High-Velocity Walk-Through Programming for Industrial Applications:

A Safety-Oriented Approach

Simone di Napoli

, Mattia Bertuletti

, Mattia Gambazza

Matteo Ragaglia

, Cesare Fantuzzi

and Federica Ferraguti

Gaiotto Automation (SACMI Group), Via Toscana 1, Piacenza PC 29122, Italy

DISMI, University of Modena and Reggio Emilia, Department of Sciences and Methods for Engineering,

Via Amendola 2, Pad. Morselli - 42122 Reggio Emilia, Italy

Keywords:

Physical Human-Robot Interaction, Cooperative Robotics, Admittance Control, Walk-Through Programming.

Abstract:

Traditionally, industrial robots are programmed by highly specialized workers that either directly write code

in platform-speciﬁc languages, or use dedicated hardware (teach-pendant) to move the robot through the de-

sired via-points. In the last years, new strategies to manually move the robot through the human input had

been introduced. During the human-robot interaction, the most limitation of this kind of use of the robot is

the velocity reduction of the machine. Taken into account the introduction of sensor-based walk-through pro-

gramming approaches as the ideal solution to reduce programming complexity and time, this paper proposes

a safety architecture for walk-through programming of industrial manipulators speciﬁcally designed in order

to reach high velocities while guaranteeing the operator’s safety. The proposed solution is validated on an

industrial manipulator.

1 INTRODUCTION

In the recent years the paradigm for robot usage has

radically changed, moving from the idea of com-

plete workspace segregation (achieved through phys-

ical barriers) to a scenario in which robots and hu-

mans share the same workspace and even collaborate

side by side (Michalos et al., 2021). In this con-

text, robots are becoming key elements in increas-

ing productivity, since continuous and fruitful phys-

ical human-robot interaction (pHRI) can deﬁnitely

help to achieve the higher production ﬂexibility in or-

der to cope with limited volumes and rapidly chang-

ing product requirements. Nevertheless, widespread

adoption of robotic technologies is still undermined

by certain well-known factors, among which the in-

herently complex and time-consuming nature of robot

programming surely plays a crucial role (Pan et al.,

2012).

In order to supply productivity and ﬂexibility ad-

vantages in pHRI, robot programming strategies usu-

ally deﬁned as “walk-through programming” (also

referenced as “lead-through programming”, or “man-

ual guidance”) have been proposed (Ang et al., 2000),

with practical applications ranging from spraying

(Ferretti et al., 2009) to welding (Ang et al., 1999).

“walk-through programming” strategies are charac-

terized by the fact that the human operator manually

moves the robot directly interacting with the robot

end-effector. Pararelly, robot records its trajectory

and, after the teaching phase, this can be reproduced.

From the control point of view, the ﬁrst appli-

cations of this programming paradigm were com-

pletely passive and relied on backdrivable actuators.

Mechanical compensation was typically provided, by

means of either hydraulic or pneumatic cylinders, in

order to help the operator lift and move the robot.

An example of this technology is represented by the

Gaiotto GA-2000 manipulator (Gaiotto Automation

Spa, ).

With the introduction of “manual guidance”

strategies, an even greater relevance is attributed to

robot safety standards, which have been updated to

address co-working scenarios as walk-through pro-

gramming. In particular, industrial settings need to

comply with strict safety requirements, given by the

standards ISO 10218-1 (ISO, 2011a) and ISO 10218-

2 (ISO, 2011b) and the most recent technical speciﬁ-

cation ISO/TS 15066 (cit, 2015). The ISO/TS 15066

provides the Hand-guiding interaction method where

the human could exploit the manually movements to

teach a machine on how to correctly complete his

di Napoli, S., Bertuletti, M., Gambazza, M., Ragaglia, M., Fantuzzi, C. and Ferraguti, F.

High-Velocity Walk-Through Programming for Industrial Applications: A Safety-Oriented Approach.

DOI: 10.5220/0012250900003543

In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023) - Volume 1, pages 503-510

ISBN: 978-989-758-670-5; ISSN: 2184-2809

503

work. Unfortunatly, the standards severely limits

the robot performances during the human interaction;

the velocity is controlled at an appropriately reduced

speed. Indeed, for many industrial application, even

during the teaching phase, the high velocity of the

process is important to see a real result. Therefore,

for these cases, the hand-guiding method should be

too liming.

1.1 Related Works

Fro a purely functional point of view, walk-through

programming architectures rely on two key elements:

a sensor system and an admittance (or impedance)

control algorithm (Villani and De Schutter, 2008).

The sensor system is responsible for measuring the

interaction forces/torques exerted by the human oper-

ator on the manipulator, while the control algorithm

ensures that the robot responds to the operator’s input

accordingly.

Assuming that an architecture like this is avail-

able, another critical issue that still needs to be ad-

dressed is the unavoidable trade-off between safety

and performance. In particular, (cit, 2015) speciﬁes

restrictions regarding the transferred energy, the bio-

mechanical limits and the robot velocity allowed dur-

ing transient and quasi-static contact. Among these

requirements, the most relevant one with respect to

the presented implementation of walk-through pro-

gramming applied to industrial robots that need to ex-

ecute continuous trajectories at very high velocities is

the Cartesian velocity limit of 250 mm/s.

Indeed, though this limitation may be suitable

for scenarios in which only via-points need to be

memorized (Ragaglia et al., 2016), it may prevent

walk-through programming to be used when contin-

uous trajectories performed at high-velocity need to

be recorded. For instance, spraying robots cannot

be manually guided at low Cartesian velocities, since

their motion needs to be synchronized with the spray-

ing system set-points that cannot be kinematically

scaled with respect to time.

In principle, solutions based on tele-operation

(Tafazoli et al., 2002) (Tanzini et al., 2016) could be

considered as a safer alternative to walk-through pro-

gramming, since they do not require pHRI. Neverthe-

less, the lack of direct interaction needs to be compen-

sated by adding tele-presence features such as haptic

feedback (Jacinto-Villegas et al., 2017) and remote

vision (Tripicchio et al., 2017), thus leading to sig-

niﬁcant equipment cost increases that can be justiﬁed

only when working in highly dangerous contexts such

as building demolition, decommissioning of nuclear

power plants, disaster recovery, etc.

1.2 Contribution Statement

As already pointed out, safety regulations establish

requirements that may prevent walk-through pro-

gramming to be used when continuous trajectories

performed at high-velocities need to be recorded.

However, in industrial applications such as spray-

ing, the robot programming has to be recorded at

high-velocities without any kinematic scaling. Con-

sequently, a control architecture which satisﬁes the

safety regulations but allows the execution of trajec-

tories at high-velocities needs to be developed.

To this regard, this work introduces a novel safety

control architecture for walk-through programming

that combines traditional safety checks with advanced

monitoring functionalities of both the robot and the

human operator in order to ensure their safety when

recording high-velocity continuous trajectories. For

the sake of completeness, a more detailed explanation

of the proposed solution is given in (Ferraguti et al.,

2023).

2 SAFETY LOGIC DESIGN

Walk-through programming is without a doubt one of

the clearest examples of physical human-robot inter-

action, it consists of two different phases:

• Teaching Phase: the human operator physically

guides the robot end-effector to teach the trajec-

tory to be executed, while the robot controller

records all the signiﬁcant poses of the trajectory

itself;

• Execution Phase: the robot plays the continuous

trajectory back.

To this regard, this work focuses on the develop-

ment of a strategy that enables human operators to

record continuous trajectories performed at high ve-

locity both at the joint and at the Cartesian space

level. Consequently, the fact that mechanical haz-

ards are the most signiﬁcant ones to be taken into

account from the safety point of view comes at no

surprise. More speciﬁcally, among the various poten-

tial consequences of the several mechanical hazards

listed by (ISO, 2011a), we took into consideration two

main scenarios: human-robot impact, and entangle-

ment/trapping of a generic operator’s body part.

In order to obtain an effective trade-off between

safety and productivity, in this paper we propose to

design a control architecture that tackles safety issues

by following a two-fold approach:

• Basic Safety Functions: upper bounds on both

joint and Cartesian space dynamic quantities are

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

504

established and the safety controller checks if

these bounds are violated during the teaching

phase;

• Dynamic Human-Robot Monitoring: the safety

controller monitors in real-time the relative dis-

tance and the relative velocity of the robot with

respect to the human operator. Threshold values

are established in order to identify situations char-

acterized by a high risk of impact and/or entangle-

ment.

In both cases the safety controller checks a logic con-

dition. If both conditions are true, the programming

phase can proceed, while if at least one condition is

evaluated as false, an emergency stop is issued by the

safety controller and the manipulator is stopped.

2.1 Basic Safety Functions

As already mentioned, basic safety functions can be

implemented in order to force an emergency stop

whenever a speciﬁc dynamic quantity (at both joint or

Cartesian space level) exceeds a prescribed threshold

value. In this work three different basic safety func-

tions have been considered, each one consisting in

checking a speciﬁc dynamic quantity against a maxi-

mum allowed positive value, deﬁned as follows:

• maximum allowed Cartesian Linear Speed at TCP



˙x



∈ R, where x

and ˙x

correspond to the

positional portion of x and ˙x, respectively;

• maximum allowed Joint Velocities ˙q

∈ R

;

• maximum allowed Joint Accelerations ¨q

∈ R

As far as velocities are concerned, the TCP linear

speed threshold depends on the speciﬁc task and it

corresponds to the maximum speed needed to cor-

rectly record the continuous trajectory that will be di-

rectly converted into a robot program.

Moving to joint velocity bounds, suitable thresh-

old values can be established by taking into account

a reference conﬁguration of the manipulator q ∈ R

and by computing, for each joint, the angular velocity

˙q

, with k = 1,...,m, that would result in the maxi-

mum allowed TCP linear speed ˙x

(which is always

positive) via multiplication by the geometric Jacobian

matrix J (·).

˙q

← min





˙q

Max

˙x

∑

w=1

(J (q))

w,k





(1)

Saturation with respect to absolute joint velocity lim-

its ( ˙q

Max

) must be taken into account to guarantee that

the safety check is effectively enforced. In this way,

this safety function acts as a redundant check with re-

spect to TCP speed monitoring.

Finally, once joint velocity bounds have been de-

ﬁned, the corresponding accelerations bounds can be

computed on the basis of the robot stopping time ∆t.

Once again, saturation with respect to absolute joint

acceleration limits ( ¨q

Max

) must be taken into account

to guarantee that the safety check is effectively en-

forced.

¨q

← min



¨q

Max

˙q

∆t



(2)

For the sake of completeness, joint velocities and ac-

celerations are checked against the discussed max-

imum values by means of their absolute value, so

that either positive and negative values are properly

checked. Consequently, the comprehensive logic con-

dition computed by the basic safety functions can be

expressed as follows:

∥

˙x

∥

≤



˙x



∧

| ˙q

| ≤ ˙q

∀ k = 1,...,m ∧

| ¨q

| ≤ ¨q

∀ k = 1,...,m

(3)

2.2 Dynamic Human-Robot Monitoring

As introduced before, in addition to the basic safety

functions, the proposed safety controller is endowed

with the capability to monitor in real-time the relative

distance and the relative velocity of the robot with re-

spect to the human operator. This way, the controller

can identify situations characterized by a high risk of

impact/clamping and issue an emergency stop accord-

ingly.

In order to perform these functionalities, a strat-

egy to model the space occupancy of both the opera-

tor and the robot is needed. To this regard, previous

contributions in the ﬁeld of safe human-robot interac-

tion (Ragaglia et al., 2015) have already explored the

application of capsule-based geometry models to the

problem of modeling the space occupancy of complex

objects, proving that this approach can be extremely

efﬁcient in approximating structured geometries. For

the sake of clarity, a capsule consists in the convex

hull of a sphere of given radius (also named “clear-

ance”), which is translated along a segment. For in-

stance, Fig. 1 shows two distinct capsules. The ﬁrst

one on the left is deﬁned by points P

1,0

, P

1,1

and ra-

dius r

, while the second one on the right is deﬁned by

2,0

, P

2,1

and radius r

. As exempliﬁed in the ﬁgure,

the minimum distance between the two capsules d can

be obtained by computing the minimum distance d

High-Velocity Walk-Through Programming for Industrial Applications: A Safety-Oriented Approach

505

between the segments P

1,0

1,1

and P

2,0

2,1

(as thor-

oughly explained in (Ericson, 2005)), and then sub-

tracting the two clearances r

and r

. Clearly, when-

ever d ≤ 0, the capsules are colliding.

Figure 1: Geometry representation of two separate capsules

and of the minimum distance between them (Ferraguti et al.,

2023).

In this work, capsules are used to model the space

occupancy of the robot and of the human operator, as

it is shown in Fig. 2. As far as the robot is concerned,

a separate capsule is deﬁned for each link and clear-

ances are assigned on the basis of the links’ geometry.

Moving to the human operator, their space occupancy

can be modeled by means of a single vertically ori-

ented capsule, whose radius and height can be param-

eterized according to the anthropometric features of

the speciﬁc operator. In addition, the operator’s cap-

sule is rigidly attached to the robot’s end-effector ac-

cording to the geometry of the handle that allows to

move the robot.

Figure 2: Example of operator’s capsule and capsule-based

geometry model of a 6 DoF offset wrist manipulator.

In principle, a more complex model could be

taken into account for the human operator, by con-

sidering all the pairs composed by a robot link and

a human link that can be computed dynamically as

proposed in (Ferraguti et al., 2020). To this aim a

tracking system that allows to detect and monitor the

current position of each link of the human operator

is required to be integrated, by following the strate-

gies proposed in (Ragaglia et al., 2018). However,

the main reason behind our choice lies in the fact that

we are interested in preventing collisions by ensuring

that there is a sufﬁcient separation distance between

the robot and the operator, rather than precisely iden-

tifying the operator’s body part involved in a colli-

sion. In addition, given the nature of the task, some

body parts (i.e. hands, wrists, and possibly forearms)

would always result in collision (or at least in close

proximity) with the robot, thus making this level of

detail not necessary.

During the execution of the teaching phase, the

safety controller updates in real-time the position of

all the capsules, computes the relative distance be-

tween each robot capsule and the operator’s one (like

it is shown in Fig. 1), and checks that the minimum

relative distance remains above a parameter threshold.

By setting this threshold equal to the distance that the

robot can cover at the maximum allowed linear speed

˙x

during the time needed to enforce an emergency

stop, the safety controller is able to prevent collisions

between the robot and the operator.

Having covered the strategy that allows to esti-

mate the relative distance between the robot and the

operator, let us focus on the monitoring of the relative

velocity between them. The usage of relative human-

robot velocity in combination with relative human-

robot distance for safety purposes has been proposed

as a safety metric by several contributions in the ﬁeld

of safe human-robot interaction, like for instance (Ra-

gaglia et al., 2014). In this work, relative human-robot

velocity is monitored in order to identify situations

where at least a single robot link is moving towards

the operator with enough speed, that an emergency

stop may not be able to prevent a collision. To this

purpose, a reasonable choice for the threshold value

is represented by the maximum allowed linear speed

˙x

Once again, the capsule-based geometry models

of both the manipulator and of the operator represent

the starting point. More speciﬁcally, the relative ve-

locity between the robot and the operator is deﬁned

as the maximum relative velocity between the robot

capsules’ end-points and the operator’s capsule.

Let us consider Fig. 3, where the solid grey bar

represents the i-th robot link, surrounded by the cor-

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

506

responding capsule robCap

. On the other hand, the

operator capsule opCap is pictured in the same way

around its axis, passing through opCap.P. For both

capsules clearances robC ap

.r and opCap.r are also

highlighted. By means of kinematic calculations, the

linear velocities vel

and vel

can be easily com-

puted, given both joint angles and joint velocities.

Then, by projecting velocity vel

(vel

) along the di-

rection that connects point robCap

(robCap

)

to the axis of the operator capsule, we can compute

the relative velocity of that speciﬁc point with respect

to the operator. It is worth mentioning that the re-

sult of this projection is a signed velocity value, that

is positive if the link end-point is moving towards the

operator, and negative when it is moving away from

the operator.

It is also worth mentioning that, as stated in (Ra-

gaglia et al., 2015), linear velocity varies linearly be-

tween the link end-points. As a result, either point

robCap

or robCap

is necessarily the link point

characterized by the maximum linear velocity. Since

the projection with respect to the operator’s capsule

consists in a linear combination of the linear velocity

coordinates, once again we can state that either point

robCap

or robCap

is necessarily the link point

characterized by the maximum relative velocity with

respect to the operator, thus making it not necessary

to check intermediate points.

Finally, please note that point robC ap

of the i-

th link corresponds to point robCap

of the (i + 1)-

th link. In addition, since the robot base (which typ-

ically corresponds to point robCap

of the ﬁrst

robot link) is normally still, the maximum relative ve-

locity can be obtained by projecting the linear veloc-

ities of each end-point robCap

of the robot links.

Figure 3: Geometric representation of the relative velocity

between the end-points of a generic robot link (solid grey

bar) and the operator’s capsule.

As a result, the comprehensive logic condition

checked by the dynamic human-robot monitoring

block can be expressed as follows:

relDistOk = 1 ∧ relVelOk = 1 (4)

3 IMPLEMENTATION

In order to validate the proposed control architecture,

we implemented the high-speed walk-through pro-

gramming strategy described in the previous sections

in a practical industrial use case. In particular, the

strategy has been implemented on the Gaiotto GA-OL

manipulator shown in Fig. 4, which is a 6 DoF indus-

trial robot designed for spraying applications (Gaiotto

Automation Spa, ). Typically, the GA-OL robot is en-

dowed with an auxiliary turning table, on top of which

the object to be sprayed is loaded and then rotated in

order to help the robot execute the spraying programs.

Figure 4: Gaiotto GA-OL manipulator (on the right, pic-

tured in blue) and auxiliary turning table (on the left, pic-

tured in grey) (Ferraguti et al., 2023).

The implementation of the safety logic lies be-

tween the safety and functional environments, as de-

picted in Fig. 5. More speciﬁcally, the basic safety

functions (see subsection 2.1) are implemented within

the “Safety Logic” domain. On the other hand, the

dynamic human-robot monitoring functionalities (see

subsection 2.2) are executed within the “Functional

Logic” domain, which is further divided into two dis-

tinct environments: the “C++ Environment” and the

“IEC Environment”. The ﬁrst one comprises a series

of libraries written in C/C++ that implement the func-

tional counterpart of the safety controller proposed in

this paper, while the latter hosts components written

in either Structured Text or FBD, like for instance

the aforementioned dynamic human-robot monitor-

ing functionalities. The communication between the

functional domain and the safety domain is realized

by means of “virtual digital IOs”, i.e. boolean vari-

ables that are exchanged between the two domains via

shared memory.

High-Velocity Walk-Through Programming for Industrial Applications: A Safety-Oriented Approach

507

Figure 5: General control architecture (Ferraguti et al.,

2023).

The Safety Logic collects all the information

needed to compute the Safe Torque Off (STO) sig-

nal, that cuts-off the electric power to the axis motors

(thus preventing them to develop torque) whenever it

is de-activated by the Safety Logic. As far as param-

eters are concerned, the following values have been

used:

•



˙x



= 1.30 m/s - maximum linear speed at

TCP, due to spraying task requirements

;

• ˙q

Max

= [200,200,200,300,300,300] deg/s -

maximum GA-OL joint velocities;

• ¨q

Max

= [450,450,450,450,450,450] deg/s

maximum GA-OL joint accelerations;

• robRad = [0.25, 0.20, 0.10,0.075,0.05,0.05,

0.075] m - robot and tool capsules radii;

• opRelP = [0.000,0.000,0.250] m - operator cap-

sule position with respect to GA-OL end-effector.

Relative orientation is considered null;

• opRad = 0.200 m - operator capsule radius;

• opH = 1.80 m - operator capsule height;

• stopT = 0.35 s - robot stopping time;

• minDistT h = 0.455 m - minimum allowed dis-

tance between operator and robot capsules. It is

obtained by multiplying stopT by ˙x

;

• maxVelT h = 1.30 m/s - maximum allowed rela-

tive velocity between operator and robot capsules.

Given stopT and minDistT h, it can be set equal to

˙x

The operator capsule radius and height have been em-

pirically selected as the medium value between the

real characteristics of the testers involved in the ex-

perimental campaign.

The value of the velocity limit has been selected as the

maximum value of the TCP reached by the robot in 40 dif-

ferent programs executed on a GA-2000 manipulator, which

is equipped with mechanical compensation systems for pas-

sive walk-through programming.

4 EXPERIMENTAL RESULTS

An extensive experimental evaluation phase has been

performed in order to demonstrate the effectiveness of

the proposed architecture, whose results are hereby

presented and discussed. For the sake of complete-

ness, experiments involved several testers selected

among Gaiotto employees.

First, let us show how the violation of the condi-

tions deﬁned by the basic safety functions results in

disabling the STO and stopping the robot. Figure 6(a)

shows that as soon as the TCP linear velocity limit

( ˙x

= 1.30 m/s) is exceeded, the STO signal (prop-

erly scaled) is disabled and the robot stops. In addi-

tion, the magniﬁcation shown in Fig. 6(b) also proves

that the robot completely stops within the declared

stopping time equal to 0.35s.

(a)

(b)

Figure 6: Violation of TCP linear velocity limit and corre-

sponding variation of scaled STO signal, with magniﬁcation

between 1.70s and 2.10s.

On the other hand, Figures 7 and 8 shows viola-

tions related to the dynamic human-robot monitoring

algorithms. Figure 7(a) shows how the violation of

the minimum relative distance condition triggers the

disabling of the STO by the safety controller, while in

Fig. 7(b) TCP linear velocity is depicted. The very

same behaviour characterizes the safety controller re-

sponse to a relative velocity violation, as it is shown

in Fig. 8.

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

508

(a)

(b)

Figure 7: Violation of the relative human-robot minimum

distance with TCP linear velocity and corresponding vari-

ation of scaled STO signal, with magniﬁcation between

59.25s and 59.75s.

4.1 Usability Evaluation

In order to properly evaluate the usability of the pro-

posed walk-through programming architecture in ac-

tual industrial scenarios, the selected testers have been

asked to answer a questionnaire right after the tests.

More in detail, testers were asked to rate several in-

dicators, by assigning each one a score ranging from

“0” (meaning either “null” or “extremely negative”),

to “10” (meaning either “extremely high” or “ex-

tremely positive”).

Figure 9 shows the results corresponding to two

distinct indicators: perceived level of safety, and in-

duced stress. More speciﬁcally, a very high overall

score has emerged with respect to perceived safety

9(a). Please also notice that the low score regarding

to induced stress 9(b) corresponds to a positive eval-

uation by the testers. Finally, we can state that the

proposed architecture represents a promising solution

not only from the theoretical point of view, but also

from the application perspective.

(a)

(b)

Figure 8: Violation of the relative human-robot maximum

velocity with TCP linear velocity and corresponding vari-

ation of scaled STO signal, with magniﬁcation between

3.50s and 3.80s.

(a)

(b)

Figure 9: Evaluation of general indicators. For each indi-

cator, the box-whiskers plot on the left shows the approx-

imated distribution of the rates assigned by the testers, by

highlighting the median (solid orange line) and the mean

(green triangle). Then, the histogram on the right displays

the normalized frequency of the assigned rates.

High-Velocity Walk-Through Programming for Industrial Applications: A Safety-Oriented Approach

509

5 CONCLUSIONS & FUTURE

DEVELOPMENTS

In this paper, the authors propose an innovative con-

trol architecture for walk-through programming of an

industrial manipulator. The proposed solution aims at

allowing human operators to program industrial ma-

nipulators by directly teaching high-speed trajecto-

ries, while guaranteeing the operator’s safety. A ded-

icated safety controller has been developed to moni-

tor the kinematic conﬁguration of the manipulator and

stop its motion whenever the risk of a collision with

the human operator becomes too high.

Given the results of the experimental validation,

the authors can state that the proposed architec-

ture successfully achieves a fruitful trade-off between

safety and productivity, by guaranteeing the opera-

tors’ safety and the possibility to directly record high-

velocity trajectories.

As far as future developments are concerned, the

authors foresee to integrate a tracking system to detect

and monitor in real-time the current position of each

link of the human operator, in order to obtain very

accurate occupancy volumes.

ACKNOWLEDGMENT

This work is partly supported by Regione Emilia-

Romagna under the agreement “Innovazione della

value proposition di Gaiotto Automation 2021-2025”

(grant number / CUP: E32C21001090009).

REFERENCES

(2015). Iso/ts 15066 - robots and robotic devices – collabo-

rative robots.

Ang, M. H., Lin, W., and Lim, S. (1999). A walk-through

programmed robot for welding in shipyards. Indus-

trial Robot: An International Journal, 26(5):377–388.

Ang, M. H., Wei, L., and Lim Ser Yong (2000). An in-

dustrial application of control of dynamic behavior of

robots-a walk-through programmed welding robot. In

IEEE International Conference on Robotics and Au-

tomation (ICRA), volume 3, pages 2352–2357 vol.3.

Ericson, C. (2005). Real-Time Collision Detection. Else-

vier.

Ferraguti, F., Bertuletti, M., Gambazza, M., and Ragaglia,

M. (2023). High-velocity walk-through programming

for industrial applications. Robotics and Computer-

Integrated Manufacturing, 81:102505.

Ferraguti, F., Talignani Landi, C., Costi, S., Bonf

e, M.,

Farsoni, S., Secchi, C., and Fantuzzi, C. (2020).

Safety barrier functions and multi-camera tracking for

human-robot shared environment. Robotics and Au-

tonomous Systems, 124:103388.

Ferretti, G., Magnani, G., and Rocco, P. (2009). Assigning

virtual tool dynamics to an industrial robot through an

admittance controller. In International Conference on

Advanced Robotics (ICAR), pages 1–6.

Gaiotto Automation Spa. Gaiotto website. https://www.

gaiotto.com/. Last checked on Sept 06, 2023.

ISO (2011a). Iso-2011-10218-1 - robots and robotic de-

vices - safety requirements for industrial robots - part

1: Robots.

ISO (2011b). Iso-2011-10218-2 - robots and robotic de-

vices - safety requirements for industrial robots - part

2: Robot systems and integration.

Jacinto-Villegas, J. M., Satler, M., Filippeschi, A., Berga-

masco, M., Ragaglia, M., Argiolas, A., Niccolini,

M., and Avizzano, C. A. (2017). A novel wearable

haptic controller for teleoperating robotic platforms.

IEEE Robotics and Automation Letters (IEEERAL),

2(4):2072–2079.

Michalos, G., Karagiannis, P., Dimitropulos, N., Andronas,

D., and Makris, S. (2021). Human Robot Collabora-

tion in Industrial Environments.

Pan, Z., Polden, J., Larkin, N., Van Duin, S., and Norrish, J.

(2012). Recent progress on programming methods for

industrial robots. Robotics and Computer-Integrated

Manufacturing, 28(2):87–94.

Ragaglia, M., Bascetta, L., Rocco, P., and Zanchettin, A. M.

(2014). Integration of perception, control and injury

knowledge for safe human-robot interaction. In IEEE

International Conference on Robotics and Automation

(ICRA), pages 1196–1202.

Ragaglia, M., Zanchettin, A. M., Bascetta, L., and Rocco,

P. (2016). Accurate sensorless lead-through program-

ming for lightweight robots in structured environ-

ments. Robotics and Computer-Integrated Manufac-

turing (RCIM), 39:9 – 21.

Ragaglia, M., Zanchettin, A. M., and Rocco, P. (2015).

Safety-aware trajectory scaling for human-robot col-

laboration with prediction of human occupancy.

In International Conference on Advanced Robotics

(ICAR), pages 85–90.

Ragaglia, M., Zanchettin, A. M., and Rocco, P. (2018).

Trajectory generation algorithm for safe human-robot

collaboration based on multiple depth sensor measure-

ments. Mechatronics (MECH), 55:267–281.

Tafazoli, S., Salcudean, S. E., Hashtrudi-Zaad, K., and

Lawrence, P. D. (2002). Impedance control of a tele-

operated excavator. IEEE Transactions on Control

Systems Technology (IEEECST), 10(3):355–367.

Tanzini, M., Jacinto-Villegas, J. M., Filippeschi, A., Nic-

colin, M., and Ragaglia, M. (2016). New interac-

tion metaphors to control a hydraulic working ma-

chine’s arm. In IEEE Symposium of Safety and rescue

Robotics (SSRR).

Tripicchio, P., Ruffaldi, E., Gasparello, P. S., Eguchi, S.,

Kusuno, J., Kitano, K., Yamada, M., Argiolas, A.,

Niccolini, M., Ragaglia, M., and Avizzano, C. A.

(2017). A stereo-panoramic telepresence system

for construction machines. Procedia Manufacturing

(PROCMAN), 11:1552 – 1559.

Villani, L. and De Schutter, J. (2008). Force Control, pages

161–185. Springer Berlin Heidelberg, Berlin, Heidel-

berg.

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

510