The Fallacy of Fencelessness, Why Cage Free is Best Part 1
By Alberto Moel (Vice President Strategy and Partnerships) and Clara Vu (CTO / co-founder), Veo Robotics
In recent months, flexibility in manufacturing has been top of mind. To keep pace with market needs, manufacturing must become more flexible to continuously adapt to change, especially during a crisis like the current one. Although the press and industry have picked up the topic recently, the need for flexibility in factories has been an ongoing concern and issue for manufacturing engineers for quite some time.
A powerful way to introduce flexibility in factories is by simultaneously taking full advantage of robots, which are the most flexible machines in a factory, and humans, which are the most flexible resource. But humans must be kept separate from robots with fences or other guarding equipment to ensure worker safety. This separation introduces frictions and inefficiencies that limit how manufacturing processes can respond to rapidly changing product and market conditions. Allowing humans and robots to easily and safely work together in the same space will enable a much more flexible manufacturing environment.
To that end, the manufacturing industry has invested heavily in the development of safe collaborative applications for humans and robots. Standards such as ISO 10218-1 and ISO/TS 15066 outline four different methods of safe human-robot collaboration: safety rated monitored stop, hand guiding, Power and Force Limiting (PFL), and Speed and Separation Monitoring (SSM).
Two of the reasons people are interested in applications with safe human-robot collaboration are to remove fences and reduce floor space, but in many cases these two goals are actually at odds with each other. In today’s post we're going to address this contradiction and show that in many applications limited fencing or guarding is desirable (or even required) to improve workcell economics, ensure workcell safety, and result in more fluid human-robot interaction.
PFL and SSM: The leading candidates for useful human-robot collaboration
The most widely available and commonly understood (and misunderstood) form of collaborative robotics in manufacturing applications are PFL robots, often known as “collaborative robots” or “cobots.”
A PFL robot controls hazards by limiting the power and force the robot can exert before stopping. PFL has had a major impact on how we think about collaborative manufacturing, but the technique does have limitations. A stop is triggered only when the robot hardware detects a collision, so this approach only works for smaller, slower, lightweight robots that won’t harm a person by coming in contact. But even a small, lightweight robot is considered hazardous when it’s carrying a sharp object, so PFL robots are also limited in end-effector designs and types of payloads.
SSM is another form of collaboration that has great promise and addresses some of the limitations of PFL: SSM works with standard industrial robots and has fewer constraints on end-effectors, speed, and payloads. With SSM, a moving robot is assumed to be hazardous and a stationary robot is assumed to be safe. SSM requires a Protective Separation Distance (PSD) between the robot and human such that it is always possible to bring the robot to a stop before coming in contact with a human. As long as the robot stops safely before human contact, the size and payload of the robot are not relevant. Hence, the SSM mode has much broader applicability for collaborative robotics.1
Collaborative robotics (including both PFL and SSM) is generating quite a bit of excitement as it extends robots’ capabilities and increases manufacturing flexibility at a time when that flexibility is very much needed. However, much of the marketing and hype around collaborative robotics (in particular, PFL robots) draw on false “conventional wisdom,” i.e., because PFL and SSM robot systems are “collaborative” and “safe,” it is OK to drop our guard and go “fenceless.”
In today’s post we look in depth at some simple SSM collaborative applications and debunk this notion that “collaborative” = “fenceless.”2 And in fact, a workcell without a fence can require even more space than it would with a fence. We argue that fences and safety guarding can improve both the possibilities of safe human-robot collaboration and the economics of manufacturing.
What does “fenceless” actually look like?
In SSM, the key parameter that determines whether the robot enters a safe state is the Protective Separation Distance (PSD). Basic physics, as well as trajectory planning implementation latencies on robot controllers, mean that a large industrial robot carrying a heavy workpiece cannot stop instantaneously. When the distance between a moving robot and a human in the workcell is less than this PSD, the robot must begin to slow down to a stop so that, if the human were to approach the robot 3 , the robot would be safely stopped by the time they meet.
Obviously, this PSD is going to be a function of how quickly the robot can stop and what distance it could travel in the direction of the human before it comes to a stop. And this stopping time and distance is going to depend both on how fast the robot is moving and how big and heavy it is (together with its end-effector plus workpiece payload). The bigger and heavier the robot, and the faster it is moving, the longer the PSD. 4
Let’s imagine a simple 5 collaborative application with a large robot (say, one of the industry workhorses, a FANUC R2000iC/165F) using SSM as the safety mode (Figure 1). In this operator load station application, the human behind the worktable brings parts to the robot, which then loads them into a machine, such as a CNC (imaginary in this case), for further processing.
In Figure 1 we show the application without any guarding or fences. The robot can only move about within a certain volume of space, determined by its physical envelope or constrained using (optional) safe axis and rate limitation functionality, such as FANUC DCS 6, but the workcell extends well beyond that.
Why is that? Because safe robot control systems, including Veo Robotics’ FreeMove, can only control robot movement, not human movement. Without fences in place, a human could enter the moving robot’s path. To account for this unpredictable human movement, fenceless SSM systems require a buffer zone (beyond the robot’s range of motion) where human motion is detected and the robot is triggered to stop or slow to maintain the PSD.
For illustrative purposes, let’s assume we’re using FANUC’s DCS to safely constrain the robot within a 2.8m x 3.5m “keep-in” zone (meaning the robot is “virtually caged” within this volume). Figure 2 shows a top view of the same workcell, but this time indicating the keep-in zone (in yellow) and the required buffer area (in green). If a human (walking at 1.6m/sec) were to approach the moving robot (in this case at 66% of full TCP speed and 100% payload), the robot would need to stop by the time the human reaches the edge of the keep-in zone.
Using the stopping time and distance numbers provided by the manufacturer, for this particular example, the PSD is 1.52m and the required “fenceless” workcell area is 37.7m2. For a robot running at full TCP speed with a full payload, the PSD extends to 1.83m, and the workcell area to 46.3m2.
Because the workcell is not fenced, anything7 entering the safeguarded space could trigger the system to slow down or stop the robot, even when it’s not warranted, expected, or desired. This would be the case if, for example, someone were to step into the buffer zone just hoping to get a better look at the application.
Fences constrain human motion, not robot motion
How can fences be useful? Consider Figure 3, where a fenceless layout allows a human to arbitrarily enter the robot envelope while the robot is moving. In order to make this situation safe without a fence, we must add a relatively large PSD buffer zone so that the robot can stop fully before a human reaches it. If the interaction is planned, the robot should already be slowing down as the human approaches, and if it is accidental, the interruption will negatively impact workcell cycle time and productivity.
Fencing off some of the workcell’s entry points where there is no need for people to approach means a human can work right next to the robot, (for example, in an adjacent workcell) without disturbing robot operation from unwanted intrusions (Figure 4).
What if we carefully add some fences to our application?
Now let’s consider the same application, but with fences on two sides of the robot envelope (Figure 5). Note that these must be actual physical barriers—human hands and feet can get through stanchions or dividers, and light curtains and 2D LIDAR scanners don’t do anything to prevent human intrusion.
Let us assume that the fences come right up to the 2.8m wide keep-in zone and are 6.5m long, corresponding to a buffer zone for the robot moving at 66% of TCP speed along its main axis (Figure 6).
By constraining the workcell’s entry points from 360 degrees to a couple of specific locations, we can reduce the size of the workcell (for a TCP speed of 66% of maximum) to 18.2m2, which is less than half that of the unconstrained “fenceless” area. And although running the robot faster does increase the required buffer area, the rate of increase is smaller than for the fenceless implementation.
What we really want is “cage-free,” not “fenceless”
Today, industrial robots are either fenced on all sides—i.e., caged—or rule over large areas bounded by light curtains and area scanners—i.e., slightly higher-tech cages. When a worker does open a cage or breach a light curtain, it is because something has gone wrong and the workcell requires maintenance or some other kind of unplanned fault has occurred. Stopping the line creates unacceptable downtime, so making tiny changes like adding a zip-tie to a fixture can be incredibly costly. Additional design and construction to reduce the need for preventive maintenance and automate periodic tasks also add to workcell costs.
But as we saw, fencelessness is not the solution either. What we really want are “cage-free” robotic applications, where humans easily interact with or work near robots as part of normal operations. This will require novel human-robot collaboration technologies such as Veo FreeMove (in some instances in combination with traditional safety guarding).
In addition to quantifiable value creation and cost savings from increased flexibility, Veo FreeMove introduces a bottoms-up, fresh mindset to designing manufacturing processes. Safe human-robot collaboration is built-in during the design phase, not just layered on top of the workcell at the cost of productivity and efficiency.
In the next blog post we’ll explore some of the finer points on implementing such “cage-free” collaborative workcells and how guarding can improve human-robot collaboration, in particular using advanced safety systems such as Veo FreeMove.
1And the market demand shows as much. We do not have a detailed breakdown of robot shipments by payload, but we estimate that more than 80% of robots shipped worldwide are for payloads of 10kg or larger. The latest and most reliable numbers available for collaborative robots (which are overwhelmingly built for payloads under 10kg) indicate that they make up single digit percentages of robot shipments.
2Some may assert that because PFL robots safely stop on human contact, the need for additional external guarding is avoided (and hence “better” than SSM). While it is possible to develop a truly “fenceless” PFL robot application, it would mean limiting the robot to a narrow range of speeds, payloads, and safe end-effectors. In all other cases, even PFL robots will need guarding or fencing. Take it from Roberta Nelson Shea, a leading safety expert and Global Technical Compliance Officer at Universal Robots, the top maker of PFL robots.
3The speed at which a human is assumed to approach the robot is set by the standards at 1.6m/sec for the body and 2m/sec for the extremities.
4The specific calculation of the PSD is fully described in the ISO/TS 15066 standard, and it includes terms for the robot stopping distance as a function of speed and payload (usually supplied by the robot manufacturer); assumptions about human speed; and the response time of the sensing system. The largest contributions to the PSD come from the robot and robot controller, not the sensing system. In practice, the PSD is calculated using worst case assumptions for speed and payload of the robot, with a lot of “slack” in allowances for robot controller and sensing latencies. Ideally, we’d like the PSD to be as short as possible and incorporate the true dynamics and paths of the robot and humans, but it is currently difficult to do so while maintaining human safety with existing technologies. Veo Robotics is working with robot manufacturers and industry partners to break through these limitations.
5 Well, more of a toy model, but it is good enough for our purposes.
6 Or the equivalent package available for all other major robot brands, such ABB SafeMove 2, Yaskawa FSU, or KUKA.SafeOperation.
7Depending on the sensing technology, something as small as a rodent or bird could be detected as “human”. The Veo FreeMove system has sufficient “intelligence” to eliminate a number of these false positives, but it would still assume a wayward forklift was a human and slow down or stop the robot in response—safely, of course.