
The joint angle response and control voltage are demonstrated. The mechanical prototype can be 3D-printed by using Fused
sinusoidal waveform input has an amplitude of 0.5 rad and a cycle Deposition Modelling (FDM) manufacturing technology, thus
time of 8 s. The data is collected within 50 s. The external making the rapid-prototyping process very economical and fast.
disturbance is created by randomly holding the robot arm by Moreover, a position control algorithm was introduced to control
hand while it is operating. This imitates the unexpected collision each joint of the arm. The control algorithm is based on a 2-loops
with a human operator while coexisting in a shared work- control mechanism. In particular, the inner control loop is
space.The angle response and the control voltage of joint 1 and designed as a model reference adaptive controller (MRAC) to
joint 2 in disturbance condition are shown in Supplement Figures deal with uncertainties in the system parameters, while the outer
S4, S5, respectively. These figures demonstrate that the control loop utilises a fuzzy proportional-integral controller (FPIC)
decentralised control algorithm can stabilise the system when to reduce the effect of external disturbances on the load.
there is unexpected torque. In addition, if the absolute value of Preliminary simulations were carried out in Matlab to validate
the deviation between the motor angle and the load angle is larger the potential of the proposed control algorithm. Successively, the
than 30° (0.523,599 radians), the motor will stop. This value can be effectiveness of the controller was also proven by conducting
tuned appropriately, which means that safety can be guaranteed. experiments with the mechanical prototype.
As future work, the possibility of replacing the DC motors
with brushless DC (BLDC) motors will be considered. Due to
6.3 Trajectory Tracking Experiment in having high torque density, BLDC motors will make it possible
Normal Condition to achieve a more compact actuation system. In addition, as
In this subsection, trajectory tracking experiment results of the shown in Figure 9, the proposed controller has a decentralised
elastic robot arm in the normal condition are demonstrated. A design. This means that each joint has its own separated
sequence of images of the experiment is illustrated in controller, both in hardware and software. Therefore, the
Supplement Figure S6. The initial position is when q1  number of DOFs can be expanded and tested easily in the
0, q2  3π
4 and the end position is when q1  π, q2  6 . The
7π future. Furthermore, the stability analysis at the system level will
desired trajectory and the response of the robot arm are shown also be considered.
in Supplement Figure S7A. It is clear that the end effector of Moreover, the possibility of realising a digital-twin of the
the robot can be following the desired trajectory. In details, presented robot will be considered to enable researchers for
Supplement Figure S7B,C, show the responses of joints 1 and designing and prototyping different control algorithms. In this
2, respectively. The tracking error is relatively small. perspective, the integration with an open-source software

Frontiers in Robotics and AI | www.frontiersin.org 16 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

framework will be investigated. To achieve this, there are different AUTHOR CONTRIBUTIONS
robotic frameworks and middleware available in recent years
(Tsardoulias and Mitkas, 2017). However, the Robot Operating Conceptualisation, TH and FS; methodology, TH and FS; software, TH;
System (ROS) (Quigley et al., 2009) has emerged as a de facto validation, TH, FS and HN; formal analysis, TH and FS; investigation,
standard for robot software architecture in the research TH and FS; resources, HN; data curation, TH; writing—original draft
community. In conjunction with ROS, Gazebo 3D simulator preparation, TH and FS; writing—review and editing, TH, FS and HN;
(Koenig and Howard, 2004) can be adopted to accurately and visualisation, TH; supervision, FS and HN; project administration, FS
efficiently simulate robots in complex indoor and outdoor and HN; funding acquisition, FS and HN. All authors have read and
environments. Therefore, these tools will be considered to agreed to the published version of the manuscript.
realise a digital-twin of the presented robot.
In the future, the design of reliable low-level control
algorithms for the proposed elastic joints could be also FUNDING
relevant for other robotic systems, such as snake like
robots similar to Serpens (Sanfilippo et al., 2019a), which is We acknowledge the support of time and facilities from Ho Chi
developed by our research group. This technology may be Minh City University of Technology (HCMUT), VNU-HCM and
essential to enable the achievement of perception-driven from the Top Research Centre Mechatronics (TRCM), University
obstacle-aided locomotion (POAL) (Sanfilippo et al., 2016, of Agder, Norway, for this study.
2017, 2018).

SUPPLEMENTARY MATERIAL
DATA AVAILABILITY STATEMENT
The Supplementary Material for this article can be found online at:
The raw data supporting the conclusion of this article will be https://www.frontiersin.org/articles/10.3389/frobt.2021.679304/
made available by the authors, without undue reservation. full#supplementary-material

Kaya, K. D., and Çetin, L. (2017). Adaptive state feedback controller design for a
REFERENCES rotary series elastic Actuator. Trans. Inst. Meas. Control. 39, 61–74. doi:10.1177/
0142331215600779
Al-Odienat, A. I., and Al-Lawama, A. A. (2008). The Advantages of Pid fuzzy Kim, M. J., and Chung, W. K. (2015). Disturbance-observer-based Pd control of
controllers over the conventional types. Am. J. Appl. Sci. 5, 653–658. flexible joint robots for Asymptotic convergence. IEEE Trans. Robot. 31,
doi:10.3844/ajassp.2008.653.658 1508–1516. doi:10.1109/tro.2015.2477957
Åström, K. J., and Wittenmark, B. (2013). Adaptive Control. North Chelmsford, Koenig, N., and Howard, A. (2004). “Design And use Paradigms for gazebo, an
Massachusetts: Courier Corporation. open-source multi-robot simulator,” in Proc. of the IEEE/RSJ International
Brito, T., Queiroz, J., Piardi, L., Fernandes, L. A., Lima, J., and Leitão, P. (2020). A Conference on Intelligent Robots and Systems (IROS), Sendai, Japan (IEEE).
machine learning Approach for collaborative robot smart manufacturing vol. 3, 2149–2154.
inspection for quality control systems. Proced. Manufacturing 51, 11–18. Lagoda, C., Schouten, A. C., Stienen, A. H., Hekman, E. E., and van der Kooij, H.
doi:10.1016/j.promfg.2020.10.003 (2010).Design of an electric series elastic Actuated joint for robotic gait
Carpino, G., Accoto, D., Sergi, F., Luigi Tagliamonte, N., and Guglielmelli, E. rehabilitation training. In 2010 3rd IEEE RAS & EMBS International
(2012). A novel compact torsional spring for series elastic Actuators for Conference on Biomedical Robotics and Biomechatronics. New Jersey,
Assistive wearable robots. J. Mech. Des. 134 (121002), 1–10. doi:10.1115/ United States: IEEE. 21–26. doi:10.1109/biorob.2010.5626010
1.4007695 Lee, C., and Oh, S. (2016).Configuration and Performance Analysis of a compact
De Luca, A., Siciliano, B., and Zollo, L. (2005). PD control with on-line gravity Planetary geared elastic Actuator. In IECON 2016-42nd Annual Conference of
compensation for robots with elastic joints: theory and experiments. the IEEE Industrial Electronics Society. (New Jersey, United States: IEEE).
Automatica 41, 1809–1819. doi:10.1016/j.automatica.2005.05.009 6391–6396. doi:10.1109/iecon.2016.7793816
devices, C. (2020). 28 Mm, 12 Bit, Single-Turn, Absolute, Spi Interface, Capacitive Li, W., Han, Y., Wu, J., and Xiong, Z. (2020). Collision detection of robots based on
Modular Encoder (Lake Oswego: CUI devices). a force/torque sensor at the bedplate. IEEE/ASME Trans. Mechatronics 25,
Donadio, F., Frejaville, J., Larnier, S., and Vetault, S. (2018). “Artificial 2565–2573. doi:10.1109/tmech.2020.2995904
intelligence and collaborative robot to improve Airport operations,” in Losey, D. P., Erwin, A., McDonald, C. G., Sergi, F., and O’Malley, M. K. (2016). A
Online Engineering & Internet of Things (Springer) (Springer, Cham). time-domain Approach to control of series elastic Actuators: Adaptive torque
doi:10.1007/978-3-319-64352-6_91973–986. and Passivity-based impedance control. Ieee/asme Trans. Mechatron. 21,
Govindarajan, V., Bhattacharya, S., and Kumar, V. (2016). “Human-robot 2085–2096. doi:10.1109/tmech.2016.2557727
collaborative topological exploration for search and rescue Applications,” in MATLAB (2018). 9.5.0.944444 (R2018b). Natick, Massachusetts: The
Distributed Autonomous Robotic Systems (Berlin, Germany: Springer), 17–32. MathWorks Inc.).
doi:10.1007/978-4-431-55879-8_2 Mokaram, S., Aitken, J. M., Martinez-Hernandez, U., Eimontaite, I., Cameron, D.,
Hua, T. M., Sanfilippo, F., and Helgerud, E. (2019). “A robust two-feedback loops Rolph, J., et al. (2017). A ROS-integrated API for the KUKA LBR iiwa
Position control Algorithm for compliant low-cost series elastic Actuators,” collaborative robot * *The Authors Acknowledge support from the EPSRC
in 2019 IEEE International Conference on Systems, Man and Cybernetics Centre for Innovative Manufacturing in Intelligent Automation, in
(SMC) (New Jersey, United States: IEEE). 2384–2390. doi:10.1109/ undertaking this research work under grant reference number EP/I033467/1,
smc.2019.8913845 and the University of Sheffield Impact, Innovation and Knowledge Exchange
Human, F. I., and Cognition, M. (2020). The institute for Human & Machine grant "Human Robot Interaction Development". Equipment has been provided
Cognition (Ihmc) Pioneers Technologies Aimed at Leveraging and Extending under the EPSRC Great Technologies Capital Call: Robotics and Autonomous
Human Capabilities (Ocala: IHMC). Systems. IFAC-PapersOnLine 50, 15859–15864. doi:10.1016/j.ifacol.2017.08.2331

Frontiers in Robotics and AI | www.frontiersin.org 17 August 2021 | Volume 8 | Article 679304

Tuan et al. Robot Arm for Safe HRI

Negrello, F., Garabini, M., Catalano, M. G., Malzahn, J., Caldwell, D. G., Bicchi, A., Actuator for knee Assistance during overground walking. In 2012 4th IEEE
et al. (2015).A modular compliant Actuator for emerging high Performance and RAS & EMBS International Conference on Biomedical Robotics and
fall-resilient humanoids. In 2015 IEEE-RAS 15th International Conference on Biomechatronics (BioRob). New Jersey, United States: IEEE. 1931–1936.
Humanoid Robots (Humanoids). New Jersey, United States: IEEE). 414–420. doi:10.1109/biorob.2012.6290271
doi:10.1109/humanoids.2015.7363567 Stmicroelectronics (2020). Nucleo-f446re development board with
nhatbon (2020). Dc H216 15a H Bridge. stm32f446re mcu
Paine, N., Mehling, J. S., Holley, J., Radford, N. A., Johnson, G., Fok, C.-L., et al. Talole, S. E., Kolhe, J. P., and Phadke, S. B. (2010). Extended-state-observer-based
(2015). Actuator control for the nasa-jsc valkyrie humanoid robot: A decoupled control of flexible-joint system with experimental validation. IEEE Trans. Ind.
dynamics Approach for torque control of series elastic robots. J. Field Robotics Electron. 57, 1411–1419. doi:10.1109/tie.2009.2029528
32, 378–396. doi:10.1002/rob.21556 Tsardoulias, E., and Mitkas, P. (2017). Robotic Frameworks, Architectures and
Penzlin, B., Enes Fincan, M., Li, Y., Ji, L., Leonhardt, S., and Ngo, C. (2019). “Design Middleware Comparison. arXiv preprint arXiv:1711.06842 (arXiv).
and Analysis of a clutched Parallel elastic Actuator,” in Actuators (Basel, Vanderborght, B., Albu-Schaeffer, A., Bicchi, A., Burdet, E., Caldwell, D. G.,
Switzerland, Multidisciplinary Digital Publishing Institute), vol. 8, 67. Carloni, R., et al. (2013). Variable impedance Actuators: A review. Robotics
doi:10.3390/act8030067 autonomous Syst. 61, 1601–1614. doi:10.1016/j.robot.2013.06.009
Pérez-Ibarra, J. C., Alarcón, A. L. J., Jaimes, J. C., Ortega, F. M. E., Terra, M. H., and Wang, M., Sun, L., Yin, W., Dong, S., and Liu, J. (2015).A novel sliding mode
Siqueira, A. A. (2017).Design and Analysis of force control of a series elastic control for series elastic Actuator torque tracking with an extended disturbance
Actuator for impedance control of an Ankle rehabilitation robotic Platform. In observer. In 2015 IEEE International Conference on Robotics and Biomimetics
2017 American Control Conference (ACC). (New Jersey, United States: IEEE). (ROBIO). (New Jersey, United States: IEEE. 2407–2412. doi:10.1109/
2423–2428. robio.2015.7419699
Peshkin, M. A., Colgate, J. E., Wannasuphoprasit, W., Moore, C. A., Gillespie, R. B., Wang, M., Sun, L., Yin, W., Dong, S., and Liu, J. (2017). Continuous robust control for
and Akella, P. (2001). Cobot Architecture. IEEE Trans. Robot. Automat. 17, series elastic Actuator with unknown Payload Parameters and external disturbances.
377–390. doi:10.1109/70.954751 Ieee/caa J. Autom. Sinica 4, 620–627. doi:10.1109/jas.2017.7510610
Plooij, M., Wisse, M., and Vallery, H. (2016). Reducing the energy consumption of Wannasuphoprasit, W., Gillespie, R. B., Colgate, J. E., and Peshkin, M. A.
robots using the bidirectional clutched Parallel elastic Actuator. IEEE Trans. (1997).Cobot control. In Proceedings of International Conference on
Robot. 32, 1512–1523. doi:10.1109/tro.2016.2604496 Robotics and Automation), 4. New Jersey, United States: IEEE, 3571–3576.
Pratt, G. A., Williamson, M. M., Dillworth, P., Pratt, J., and Wright, A. (1997). Xiao, J., Zhang, Q., Hong, Y., Wang, G., and Zeng, F. (2018). Collision detection
“Stiffness isn’t everything,” in Experimental Robotics IV (Berlin, Germany: Algorithm for collaborative robots considering joint friction. Int. J. Adv.
Springer), 253–262. Robotic. Syst. 15, 1729881418788992. doi:10.1177/1729881418788992
Pratt, G. A., and Williamson, M. M. (1995). Series elastic Actuators. In Proc. of the XYTmotor (2020). 12v Dc Motor Servo Encoder (Guangdong: XYTmotor).
IEEE/RSJ International Conference on Intelligent Robots and Systems, Yin, W., Sun, L., Wang, M., and Liu, J. (2016). Position control of series elastic Actuator based
Pittsburgh, PA (IEEE). vol. 1, 399–406. on feedback linearization and rise method. In Proc. of the IEEE International Conference
Pratt, J., Krupp, B., and Morse, C. (2002). Series elastic Actuators for high fidelity on Robotics and Biomimetics (ROBIO), Qingdao, China (IEEE). 1203–1208.
force control. Ind. Robot: Int. J 29 (3), 234–241. doi:10.1108/ Yu, H., Cruz, M. S., Chen, G., Huang, S., Zhu, C., Chew, E., et al. (2013a).
01439910210425522 Mechanical design of a Portable knee-Ankle-foot robot. In Proc. of the IEEE
Quigley, M., Conley, K., Gerkey, B., Faust, J., Foote, T., Leibs, J., et al. (2009). “ROS: International Conference on Robotics and Automation (ICRA), Karlsruhe,
an open-source robot operating system,” in Proc. of the IEEE International Germany (IEEE). 2183–2188. doi:10.1109/icra.2013.6630870
Conference on Robotics and Automation (ICRA), workshop on open source Yu, H., Huang, S., Chen, G., Pan, Y., and Guo, Z. (2015). Human-Robot Interaction
software, Kobe, Japan (IEEE). vol. 3, 5. Control of Rehabilitation Robots With Series Elastic Actuators. IEEE Trans.
Radford, N. A., Strawser, P., Hambuchen, K., Mehling, J. S., Verdeyen, W. K., Robot. 31, 1089–1100. doi:10.1109/tro.2015.2457314
Donnan, A. S., et al. (2015). Valkyrie: NASA’s First Bipedal Humanoid Robot. Yu, H., Huang, S., Chen, G., and Thakor, N. (2013b). Control design of a novel
J. Field Robotics 32, 397–419. doi:10.1002/rob.21560 compliant Actuator for rehabilitation robots. Mechatronics 23, 1072–1083.
Rao, K. G., Reddy, B. A., and Bhavani, P. D. (2010). Fuzzy Pi and integrating type doi:10.1016/j.mechatronics.2013.08.004
fuzzy Pid controllers of linear, nonlinear and time-delay systems. Int. J. Comp. Yu, H., Huang, S., Thakor, N. V., Chen, G., Toh, S.-L., Cruz, M. S., et al. (2013c). A
Appl. 1, 41–47. doi:10.5120/144-264 novel compact compliant Actuator design for rehabilitation robots. In Proc. of
Sanfilippo, F., Azpiazu, J., Marafioti, G., Transeth, A. A., Stavdahl, Ø., and the IEEE 13th International Conference on Rehabilitation Robotics (ICORR),
Liljebäck, P. (2016). A review on Perception-driven obstacle-Aided Seattle, WA (IEEE). 1–6. doi:10.1109/icorr.2013.6650478
locomotion for snake robots. In Proc. of the 14th International Conference Zollo, L., Siciliano, B., De Luca, A., and Guglielmelli, E. (2007). PD control with on-line
on Control, Automation, Robotics and Vision (ICARCV), Phuket, Thailand. gravity compensation for robots with flexible links. In Proc. of the IEEE European
1–7. doi:10.1109/icarcv.2016.7838565 Control Conference (ECC), Kos, Greece (IEEE). 4365–4370. doi:10.23919/
Sanfilippo, F., Azpiazu, J., Marafioti, G., Transeth, A., Stavdahl, Ø., and Liljebäck, P. ecc.2007.7068344
(2017). Perception-Driven Obstacle-Aided Locomotion for Snake Robots: The
State of the Art, Challenges and Possibilities †. Appl. Sci. 7, 336. doi:10.3390/ Conflict of Interest: The authors declare that the research was conducted in the
app7040336 absence of any commercial or financial relationships that could be construed as a
Sanfilippo, F., Helgerud, E., Stadheim, P. A., and Aronsen, S. L. (2019b).Serpens, a potential conflict of interest.
low-cost snake robot with series elastic torque-controlled Actuators and a
screw-less Assembly mechanism. In Proc. of the IEEE 5th International Publisher’s Note: All claims expressed in this article are solely those of the authors
Conference on Control. Beijing, China: Automation and Robotics (ICCAR). and do not necessarily represent those of their affiliated organizations, or those of
133–139. doi:10.1109/iccar.2019.8813482 the publisher, the editors and the reviewers. Any product that may be evaluated in
Sanfilippo, F., Helgerud, E., Stadheim, P. A., and Aronsen, S. L. (2019a). Serpens: a this article, or claim that may be made by its manufacturer, is not guaranteed or
highly compliant low-cost ROS-based snake robot with series elastic Actuators, endorsed by the publisher.
stereoscopic vision and a screw-less Assembly mechanism. Appl. Sci. 9. , 2019a
Art. no. 396. doi:10.3390/app9030396 Copyright © 2021 Tuan, Sanfilippo and Hao. This is an open-access article
Sanfilippo, F., Stavdahl, Ø., and Liljebäck, P. (2018). SnakeSIM: a ROS-based distributed under the terms of the Creative Commons Attribution License (CC
control and simulation framework for Perception-driven obstacle-Aided BY). The use, distribution or reproduction in other forums is permitted, provided the
locomotion of snake robots. Artif. Life Robotics 23, 449–458. doi:10.1007/ original author(s) and the copyright owner(s) are credited and that the original
s10015-018-0458-6 publication in this journal is cited, in accordance with accepted academic practice.
Sergi, F., Accoto, D., Carpino, G., Tagliamonte, N. L., and Guglielmelli, E. No use, distribution or reproduction is permitted which does not comply with
(2012).Design and characterization of a compact rotary series elastic these terms.

Frontiers in Robotics and AI | www.frontiersin.org 18 August 2021 | Volume 8 | Article 679304