Collaborative robots are the latest technology, and may provide just the solution you are looking for to improve your production process, but you must still make sure any risks associated with your robotic application are recognized and managed appropriately.

Safety for Collaborative Robots: New ISO/TS 15066

Contributed by | Robotiq


With the new technology of collaborative robotics that allow robots to work alongside humans; workplace safety concerns are a lot different today. In fact since these robots have been designed to work alongside humans, the force and speed thresholds that are considered safe for cobots are very different than those of traditional industrial robots. To understand what this means, we must look at the new ISO technical specification that just came out. It contains a lot of data that can help with implementing a cobot and insuring that your workplace is safe.

The ISO/TS 15066 Robots and robotic devices – Collaborative Robots is the new technical specification developed by experts from the robotic industry. The new addition to the standards, since it is a technical specification, contains guidelines and recommendations for robotic end users and robotic manufacturers. However, this is not a legislative requirement. It is only a guide to help you to create a safer product or a safer working environment. It does provide useful data so that workplace injury can be minimized and/or avoided, so it is a valuable addition which will clarify and support robot use in the workplace.

The main improvements to the previous standard on robot safety (ISO 10218 – Robots and robotic devices – Safety requirements for industrial robots) include more precise definitions and complementary data on the 4 different types of human robot collaboration.


These different types of collaboration being:

Safety Monitored Stop: This first collaboration type is when a robot is used to do heavy duty applications (carry a car for example) and a dexterous application has to be done (fixing a small bolt) on the large part. The robot brings the large part to the human, stops (but remains active) while another action is completed, then continues its trajectory once the human tells it so. The reason why this application is considered collaborative is that the robot remains active (in operation) while the human is interacting around it. Historically, this type of application required a shutdown or brakes engaged on the robot before it was restarted after the operation was completed.

Hand Guiding: The hand guiding mode is the only condition where the robot is in direct contact with humans, in this case the robot must have a special force recording sensor (AKA Force Torque Sensor). This sensor is used by the human to direct the robot and displace it manually. Note that many hand guided robots are still fenced during their normal use and the human is simply present for the programming side of things.

Speed and Separation Monitoring: This type of collaboration allows for the removal of physical barriers between the human and the robot. This setup requires an area scanner that delimits safety zones. Imagine a green, yellow and red zone where you have respectively a high speed zone, a reduced speed zone and an almost stopped zone. This type of collaboration allows the robot to run as fast as possible when the human is not around and to reduce its speed as the human gets closer. Makes sense right?!

Power and Force Limited: This is the so-called collaborative robot or cobot. In fact, this type of robot ‘feels’ its surrounding and is capable of determining abnormal forces during its normal use. In other words, it is the only type of collaboration that allows the robot to run normally, albeit with power and force limitations, in the presence of humans.

The first three types of collaboration can basically be industrial robots with add-ons such as, area sensors or force-torque sensors. The 4th type is the most innovative part of the new proposition. In fact, power and force limited robots are the new technology that allow for the removal of fencing and even accommodates contact between the robot arm and the user. That being said, major attention should be given to the introduction of any robot on a work floor.


In order to make the introduction of a collaborative robot as safe as possible, the standard requires the performance of a risk assessment; not only for the robot, but for the whole robotic cell. In fact, for any collaboration you decide to use in your production process, a risk assessment is mandatory. You might have imagined that a risk assessment was only necessary when direct interaction at full speed was a possibility and that greater attention was given to this type of application, but no, every robotic application requires a risk assessment regardless of the intensity of the robotic interaction.


What is a risk assessment?

The risk assessment process basically analyzes all the motions, interactions and operations the robot is going to perform during its normal operation. The complete operation is sectioned into separated tasks and risk analyses are done for each task. These tasks and their associated risks are evaluated and rated. With certain criteria these risks are classified and those that are considered dangerous are evaluated and ways are sought to reduce their risk.

That is the brief description, however, this can be a complex process where any motion is considered a risk. In fact, the simple motion of closing a gripper is considered a risk. In order to know what type of risk is associated with the closing action of a gripper (for example) you need to consider a couple of aspects.

The severity of the risk, the possibility of avoiding the risk and the redundancy of the risk. Each of these criteria can be evaluated in the following manner:

Severity of Risk: The principal way to evaluate this criteria is by evaluating the force and/or pressure that is applied on the human body part by the robot. The tech spec provides a complete chart for human body parts and their associated pain value. There is also a difference between a quasi-static impact and a transient impact, with a quasi-static impact being an impact against a fixed object and a transient impact being an impact involving the free movement of the body part. In general, a factor of 2 is added to the maximum allowable force/pressure when you are free to move; which basically allows the robot contact to be 2 times stronger when you are free to move. The least severe or more acceptable risks are rated the lowest.

Possibility of Avoidance: This factor is normally inversely proportional to the speed of the robot. In an example where a robot is running at full speed VS a robot that is running at 50% speed, you clearly have more chance to avoid the slower robot than the faster one. Therefore, the faster a robot is moving the lower your chances of avoiding it are and thus the higher your risk. If the robot is moving slowly then you have more chance to avoid contact and have an increased possibility of avoidance, thus lowering your risk.

Redundancy of the Risk: This factor is evaluated according to the amount of time the risk is possible calculated over a fixed amount of time. For example, does this risk happen every 2 minutes or once per month. The less frequent the risk is, the lower the number you will get; but all risks need to be analyzed in relation to each other. I.e. do you have the possibility of small impacts frequently or a large injury infrequently?


*Note that data for specific body parts and the forces that they can withstand without injury are listed in the ISO/TS 15066.


As explained earlier, the score you get represents the danger level of the risk for a particular section of the robotic cell. The highest scores should be reduced to acceptable levels. You need to keep in mind that every time you reduce one risk you need to make sure that it does not increase another risk somewhere else in the process.

At the end of the day, there is no rigid guideline on what is dangerous and what isn’t. It is the end users responsibility to make sure the robotic cell that they set up is safe enough for human robot collaboration, regardless of whether the robotic cell is using collaborative or industrial robots. To have more information, download this guide on ‘How to do a risk assessment’. When considering risk, a quote that I like to refer to is the following:

‘’Risk assessment is not about avoiding hazards altogether, but about choosing one risk over another. For example, if you're looking to cross the street and a bus is coming and there's a 5% chance that you would be run over, then it's unlikely that you would take the risk. But if there is a child running after a ball crossing the street in front of the bus, you would not hesitate to run in front of the bus to save the child.’’ – Lasse Kieffer, Universal Robots.


It's the same with collaborative robotics: you may choose to accept the risk of a bruise, in order that you avoid the risk of more serious injury somewhere else in the production process.

To conclude, collaborative robots are the latest technology, and may provide just the solution you are looking for to improve your production process, but you must still make sure any risks associated with your robotic application are recognized and managed appropriately. You need to have the safest robotic cell you can think of and still be able to run the robot normally. The standards are simply there to guide you and give you information on what could pose a danger and how you can calculate these dangers in a technical way. But the practical side of things still remains up to you and your co-workers. 


Comments (0)

This post does not have any comments. Be the first to leave a comment below.

Post A Comment

You must be logged in before you can post a comment. Login now.

Featured Product

Maplesoft - Free Whitepaper: Developing a Robot Model using System-Level Design

Maplesoft - Free Whitepaper: Developing a Robot Model using System-Level Design

This paper uses NAO, the humanoid robot from Aldebaran Systems, to demonstrate how MapleSim can be used to develop a robot model, and how the model can be further analyzed using the symbolic computation engine within Maple.