An industrial robot can cost hundreds of thousands of dollars just for the hardware. It can cost three to five times more to add it to an assembly line and get it working. That’s why industrial robots are typically only used by large, wellcapitalized manufacturers. That’s also why Rodney Brooks, one-time robotics professor at Massachusetts Institute of Technology and Director of its Computer Science and Artificial Intelligence Laboratory, founded Rethink Robots in 2008. He wanted to develop a relatively low-cost factory-floor robot that could be trained or programmed by a factory-floor worker.
That $22,000 robot, Baxter, is now rolling off the U.S. assembly lines and is ready to go to work.
Baxter is equipped with two arms on a turnable torso. Each arm measures 41 in. from shoulder to wrist and has seven degrees of freedom (dof). The arms move at up to 3.3 fps empty and 2 fps when loaded. They come with two interchangeable hands or end effectors that can pick up 5-lb payloads.
Electrical parallel grippers with interchangeable fingers of different lengths, as well as interchangeable fingertips, can be customized to pick up and handle specific objects. The grippers add 1 dof and open or shut within 1sec. They include a force sensor, which detects when an object is in the robot’s grasp and how much force it is exerting on that object.
Vacuum-cup grippers work on smooth, hard-to-grab objects such as mirrors, plastic sheets, and larger packages. The cups come in a variety of sizes and an air-pressure sensor detects when they have “grabbed” an object. The cups reduce the risk of marring an object’s finish, but they do require that users hook the robot to an airpressure supply.
The arms can be “trained” or programmed to work together, or they can each carry out tasks independently of what the other arm is doing.
The torso can be bolted to a table or other stable surface, but can also ride on a 141-lb pedestal from the company. The pedestal has two adjustable heights and is sized to fit through standard doorways. It lets the 165-lb robot be rolled from workstation to workstation on industrial-grade casters. The casters come with legs that lock it into place and adjust to level and stabilize the robot.
For safety and to carry out tasks, Baxter has 360° sonar and machine vision with one main “face” camera and access to four others. The wrists, for instance, carry a camera so Baxter can closely monitor what its end effectors are doing. Each wrist also carries a range finder.
Baxter operates in temperatures from 32 to 104°F and is rated IP50, which means it is protected against dust but not water and other liquids. It runs on 120 Vac and 10 A, and is rated for 6,500 hr.
What makes Baxter different
Baxter is different from other industrial robots in that it uses behavior- based intelligence, a concept Brooks developed throughout the 1980s and 1990s. He used the concept for all the robots developed and built at his previous company, iRobot (makers of the Roomba). “You program in a collection of parallel behaviors, each running independently, that look at the environment and input conditions, then decide if they have an action to perform.”
Conventional robots come with little or no programming, so users have to define and program all error conditions. Baxter comes preprogrammed with certain “instincts” right out of the box that always run in the background when Baxter is working. This lets users program tasks rather than a string of repetitive motions.
For example, if you show Baxter an object and position its gripper so that it can pick the object up, the robot builds a visual model of the object and stores this data. Later, Baxter will be able to visually identify the object, even if it’s in a different orientation and amidst other objects and clutter, and pick it up.
Another one of these basic instincts is self-preservation, “For example, the arms never collide with one another, thanks to the preprogramming,” says Brooks. “Even if you grab both arms and try to bring them together, you will feel a force, one that gets stronger the closer the arms get to one another, resisting you. This behavior keeps the robot safe from itself.”
“Another built-in behavior is that Baxter knows it can’t put something down if it doesn’t have something in its hand,” says Brooks. “So if the robot picks up an object to place in a box, and a worker grabs the object from its gripper, the robot doesn’t try to put that object in the box. Instead, the robot goes to the designated area to pick up another object.”
“To get the robot to pick up an object, a worker simply takes its gripper and places it over the object,” says Brooks. “Pushing a button on Baxter’s sleeve closes the gripper on the object. The worker then lifts the arm. The robot now knows it is supposed to pick up one of these objects in this area.”
“And if you’re training the robot to pick up an object and put it in a box, you can first train it to put the object in the box, then where to pick it up,” explains Brooks. “You don’t have to do it in order. Baxter figures out the proper sequence.”
Baxter is also adaptable. If you train the robot to pick up an object from a moving conveyor belt and place it in a box, you don’t have to tell the robot how fast the belt is moving or exactly where the belt or it, the robot, is stationed. You can even change the belt speed. Baxter can see the belt and object and adjust to pick up the object.
“A line worker can quickly and simply program Baxter for new tasks,” explains Brooks. “ You don’t need an engineer, and you don’t need to set up the environment or have accurate metric data for the workstation. You just show the robot what to do and it adapts through vision and its other senses. And if you move the robot a bit, it keeps on working job because it uses task coordinates, not robot coordinates.”
Baxter is said to be safe for humans to be working around it. Its sonar and moving-target indicator alerts it to where people are and the robot avoids those areas. If one of the robotic arms happens to strike a human, a wall, or practically anything of size, sensors in its joints immediately signal Baxter to go into zero-force mode. This eliminates any forces on the arms. The robot also compensates for gravity and the arms remain motionless. (They can be gently pushed or nudged into any position and will maintain that position.) This protects nearby workers and the robot itself. Another safety feature slowly lowers the arms if power is removed or lost.
And there’s no reason for workers to override or work around safety measures. “There’s really nothing for workers to bypass,” says Brooks. “If they throw a bag over the ultrasound sensor, for example, the robot just shuts down. We’ve tried to make it as safe and foolproof as possible.”
“I have worked in industries where people were paid piecework, and they did try to get around safety devices to work faster,” notes Brooks. “But with Baxter, bypassing safety features doesn’t speed production or improve performance.”
Baxter is designed for expansion and growth. Its wrists, for instance, have end-effector plates that accept third-party grippers. There’s also Ethernet and USB connections which will let users plug other devices and capabilities into the robot. And Rethink Robotics plans on releasing new software.
“Next year we will release software that lets researchers use the hardware platform to program new capabilities using our lowlevel software,” says Brooks. “And later we will have an app-level kit that works with high-level software for new manufacturing features.”
“We also plan on bringing out software next year that lets Baxter understand how to pack boxes,” says Brooks. “If you show the robot a box and train it to pack items in a 3 × 4 array, it will do so even if you turn the box 30°.”
The company is targeting smaller companies as its first customers, and they’ve built a scalable distribution network around that plan. “We have gotten interest from larger companies with national brands,” notes Brooks. “But those companies don’t buy one or two robots, they buy dozens, even hundreds. So it takes time to make a sale as they want to try out the robots, evaluate them, and see how the robots might work in their plants.”
But there can be problems with small companies as well. It can be a challenge convincing owners and managers that they won’t need expensive consultants or nonstop maintenance to keep Baxter up and running.
“At one small plastics factory, the manager watched as one of our engineers went up to one of our robots that was packing boxes and moved it by about half a foot,” recalls Brooks. “The manager’s face fell and he said, ‘Well now you will have to teach that task to the robot again.’ He was quite surprised when the robot kept on packing boxes. Seeing that the robot adapts to its environment and doesn’t need specialists really gets the message across.”
One of reasons Brooks is so adamant about his new robot is that he believes it can bring manufacturing jobs back to the U.S. This may seem counterintuitive as many people take it for granted that robots replace human workers.
“But our robot is not made to replace people, just like the PC didn’t replace office workers,” says Brooks. “This robot can change factory workers’ jobs, letting them offload simple, repetitive, dull tasks, and let them — cognitive beings with dextrous hands — do the higher-value-added work and thereby be more productive. So it’s a tool ordinary workers can use, not a way for management to displace workers.”
What Baxter can do
Rodney Brooks tells of how his team regularly takes Baxter out of its packing crate at a customer’s facility and has it set up and working on a line they’ve never seen before within an hour. But what kinds of tasks does Brooks believe Baxter will be widely used for, at least at first?
• Material handling: Moving objects from one location to another, counting, and reorienting.
• Loading and unloading lines: Putting parts on or removing them from conveyor belts or fixed surfaces.
• Inspecting, testing, and sorting: Checking parts for weight or shape, evaluate them against a criteria, and perform different actions depending on results.
• Operating machines: Watching machinery and performing sequences of actions based on input or alarms. For example, it could insert a part into a machine, push a button, then remove the part.
• Packing and unpacking: Picking up an object, bag, box, or tray and arranging them in an array for packing. It can also unpack containers.
• Light assembly: It can be quickly trained to align and snap fit parts together and insert them into containers.
• Finishing: Grinding, polishing, and other finishing operations.