This article presents a study on the evolution and future of low cost and flexible new generation of robots. These collaborative robots – designed to work safely with and around people – combine low cost and ease of use. They are finding their way into large plants and into small factories, which can now automate batch runs that would never be economical with a conventional industrial robot. Most of these robots are similar to conventional industrial robots, but are smaller, lighter, and simpler to use. As robotics technology has matured, programming the robots (and just about everything else) has grown easier and more intuitive. Key components, such as motion and impact sensors and vision systems, combine better performance with costs low enough to sell in cheap video game controllers. Today’s robots may not be very interactive, however, the first wave of collaborative robots are finding simple tasks where they can make a difference.


Rethink Robotic's Baxter (above) and Universal Robots’ UR3, rated for 3 kg loads, (right) are two next-generation “inherently safe” robots able to work near people.

Walking through a large automobile factory is like passing between two worlds. On one side, powerful robots rapidly position and weld sheet metal into framed car bodies. On the other, thousands of workers add doors and other moving parts, power systems, chassis subassemblies, and trim.

That division is there for a reason. Up to now, industrial robots have been big, fast, dangerous, and dumb. They have to be kept in cages and away from people. The danger has limited their uses among workers on assembly lines.

An emerging generation of robots is promising to change all that. They can be let out of the cage. They are nimbler, easier to program, and far less dangerous.

These collaborative robots—designed to work safely with and around people—combine low cost and ease of use. They are finding their way into large plants and also into small factories, which can now automate batch runs that would never be economical with a conventional industrial robot.

Like the first PCs, this new generation of robots is a general solution. Engineers do not have to redesign their entire assembly line to use them. As a result, they can apply them in many different ways.

For instance, Glidewell Laboratories of Newport Beach, Calif., used a robot to reconfigure its manufacturing process.

Glidewell is a large producer of dental crowns from impressions of patient's teeth. Every crown is a custom order, and the workflow was almost artisanal in its many manual steps. Over the past few years, however, Glidewell has added engineering software and automation to improve productivity. Robots were the next logical move, and Glidewell had an obvious application.


“We’re making 30,000 zirconia crowns per week, and it's really hard to keep up when you’re making a custom product,” said David Leeson, Glidewell's engineering manager.

It takes several steps to make a crown that matches a patient's tooth shape and color. When the company receives an impression from a dentist, it casts a replica of the patient's mouth and scans it into a 3-D CAD file. Glidewell designs the crown in CAD and sends the output to a CNC mill for machining.

Before installing the robot, Glidewell put each replica into a case box and sent it to the mill. It took the mill about 10 minutes to shape each crown. Rather than have an operator standing around waiting, it made sense to machine 15 crowns at a time from a single disc of compressed zirconia powders.



Once the boxes arrived at the mill, the operator would group the molds by size and color to optimize material utilization. He also had to record the location of each crown on the disc, and this manual process sometimes led to errors.

These manual processes created a logjam at the mill. Leeson broke it by using a collaborative Universal Robotics UR5 robot to tend four CNC mills. The robot made it economical to switch from processing a batch of 15 crowns to milling a single crown at a time. Now, CAD models go directly to the mill. Instead of waiting until the operator has enough molds to sort and group, the robot matches the colored blank and begins machining.

“People around it feel safer because the robot slows down when they come near.”

— Jean-Francois Rousseau, plant engineer at Etalex

“This gives us single part flow,” Leeson said. “It used to take three hours for the cases to go from design into the mill. Now the CAD data transfer is instantaneous.”

The system also improves quality. Instead of having the operator manually record the position of each crown on the disc, the robot, working with Glidewell's manufacturing information system, matches the crown preform with the CAD model and records its location when placed into a barcoded tray for sintering into a final crown.

“We get way fewer returns now,” Leeson said.

So far, this could be a story about any industrial robot. In fact, Glidewell also uses a conventional robot in another part of the plant.

“The big difference between that robot and the Universal is that the conventional robot is caged and requires a constant level of engineering support,” Leeson said.

Limited Force

Eliminating the cage simplifies layout and workflow. Conventional robots are caged because they are dangerous. They are fast and powerful, and most have no way to detect collisions with anything short of a large hunk of metal.

The UR5 is a power- and force-limiting robot, one that operates at low speeds and light payloads. With dual sensors in every joint, the robot stops within milliseconds of the slightest impact. At worst, it delivers a modest shove. Workers can safely deliver new zirconia blanks and remove milled parts without stopping the robot and losing time.

The UR5 is also easy to use, Leeson said. It includes features, such as Ethernet connectivity, that cost extra on conventional robots. He said the user interface is easy for anyone with programming or engineering skills to learn, and it comes with a simulator to test scripts.

Above all, the UR5 is flexible. Like a PC, it is a general-purpose device that does a variety of tasks. Leeson, for example, used it as an alternative to dedicated automation. He estimated that it would cost about $150,000 to automate the Glidewell's zirconia crown work cell. The UR5, including modifications, programming, and setup time, cost little more than one-third of that. The savings alone would fund roughly two more robotic work cells.

As robotics technology has matured, programming robots (and just about everything else) has grown easier and more intuitive. Key components, such as motion and impact sensors and vision systems, combine better performance with costs low enough to sell in cheap video game controllers.

The result is a change in the robot paradigm. The installed cost of a typical industrial robot is two to three times the robot price. New-generation robots generally cost far less, $20,000 to $40,000, and as Leeson noted, cost far less to install. Universal and another manufacturer of safer industrial robots, Rethink Robotics, founded by Rodney Brooks, the inventor of the Roomba, claim most customers earn back the investment in their robots in six to nine months.

The UR5 was Universal Robots’ first model, introduced in 2009. The company was the first manufacturer to claim its industrial robots were inherently safe because they limited their power, force, and speed. Other power- and force-limiting robots today include Rethink Robotics’ Baxter, ABB's Yumi, KUKA's LBR iiwa, and Precise Automation's line of articulated, SCARA, and Cartesian robots.

Most of these robots are similar to conventional industrial robots, but are smaller, lighter, and simpler to use. They carry small payloads, usually 3 to 10 kg, at relatively low speeds. They also contain sensors sensitive enough to detect a low-speed collision and fast pathways that stop motion within milliseconds. This reduces the likelihood of injury if they hit a worker.

They are also easy to use. Programming, for example, involves moving the robot arm between points to teach it a job, then using code to refine the task and tell the grippers or other end effectors what to do.

The robots are easy to program for simple tasks, such as pick and place, machine tending, and product testing. Precise Automation specializes in moving laboratory samples.

Uncaged in Tight Spaces

Because they can work uncaged around people, power- and force-limiting robots can function where space is limited and must be shared. That ability solved a dilemma faced by Jean-Francois Rousseau, a plant engineer at Etalex, a Montreal-based manufacturer of steel storage systems.

Rousseau had a huge metal punch located at the front of his plant. Removing punched metal is dangerous work, and he wanted to automate the process. The 12-foot-long punch, however, was only six feet from the main aisle.

“We didn’t have any space to put in a safety guard, and forklifts and employees pass down that aisle all day long. And we didn’t want to move the press, because it is such a big setup,” Rousseau said.

He turned to Universal because it was similar to the 30 Fanuc industrial robots he already managed, and he liked its smooth motion and precision. Rousseau handed the programming over to a junior engineer. Within a month, the robot was removing and stacking long bent strips of metal as they come out of the punch.

The robot works largely on its own. Workers periodically enter the space to measure a part and remove carts of stacked parts. A laser scanner senses when workers approach and slows the robot down by 30 percent.

“People around it feel safer because the robot slows down when they come near,” Rousseau said.

Eliminating fences also makes it easier to change tooling. With conventional robots, Rousseau would have to remove the fence and robot to get a forklift carrying the massive tooling through. Now, he unlatches a bracket and slides the robot out of the way to access the center of the press.

In most factories, the only difference between power- and force-limiting robots and conventional industrial robots is the absence of fences. With Baxter (and its new Sawyer robot for machine tending), Rethink Robotics hopes to bring greater flexibility to the shop floor.

“We designed it to roll the robot off, train it easily, and change tasks from month to month,” said Rethink product manager Brian Benoit.

Baxter comes with several innovative features. One is the robot positioning system. By using a camera embedded in its wrist, it can orient itself by locating markers posted around the workspace.

“Now you can design tasks in two different cells, turn on the positioning system, and the robot will orient itself and get right to work,” Benoit said.

Baxter also has what Benoit calls “adjustable behavior.”

“Baxter knows where the part is in the work cell, and knows if it is holding a piece or misses a pick,” Benoit said. “It won’t just continue with the process. It will pick up the part or repack it again. It has sensors on board that make it possible to put these behaviors in place.”

Unlike most robots, whose joints use gear boxes with hard stops, Baxter has elastic actuators. They make it slightly compliant. If its arm bumps into someone, it gives a little bit.

That gives Baxter an extra margin of safety. Compliance, when coupled with built-in force sensitivity, also lets Baxter perform more tasks. It could, for example, test a circuit board. This usually involves inserting the tester guide pin through a hole on the board.

“A human would use the guide pin to feel his way in and give it a shake to mount it,” Benoit said. “A typical robot would have to go top down. Baxter comes in at an angle, like a person, and when it feels the force feedback, it has another move that seats it.”

At Du-Co Ceramics in Saxonburg, Pa., process engineer Josh Rupp is taking advantage of Baxter's flexibility to add automation to short production runs.

Du-Co created a small work cell with a simple conveying system. Ceramic preforms come down the line, one at a time and oriented in the same direction. When they arrive at a known location, the system controller tells Baxter to pick them up. The robot then places them in trays for sintering. Workers periodically enter to deliver empty trays and remove full ones.

“We can look at production runs we could never automate before, because the time required to set up the work cell wouldn’t have justified it. In the past, we would have done it with people. Now, we can create a versatile cell that we can move from machine to machine,” Rupp said.

As with other power- and force-limiting robots, programming was simple. In fact, Rupp hopes he can eventually train workers to set up Baxter.

“That's the best way to utilize its flexibility,” Rupp said.

Emerging Safety Standard

While power- and force-limiting robots have many boosters, their payloads are limited. At most, they might lift 14 kilograms. That's a problem for companies that work with large or heavy parts.

General Motors’ principal robot engineer, Marty Linn, sums it up succinctly: “Inflatable hammers are inherently safe, but not very good at pounding in nails.”

That is why engineers are developing ways to instrument conventional industrial robots to work safely near humans. A new standard under development, ISO 15066, Safety Requirements for Industrial Robots—Collaborative Operation, outlines four distinct safety strategies.

One, of course, is limiting power and force. The second is the safety-rated monitored stop, where the robot stops as soon as a worker triggers a sensor (such as a light curtain or pressure mat) by walking into its space. It resumes work once the human leaves.

Third is hand guiding, which allows humans to teach an arm a new set of motions while it is in live automatic mode. Fourth is speed and separation monitoring, which uses sensors to track workers and to slow or stop the robot if humans get too close.

“We designed it to roll the robot off, train it easily, and change tasks from month to month.”

— Brian Benoit, Rethink product manager

Many industrial robots are already equipped with these safety-rated sensors. “The major industrial robot makers have been doing functional safety for more than four years,” said Erik Nieves, technology director for Yaskawa Motoman.

Those robots could be used in collaborative applications once ISO 15066 is adapted, he added.

In fact, some companies are already using industrial robots collaboratively. Nieves described one customer that uses a worker-robot team to move 400-pound castings.

“The robot can’t pick up the parts blind, because they are unwieldy and come on a pallet in a variety of positions,” Nieves said. “The robot has an autonomous program and moves where the part ought to be, then waits for the operator to move the robot into position to pick it up.

“After the operator walks out of cell, the robot resumes the autonomous part of the work.”


GM's Linn would like to put collaborative robots on his assembly lines, especially in tasks that are dull, dirty, or dangerous for humans. He is intrigued with BMW's demonstration of a robot that can glue a water deflector to the side of a door using constant pressure, a job that he calls “ergonomically challenging” for workers.

Yet, after GM's experience with recalls last year, the company has doubled down on safety throughout the organization. GM will not consider collaborative robots without third-party safety certification to internationally recognized standards and specifications.

The lack of existing safety standards leads Linn to question the safety of “inherently safe” robots. It depends on the jobs and the individuals, he argues. Bumping into someone with a 5 kilogram load might not hurt, but a lighter metal sheet with a sharp edge could slash through skin.

Researchers at Germany's University of Mainz are trying to answer those questions through a research program that evaluates pain and injury thresholds. The work is complex, and has slowed the adoption of ISO's collaborative robot standard. Linn is not happy about it.

“Without clarity there, it handcuffs us quite a bit,” he said. “Our company believes in standards. If you go to any trade show, robot manufacturers are all working on something, but to be very candid, they have not gotten to the level of certification we need.”

Nor are collaborative robots very collaborative. In fact, most simply coexist with humans. They modify their behavior when humans enter or leave their space. Perhaps one day soon they can do kitting, grouping parts from bins so a worker can install them. Still, that is about as far as the present generation is likely to go.

Researchers, such as Aaron Bobick, who founded Georgia Tech's School of Interactive Computing, and Julie Shah, who heads MIT's Interactive Computer Group, hope to change that.

Bobick, for example, has trained robots to know a task's sequence, so they can then take the right actions at the right time. Shah, meanwhile, has focused on teaching robots to anticipate and respond to human actions.

“Robots use repetitive motion, but people don’t,” Shah said. “People go a little faster or slower. That means the robot is starting and stopping a lot, and we lose the economic benefit of automation if it does that.”

Those advances are still years away. Today's robots may not be very interactive, but neither were the first PCs. It took years for today's intuitive interfaces, Internet, and smartphones to evolve. Yet even in their earliest forms, PCs made it easier to perform some tasks, like creating spreadsheets or writing and revising documents. Similarly, the first wave of collaborative robots are finding simple tasks where they can make a difference.

On the shop floor, the revolution is only beginning.