What will 5G bring to industrial robotics?

If you ever have the chance to enter a factory shop floor and see a robot in action, this is a must-do. The combination of strength, speed and extreme precision is nothing else than impressive. In the last years, they became general-purpose tools for automating production processes. Since their application possibilities are constantly being expanded they are used everywhere as long as the benefit of automating is higher than the efforts and costs of buying, installing and programming the robot.  

At present, there are basically four types of industrial robots: heavy single-armed robots (we can call them HEAVY), collaborative one-armed or two-armed robots (COOP), robots with limited capabilities for dedicated use cases (LIMITED), and the exotic ones (EXOTIC).

The yearly Hannover Industry Fair usually gives a good overview about the robotics market and this year HEAVY and COOP have been omnipresent. Robots of type COOP became quite prominent recently as they enable an easier integration into the production facilities, as they usually don’t require fences for safety reasons, and allow an alongside operation with their human colleagues.

Type LIMITED robots comprise fast pick-and-place models, that are sometimes called ‘spider’ or delta robots and Scara-models (Selective Compliance Assembly Robot Arm). Scara robots are designed like a human arm but with movement limited to the horizontal plane only. Both robot models are rather tight to a specific use case. 

Finally, type EXOTIC robots - there one finds different concepts like mobile industrial robots of various kinds and some very interesting bio-inspired robots.

The typical vendor-independent anatomy of type HEAVY and COOP robots is to have a physical robot comprising of joints and drives complemented by a robot controller where all control logic is deployed. Both elements are connected by wires for power supply and low-level control of all drives.

A human-machine interface, which might be a standard PC can be used to program the robot. Robots of type COOP also have a lot of sensors in-built to allow a safe interworking with humans and a design avoiding sharp edges.

The differences from vendor to vendor are in performance, intended areas of usage, programming possibilities and in the way the robot controller can be interfaced, and to what extent it is possible to influence the robot program by for example the integration of external sensors.

In robotics research, Cloud Robotics is a major topic. Basically, it is about connecting robots to the cloud, using communication technologies. The benefits, described for example in the Ericsson Technology Review article Cloud robotics: 5G paves the way for mass-market automation as well as several other publications, include:

1. Usage of more powerful computing resources in the cloud (e.g. for AI tasks)
2. Use data from the internet for decision making and learning (including digital twins)
3. Lower cost per robot as functionalities are offloaded to a central cloud
4. A possibility to perform a failover in case one robot physically breaks from an up-to-date backup in the cloud
5. Reliability of functions can be improved by running multiple instances as hot standby in the cloud and the operation can immediately be taken over from faulty primary function without interruption
6. Makes the operation and maintenance easier (software updates, configuration change, monitoring, etc.)

But what has 5G to do with cloud robotics? A static robot on the shop floor could as well be connected using industrial Ethernet to achieve most of the benefits listed above.

So why do we need 5G communication for industrial robots?

In a nutshell: if a robot shall be used flexibly and maybe be mobile, cables are a burden.

Cellular connectivity is the best choice because it is standardized technology with a very large, global ecosystem. 4G works for many use cases, but 5G will support also the absolutely most demanding ones, such as cloud robotics in cases where the cloud’s processing is relevant for the immediate motion of the robot (you can find some remarkable examples below).   Ultra-reliable and low-latency communications (URLLC) is a new service category that will be supported in 5G New Radio (NR), created to meet requirements for 5G applied in industrial domains. The 3GPP report TR 22.804 defines use cases and requirements for 5G applied in the vertical domain, like manufacturing. Therein, ‘Mobile Robots’ is specified as one use case

In the Ericsson Technology Review article mentioned above, our colleagues elaborate on the topic of flexibility as a success factor for industries, and the role that 5G can play in that context.

Our latest projects

Virtualized Robot Control (Comau Collaboration)

A notable example is our collaboration with robot manufacturer Comau and other partners to investigate the concept of virtual robot control using 4G/LTE (read more in the Ericsson Technology Review article 5G and Industrial Automation and references therein). Various parts of a robot’s motion control calculation can be outsourced to a cloud (or edge cloud) system to some extent. 

In our collaboration with Comau, they provided us a real robot cell with two large robot arms, a conveyor belt and some other industry devices. For the communication, Comau used ProfiNet; an Industrial Ethernet variant. First, we explored what level of robot control can be virtualized over 4G and 5G technologies. The high-level control that is typically done by the Programmable Logic Controller (PLC) turns out not to be very delay critical – it has a latency requirement of several tens of milliseconds depending on the configuration. However, the whole communication is very sensitive to delay variations (jitter) and packet losses. For instance, three consecutive packet losses or equivalent delay variations cause the whole robot cell to stop. Those requirements are straightforward to fulfill from dedicated hardware components though can be challenging using virtualized execution over wireless technologies.

Comau aimed to replace the hardware (HW) PLC with the software version and run that in a virtualized environment/cloud on commodity HW components. From the cloud platform perspective, one of the main challenges that virtualized control brings in is the execution of real-time applications. Some software PLC platforms use Windows as an operating system (OS), next to a real-time OS (RTOS) which is responsible for executing the PLC code. Both run in parallel and communicate via inter-process communication (IPC). The control logic implementation is always executed by the RTOS and Windows is often used as a user interface. The RTOS typically has some specific requirements to ensure the necessary performance such as precise timers and specific network interface cards. We addressed all these requirements and created a virtualized environment that can host the software PLC platform and execute the same control logic as the one was running on HW PLC.

We showed in a real factory environment that placing the PLC-level of control logic into an edge-cloud platform is feasible. However, if we investigate applications such as trajectory planning, inverse kinematics and control loops that accurately steer the speeds, accelerations or positions of the actuators it requires significantly lower latency, in the range of 1-5 ms. To support those applications, the ultra-reliable and low-latency service of 5G is essential together with a real-time edge-cloud platform deployed geographically close to the devices as illustrated in the figure below.

Robot sensor communication

This project was another proof point that low latency control of robots is possible over 5G.  The benefit over the legacy way of robot control is of course the possible virtualization of the robot control and the possibility to gather sensor data wirelessly in real-time for that. Last year, we used low-latency radio prototypes (‘WARP’) for ‘sensor - robot controller - virtual PLC’ communication of a Bosch APAS robot as part of the publicly funded German project Koordinierte Industriekommunikation (KoI). The Bosch Apas is a type COOP robot that offers an Ethernet interface that can be used with the Modbus-TCP protocol. To embed the sensor values into the robot’s trajectory planning in real-time it was necessary to transmit data from sensor to PLC to the robot controller every 5ms. This clearly would not have been possible without the use of low-latency wireless communication.

5G for Robotic Automation Cells

One typical scenario where low latency communication is required is sensors at the tool center point (so the robot’s ‘wrist’) towards the robot controller. This is usually used for applications that demand a higher precision like for example a torque sensor used to not break fragile workpieces. A 5G network offers the small latencies that allow a wireless connection. We investigated this scenario from a radio engineering perspective. To summarize the results briefly – using certain radio communication methods explained therein, it was possible to achieve a reliable low-latency (< 1ms) communication on radio links in an automation cell.  The measurements for our study were made in an active type HEAVY robot’s cell. 

Latency Impact in Robot Control (with a UR5 Robot)

We evaluated the closed-loop control performance of a UR5 industrial robot arm (also a type COOP robot) through a modelled 5G link. The focus was put on the effect of the link delay on the performance of the robot arm movement quality measured by specific key performance indicators (KPIs). The UR5 industrial robot arm has an externally accessible velocity control interface that accepts velocity commands for each joint (servo) and publishes joint state information with 8 ms update time. Investigated KPIs are response time and precision of trajectory execution, i.e. spatial and temporal deviations from the planed trajectory. The measurement results have shown that the network delay lower than 4 ms has no significant performance impact. This is because the internal operation of the robot ends in about 2 ms standard deviation in response time due to the internal sampling used in the robot, and the ticks of the robot and the controller are unsynchronized.

The required delay also depends upon the robot’s task:

• Reaction on external events: low network delay is desired, because the network delay between robot and controller directly increases the reaction time.
• Realtime trajectory refinement (i.e., accurate positioning of the end of the robot arm): Deadline on trajectory execution time leads to requirement on maximum tolerable network delay. In general, higher network delay makes the refinement time longer and, in this way, increases the total trajectory execution time.
• Trajectory Accuracy: Some tasks require accurate movement along the path such as welding, and not only at the final position. Another example is the collaboration of more robot arms where the precise and synchronized movements are crucial. For these tasks, low network delay is desired if external information shall be respected in the trajectory planning.

What’s next?

Our vision is that in the next years, most industrial robots and robotic supplies will be equipped with 4G and/or 5G communication modules as cellular communication technology is becoming an essential part of modern factories. But from our hands-on experience with industrial robots, to really exploit the flexibility wireless communication offers, there is also a need to rethink some robot control concepts like introducing virtualized control functions and to allow more flexibility in the interfacing of robot and robot controller from an external system.  Maybe some concepts of type EXOTIC will become more popular as well in the near future. Many cloud robotics concepts are based on AI and therefore, it will be a fundamental part of robotics quite fast. Robots, Cloud, 5G, and AI sounds like a buzz-words meetup but becomes real in this context. A long-term question will be how much autonomy will be introduced to industrial robotics and what impact this has on the way industrial robots communicate. Anyway, real-time-communication and real-time cloud computation is one of the six key challenges defined by our colleagues in their article: Artificial intelligence and machine learning in next-generation systems.