A Review of Safety Methods for Human-robot

Collaboration and a Proposed Novel Approach

Ansuri Reddy, Glen Bright and Jared Padayachee

Discipline of Mechanical Engineering, University of KwaZulu Natal, King George V Ave, Durban, South Africa

Keywords: Obstacle Avoidance, Trajectory Planning, Sensory Systems, Environment Mapping, Serial Manipulators.

Abstract: Industrial robots offer the advantage of flexible manufacturing and increased efficiency when paired with

human workers. However, this means breaking well-established safety procedures such as safety fences and

workspace separation. Robots present a danger to humans as they work at high speeds with sudden motions.

It is therefore necessary to ensure safe interaction during collaboration. This paper presents a collection of

sources that explain the trends and advances in the field of industrial robotics specifically to safety in human-

robot interaction. Major trends and popular methods lean towards obstacle avoidance using a sensory planning

method of polynomials and a sensory system that is able to map the robot workspace. The goal of these

methods is to ensure that the human is kept safe. These methods were used to develop a novel approach to

safe interactions. This approach uses a LIDAR sensor for obstacle detection and tracking.

1 INTRODUCTION

Industry 4.0 has brought with it advances and

advantages to factories and their methods of

production specifically in flexible and reconfigurable

manufacturing (Shiyong Wang, 2016). Industrial

Robots perform a variety of repeatable tasks at

consistent quality resulting in decreased waste and

production costs (Fryman and Matthias, 2012).

Factories are able to increase their production rate and

throughput of components that meet all the quality

standards leaving the humans to perform more

complex tasks. However, the flexibility of robot

systems is limited by programming and part feeding

challenges. A human worker resolves this challenge

by monitoring or working collaboratively with the

robot. Humans perform maintenance tasks, check the

quality of parts and set up the workspace of the

robots. This close working relationship requires the

robot to be aware of the human in its workspace.

Industrial robots are made of steel, are extremely

heavy and move at high speeds with sudden

movements. Without safety fences, these

characteristics make it dangerous for a human to be

in close proximity to an industrial robot as it performs

its tasks (D. Gao, 2009). The human could easily be

injured or killed by being hit with the robot arm or

struck with the work tool. To ensure human safety,

research is being conducted to discover ways of

keeping humans safe within a robot production

environment.

This paper presents the various safety methods

used in industry as well as approaches developed by

researchers. These methods are in accordance with

ISO/TS 15066:2016 which are the methods of

collaborative robotics. The safety method themes

presented in this paper are robot vision, obstacle

detection, obstacle avoidance, and trajectory

planning.

A novel approach to safe Human Robot

Interaction is discussed. The objectives of this

research study were to research human-robot

interaction, develop a sensory system for human

detection, algorithms for data processing and

predicting the location of the human in the workspace

and to develop algorithms that allow the robot to

modify its work routine in a safe, reactive manner.

This paper contribution is a literature review of the

state of the art methods of safety for humans in

industrial robot production environments. The

objective of safe interaction is to eliminate the risk of

collisions between the human and the robot.

Reddy, A., Bright, G. and Padayachee, J.

A Review of Safety Methods for Human-robot Collaboration and a Proposed Novel Approach.

DOI: 10.5220/0007840502430248

In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 243-248

ISBN: 978-989-758-380-3

243

2 COLLABORATIVE ROBOT

SAFETY METHODS

ISO/TS 15066:2016, the technical standard for

Human-Robot Interaction, states the basic safety

methods of collaborative robots for industrial

application. These methods are Hand Guiding, Speed

and Separation Monitoring, Power and Force

Limiting and Safety Rated Monitored Stop (Marvel,

2017). Hand Guiding allows the operator to transmit

motion commands by showing the robot physically

how to move when performing a task. The speed and

separation monitoring condition continually monitors

the proximity of the robot to the obstacle and

maintains a set distance away. This condition is the

most usable with regards to collision avoidance. The

safety monitored stop condition stops the robot before

the human enters the workspace. This can be

implemented as a trip switch when a human enters the

environment. Power and Force limiting is used to

ensure that the force felt by a human worker is very

small and does not injure the human in any way.

Industrial application collaborative robots have

been developed by companies such as Rethink

Robotics, Fanuc, Kuka, and ABB.

Rethink Robotics collaborative robot sawyer is a

high-performance single arm robot designed to work

on tasks that require high precision. Sawyer is a fully

integrated collaborative robot solution, embedded

with Cognex Vision System located in its arm. The

vision system combined with built-in force sensors

allow the robot to make adaptive decisions and work

precisely േ 1mm away from the human for safe

collaborative operation. Elastic actuators on each

joint minimise contact force(robotics, 2017). The

sawyer robot uses the Speed and Separation

monitoring and the Power and Force Limiting safety

method.

Fanuc’s collaborative model cr-35ia is aware of

its surroundings and stops safely when contact with a

human operator is detected. The Dual Check Safety

system can be set up to perform checks on Position,

Safe Zones, Speed and Cartesian Position. This

system decreases the amount of floor space needs for

safe operation and eliminates the need for

fences(Robotics, 2019). This robot also used the

Power and Force Limiting safety method of Human

Robot Collaboration.

Kuka LBR iiwa (intelligent industrial work

assistant) robot has been designed for close human

collaboration. This robot is light weight and able to

react quickly if human contact is detected by its joint

torque sensors(KUKA, 2014). Kuka LBR iiwa uses

the Power and Force Limiting technique.

ABB’s robot Roberta is a collaborative robot

designed to suit Small to medium size enterprises.

This agile, light weight robot features a camera vision

system that can detect the object in its gripper and is

able to decipher if it is a human hand or a tool. It is

also equipped with fingertip force sensors that slows

down or stops the robot when contact is detected with

a human(Robotiq, 2014). Roberta uses the Hand

Guiding and Power and Force Limiting techniques.

3 ROBOT VISION

To perform obstacle avoiding tasks the robot requires

complete awareness of its workspace and any

potential obstacles in the field. Robot awareness is

created by implementing a vision system that covers

the entire workspace. Vision systems such as stereo

cameras, RGB Vision systems, proximity sensors and

ultrasonic sensors provide decent awareness of the

robot workspace environment. The sensors detect the

presence of an obstacle in the environment. Obstacle

tracking data provides information about human

intention. This is achieved by tracking and

interpreting human motions and gestures (Billard and

Dillmann, 2006).

The workspace is created using grid

representation. The location of an obstacle within the

grid is communicated to the robot to facilitate

obstacle avoidance and trajectory planning.

Constructive solid geometry is primarily used to

model the robot workspace environment (Zacharias

et al., 2007).

Localisation and position are primarily based on

visual and positional data derived from the visual

sensors of the robot. There are two approaches that

exist: Continuous geometric mapping and Discrete

cell-based mapping. Continuous geometric mapping

represents the environment more accurately, while

Discrete cell-based mapping method represents the

environment in discrete cells that form a grid. Each

cell represents a square area of the environment and

stores a value that indicates the occupied state of that

area. (Marvel, 2017)

4 ROBOT CONTROL

Robot control is essential to ensure the robot performs

appropriate actions around humans in the workspace

and completes tasks while avoiding all collisions. It is

necessary for the robot to be programmable in order

to achieve a variety of tasks. Adaptability is also

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

244

necessary for the robot to modify its own behaviour,

meet its goal and have good reactivity to sudden

obstacles (Albu-Schäeffer et al., 2005).

4.1 Obstacle Detection

Obstacles within the workspace can be detected by

proximity or depth sensors. Ultrasonic sensors emit

an ultrasonic sound wave and receive the echo. The

time taken for the echo to return can be used to

calculate the distance of an obstacle away from the

sensor. Other sensors such as stereoscopic sensors use

3D imaging to create the illusion of depth offering the

advantage of depth perception of the environment

(Pérez et al., 2016). Infrared Sensors detect motion of

a human in the workspace and can only be used for

dynamic objects.

Capacitive sensors are used to detect the presence

of a human. These sensors are capacitors that detect

and track a human with good accuracy. The robot

manufactures at Fogale Robotics have used this

technology to create a skin that covers the robot to

detect a human at any point on its body. Scanning

range finders such as LIDAR or RADAR are capable

of detecting an obstacle as well as counting, locating

and tracking them. Cameras and imaging devises are

the most popular form of obstacle detection and

tracking in a robot workspace. They are also the most

affordable to implement. A depth camera captures the

depth of an object. A study by (Flacco et al., 2012)

has shown that a robot can avoid an obstacle easily

and consistently when using depth camera imaging.

4.2 Obstacle Avoidance

Obstacle avoidance algorithms instruct the robot to

avoid a collision with a human and the path planning

algorithm plans a new path to allow the manipulator

to contour around the human towards its goal

configuration. The kinematic calculations determine

the position and orientation of the new goal pose once

the path has been planned. Robot motion is restricted

due to singularities. The path planner considers the

singularities when planning the trajectory of the

robot.

Real-time collision avoidance methods allow for

safe human-robot interaction as the robot can adjust

itself as the human moves within the workspace. A

method described by (Fratu et al., 2010) continuously

measures the proximity between the robot and

dynamic obstacles and uses the data to generate

repulsive vectors. The vectors control the robot as it

manoeuvres and performs a task. An obstacle

avoidance algorithm by (Flacco et al., 2012) uses a

stereoscopic depth sensor to capture the environment.

The obstacle avoidance algorithm incorporates

different reactions that have been set up for the end

effector and other joints of the manipulator. This

allows the robot to react in a number of ways to avoid

an obstacle.

To avoid collisions with dynamic objects, free

space must be considered around these objects.

Collision avoidance is an important factor in Path

Planning. In a case where there is no automatic

collision avoidance, the robotic workspace must be

engineered to be collision-free another option is to

have sub-optimal paths which are selected by the

human programmer (Mohammed et al., 2017). The

feedback to the planner is used to pause paths where

collisions are imminent.

4.3 Trajectory Planning

Trajectory planning is fundamental to collision

avoidance techniques. To re-rout a robot and avoid an

obstacle the robot needs to search for a new path and

plan a trajectory that will allow it to contour around

the human and meet its goal without collision. The

term trajectory refers to a path that is executed in a

specific time interval. Trajectories can be planned

using a variety of equations to specify the shape of the

path. The most common being the polynomial

trajectory. Polynomial trajectories are commonly

formulated using a polynomial equation these

equations are iterative and plot a spline when the time

is specified. The coefficients calculated are the

velocity, acceleration and position values for the

trajectory (Letla, 2008 ). The degree of the

polynomial depends upon the complexity of the path.

A study by (Boryga et al., 2015) has used a

method called PR-RPT (Planning Rectilinear-Arc

Polynomial Trajectory) which is a method of

trajectory planning that links rectilinear lines that

intersect with a curve of a set radius. This method

uses a seven-degree polynomial so that the jerk is

equal to 0 when the robot transitions between speeds.

Seven types of path planners exist. These include

knowledge-based simple path planners, knowledge-

based hybrid path planners, sensor-based path

planners, static knowledge and sensor-based

hierarchical path planners, dynamic knowledge and

sensor-based path planners, path planners based on

offline programming, path planners based on online

programming.

A Review of Safety Methods for Human-robot Collaboration and a Proposed Novel Approach

245

4.4 Trajectory Planning Methods

There are three types of trajectory planning

algorithms: road map, cell decomposition and

potential field method. The road map method

represents the free configuration space and its

connectivity to other free paths. Cell decomposition

can be broken down into two aspects: exact cell

decomposition and approximated cell decomposition.

The Octree method is an approximated cell

decomposition method where the cells are

consecutively subdivided until there are no mixed

cells on the map (Sousa e Silva et al., 2013). A paper

by (Barcellini et al., 2012) defines a virtual wall

around an obstacle where a robot senses repulsive

forces which signal it to turn away from the obstacle.

The potential field method is the earliest and most

popular method used for obstacle avoidance. This

method is implemented by defining a potential field

of repulsive forces around the obstacle in the

workspace. The robot is able to sense this method and

adjust its posture to avoid the obstacle in the field.

Combining octree and potential field methods, the

manipulator is repelled by the obstacle and attracted

towards the goal configuration.

A study by (Leutert et al., 2012) implemented

Photonic Mixer Device that gathered depth

information of the robot environment. This

information was used to build an environment model

which served as an input for the path planning model.

The robot uses the data to autonomously select a path

to navigate towards the target position while avoiding

all static, dynamic and obstacles and continuously

optimising its trajectory.

A study by (Nieto et al., 2010) uses the RRT

(Rapidly-exploring Random Tree) method to plan a

path in a dynamic environment. New paths are

planned based on the mapped area of the robot

workspace. The main aspect of this research is

centred around an approach to formulating a cost

function for a motion planner for human-robot

collaboration. This method quantifies the consistency

of the robot’s motion so that it is predictable.

(Mišeikis et al., 2016) developed an algorithm that

avoids a human using lane differentiation. The robot

and human are represented in separate lanes and the

robot avoids the human without collision while

considering its tasks. This method was adapted from

aeronautic planes avoiding each other on runways.

Shorter and smoother trajectories where produced

when compared to reactive trajectory planning. A

method by (Jin et al., 2005) avoid obstacles while

positioning the end effector with on-line line collision

avoidance. The motion planning method is sensor

based and operates around unknown obstacles of

arbitrary shape. This method is an online collision

avoidance method that requires no prior knowledge

of the obstacles. A trajectory scaling algorithm for

safe Human-Robot Interaction that relies on a real-

time prediction of human occupancy was developed

by (Eder, 2014). By knowing the space that the

human will occupy and the robots stopping time, the

controller is able to scale the manipulator's velocity

for safe interaction.

4.5 Pre Collision Methods

Pre Collision methods are considered as preventative

methods and are intended to ensure safety during

Human-Robot Interaction. These methods are

implemented by monitoring the human and the robot

and then adjusting the robot controller according to

the feedback. The most common techniques and

methods are Quantitative Limits, Speed and

separation Monitoring and Potential Field Methods.

Quantitative Limits are described as a guarantee that

a robot cannot pose any threat to a human even if a

collision occurs. This is done by limiting parameters

such as joint velocity, energy and potential exertion

of force (Lasota et al., 2017).

When there is no human presence in the

workspace the motion can be maintained at the

maximum programmed speed. At this point, all tasks

and actions are taken as autonomous state behaviour.

When a human enters the workspace and is detected

by the robot the collaborative behaviour is activated.

The distance from the human to the robot manipulator

arm is constantly monitored. As the human

approaches the robot, the safety constraints cannot be

guaranteed while maintaining the production at

maximum level. The robot reduces its speed and

modifies its behaviour accordingly. If the robot

comes to a point where it could collide with the

human, the robot stopped. This should happen at 0

speed as the human is almost in contact with the

manipulator (Zanchettin et al., 2016).

5 SAFETY IN HUMAN-ROBOT

INTERACTION

This study contributes to the design of algorithms for

safe human-robot interaction. The proposed approach

is to develop a sensory system to detect humans in the

workspace of a robot and develop obstacle avoidance

and trajectory planning algorithms. The project will

be developed as described in Figure 1.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

246

Figure 1: Development of Research Project.

This novel approach uses a LIDAR sensor. The

LIDAR model that will be used is a Garmin v3 Lite.

This sensor has a range of 10mm to 40m and a

frequency of 50 to 500Hz (Garmin, 2016). The

LIDAR is mounted on a pan-tilt mechanism to detect

the presence of humans in the workspace. The

LIDAR sends and receives a laser signal. When

connected to a microcontroller the distance data is

displayed as a 3D point cloud. When a human is less

than the specified distance away from the robot, the

robot will slow down and begin planning a new

trajectory. This approach is unique as LIDARs are

traditionally used to detect and track humans and

objects for mobile robotic applications. For this case,

it will be applied to a serial industrial robot

application.

The pan-tilt motion of the mechanism is created

using servo motors. The servo motor position is fed

into the control system and the location of the

obstacle is found using trigonometry.

The kinematics of the robot is calculated using

Peter Cork’s Toolbox. The algorithms will be

implemented on a six degree of freedom serial arm

industrial robot.

The environment will be modelled using an

octree. This method allows for fast path planning and

obstacle avoidance. The octree indexes three-

dimensional space such that the occupied state of

each region can be determined.

The path planning module will apply the A*

search algorithm in combination with local Hill

Climb and Simulated Annealing. This combination

was found to be most efficient by (Leutert et al., 2012)

The planner operates by searching for a path from the

starting position to the goal position using partial

local search algorithms. If no path is found the

planner uses the complete A* algorithm with a

modified distance estimator. The A* algorithm also

avoids singularities in the robot’s architecture. The

obstacle will be bound by a bounding volume

represented by geometric objects. When a robot

coincides with the bounding volume, the robot is too

close to the object and a new path should be planned.

The calculations, instructions, commands,

detection, mapping, path selection will be calculated

and simulated on MATLAB using Peter Corke’s

Toolbox. The virtual implementation allows safe

testing before the algorithms are implemented in a

real-world scenario. Peter Corke’s Toolbox will be

used as a solver to perform all kinematics

calculations. The proposed architecture setup is

presented in Figure 2.

This project addresses the issue of human safety

when working within the workspace of a robot. The

algorithm combination makes for suitable

pathfinding and obstacle avoidance. It is fast and

accurate providing reliable results. This method will

ensure that safety of the human in the robot

workspace.

Figure 2: Proposed Architecture Setup.

6 CONCLUSION

Autonomous factories are advancing swiftly with

more robots being implemented without fences. With

humans working in close proximity to the robot, it is

important to ensure that there are no injuries or

fatalities. A method of robot awareness and reactivity

are essential to ensuring safety. Visual systems notify

the robot of a human presence in its workspace.

Control systems allow the robot to adjust its path and

to avoid the human and generate a new path towards

its goal. The proposed method of obstacle avoidance

and trajectory planning will be tested and validated.

This method combines octrees and A* algorithm with

Phase 1

• Develope sensory system for obstacle

detection and visual awareness

Phase 2

• Develope safe control methods for

Obstacle avoidence and trajectoty planning

in real-time.

Planning

• Design algorithms for data processing and

predicting the location of the human in the

workspace

Testing

• Implement the algorithms for a serial robot

architecture

A Review of Safety Methods for Human-robot Collaboration and a Proposed Novel Approach

247

local Hill Climb and Simulated Annealing for reliable

obstacle avoidance and trajectory planning.

ACKNOWLEDGEMENTS

Gratitude and sincere thanks to the MR2G research

group for providing the environment and

infrastructure to make this research possible.

REFERENCES

Albu-Schäeffer, A., Bicchi, A., Chatila, R., Luca De, A.,

Giralt, G., Hirzinger, G., Mattone, R., Oriolo, G.,

Schiavi, R., Siciliano, B., Tonietti, G., Vendittelli, M.

& Luigi, V. 2005. Physical Human-Robot Interaction

In Anthropic Domains: Safety And Dependability.

Barcellini, A., Bascetta, L., Raymo, M., Rocco, P.,

Zanchettin, A. M. & Robertsson, A. 2012. Integrating

An Anti-Collision System Based On Laser Time-Of-

Flight Sensor In An Industrial Robot Controller.

Billard, A. & Dillmann, R. 2006. Social Mechanisms Of

Robot Programming By Demonstration.

Boryga, M., Graboś, A., Kołodziej, P., Gołacki, K. &

Stropek, Z. 2015. Trajectory Planning With Obstacles

On The Example Of Tomato Harvest. Agriculture And

Agricultural Science Procedia, 7, 27-34.

D. Gao, C. W. W. 2009. Head Injury Criteria: Assessing

The Danger Of Robot Impact.

Eder, K. H., Chris;Leonards, Ute 2014. Towards The Safety

Of Human-In-The-Loop Robotics: Challenges And

Opportunities For Safety Assurance Of Robotic Co-

Workers'. The 23rd Ieee International Symposium On

Robot And Human Interactive Communication, 660-665.

Flacco, F., Kröger, T., Luca, A. D. & Khatib, O. A Depth

Space Approach To Human-Robot Collision Avoidance.

2012 Ieee International Conference On Robotics And

Automation, 14-18 May 2012 2012. 338-345.

Fratu, A., Vermeiren, L. & Dequidt, A. Using The

Redundant Inverse Kinematics System For Collision

Avoidance. 2010 3rd International Symposium On

Electrical And Electronics Engineering (Iseee), 16-18

Sept. 2010 2010. 88-93.

Fryman, J. & Matthias, B. Safety Of Industrial Robots:

From Conventional To Collaborative Applications.

Robotik 2012; 7th German Conference On Robotics,

21-22 May 2012 2012. 1-5.

Garmin. 2016. Lidar Lite V3 Operation Manual And

Technical Specifications. Europe Patent Application.

Jin, B., Wang, S. & Yili, F. Sensor-Based Motion Planning

For Robot Manipulators In Unknown Environments.

2005 Ieee/Rsj International Conference On Intelligent

Robots And Systems, 2-6 Aug. 2005 2005. 199-204.

Kuka. 2014. Lbr Iiwa [Online]. Available: https://www.

Kuka.Com/En-Cn/Products/Robotics-Systems/

Industrial- Robots/Lbr-Iiwa [Accessed].

Lasota, P., Fong, T. & A. Shah, J. 2017. A Survey Of

Methods For Safe Human-Robot Interaction.

Letla, T. 2008 Introduction To Robotics

Leutert, F., Freier, C. & Schilling, K. 3d-Sensor Based

Dynamic Path Planning And Obstacle Avoidance For

Industrial Manipulators. Robotik 2012; 7th German

Conference On Robotics, 21-22 May 2012 2012. 1-6.

Marvel, J. A. Sensors For Safe, Collaborative Robots In

Smart Manufacturing. 2017 Ieee Sensors, 29 Oct.-1

Nov. 2017 2017. 1-3.

Mišeikis, J., Glette, K., Elle, O. J. & Torresen, J. Multi 3d

Camera Mapping For Predictive And Reflexive Robot

Manipulator Trajectory Estimation. 2016 Ieee

Symposium Series On Computational Intelligence

(Ssci), 6-9 Dec. 2016 2016. 1-8.

Mohammed, A., Schmidt, B. & Wang, L. 2017. Active

Collision Avoidance For Human–Robot Collaboration

Driven By Vision Sensors. International Journal Of

Computer Integrated Manufacturing, 30, 970-980.

Nieto, J., Slawiñski, E., Mut, V. & Wagner, B. 2010. Online

Path Planning Based On Rapidly-Exploring Random

Trees.

Pérez, L., Rodríguez, Í., Rodríguez, N., Usamentiaga, R. &

García, D. F. 2016. Robot Guidance Using Machine

Vision Techniques In Industrial Environments: A

Comparative Review. Sensors (Basel, Switzerland), 16,

335.

Robotics, M. C. 2019. Dual Check Safety (Dcs) Explained

[Online]. Available: https://motioncontrolsrobotics.

com/Dual-Check-Safety-Dcs-Explained/ [Accessed].

Robotics, R. 2017. Our Cobot Difference [Online].

Available: https://www.Rethinkrobotics.Com/ Sawyer/

[Accessed 8/05 2019].

Robotiq 2014. Colaborative Robot Ebook.

Shiyong Wang, J. W., Di Li, And Chunhua Zhang 2016.

Implementing Smart Factory Of Industrie 4.0: An

Outlook. International Journal Of Distributed Sensor

Networks, 2016.

Sousa E Silva, J., Costa, P. & Lima, J. 2013. Manipulator

Path Planning For Pick-And-Place Operations With

Obstacles Avoidance: An A* Algorithm Approach.

Zacharias, F., Borst, C. & Hirzinger, G. 2007. Capturing

Robot Workspace Structure: Representing Robot

Capabilities.

Zanchettin, A. M., Ceriani, N. M., Rocco, P., Ding, H. &

Matthias, B. 2016. Safety In Human-Robot

Collaborative Manufacturing Environments: Metrics

And Control. Ieee Transactions On Automation Science

And Engineering, 13, 882-893.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

248