Will Baxter be the Model T of the robotics industry?

Rodney Brooks has been working in the field of robotics since the 1970s, and is considered one of the world’s foremost roboticists. In 1990 Brooks cofounded irobot, which has produced a number of innovative robots including the roomba vacuum cleaner. In 2008 Brooks co-founded Rethink Robotics with the aim of creating inexpensive, versatile, and easy to use manufacturing robots that could work directly with humans.
Rethink Robotics recently unveiled the highly configurable Baxter robot, which should be the first in a series of increasingly sophisticated and autonomous robots. In an interview with Sander Olson for Next Big Future, Brooks discusses the capabilities of Baxter, why he believes that Baxter is a game changing robot, and what industries will be transformed by robotics in the coming decade.
Rodney Brooks
Question: Your company recently changed its name from Heartland Robotics to Rethink Robotics. This implies that a new generation of robots is emerging.
The first manufacturing robots emerged in 1961, when sensing and computation were expensive. Since then, robots have been used in welding and painting, but are generally not deployed in the assembly line. This is because robots were too dangerous, and difficult to interact with. Just as the PC transformed the way people interacted with computers, these new generation robots are inexpensive, and are designed to safely interact with humans, to be trainable, and to be an easy to use tool.
Question: What is the optimal ratio of humans to robots?
Our robots are designed to do dull, repetitive tasks, and to be very easy to use. These robots will have screens on their faces, and will contain numerous cameras. So in many situations a single human could oversee a large number of robots.
Question: Will these robots respond to verbal commands? Can it see?
Most factories are far too noisy for voice commands to be effective. There are buttons and knobs on it, and menus on the face/screen. The robot sees using five cameras, range sensors, motion sensors, and every joint has force sensors. Those force sensors are one of the things that make it so safe – if it encounters an unexpected force it stops moving.
Question: How many objects can the robot recognize? Can it recognize faces?
The robot is a platform that can be regularly upgraded with software updates. The robot is designed to recognize particular objects, such as conveyor belts and boxes. It is not designed to recognize faces, but it can be trained to recognize hundreds of different objects, assuming that these objects look different.
Question: What type of hands does the robot have?
We ship the robot with two different types of hands, one which is a suction gripper, and another with two fingers. The kit that ships with the robot has multiple finger sizes, multiple attachment points, and many different finger pads, and cross guards so there are probably at least a hundred different combinations of hands that could be made. But we also have an interface that allows our customers to design and build their own hands.
Question: And the robot has a camera built into its wrist?
Yes, that camera allows the robot to see what it is about to grasp, but it is also able to see its own fingers, so it needs to handle the fingers getting in the way of the object in its recognition algorithms. We developed an algorithm to be able to sense its fingers moving, regardless of the particular hand configuration. So it quickly adjusts to whatever hand it has and then is able to recognize objects even when they are partially occluded by the fingers, and then lift bottles, boxes, and a wide variety of other objects.
Question: What type of AI does the robot employ? What safety mechanisms does it employ?
The robot uses a behavior-based AI system. It has a modicum of common sense, so it know that if a person grabs something that it is carrying, it should go back to the box to get another one. Regarding safety, the robot is designed to slow down when a person gets close, and to shut down if the person gets really close. In fact, even if the power is lost with a human directly under the arm, the arm will descend so slowly that it will do no damage.
Question: Is there an emergency stop button for this robot?
There is, but I don’t anticipate anyone using it. Early mainframe computers also had a stop button, but these buttons were eliminated when it became obvious that they were completely superfluous.
Question: Can this robot be employed outside a factory setting?
Yes, we anticipate this robot being used in a variety of non-factory applications. The robot comes with a Software Development Kit (SDK). We expect many robotics labs to buy it because it is so inexpensive. So many people will do research on it, and they will undoubtedly devise some economically viable applications.
Question: Could this robot have military applications?
Although this robot could be employed in logistical operations for the military, we haven’t been pursuing that. When I worked at irobot, we aggressively pursued the military market, but we are positioning this machine in the civilian sector.
Question: How much will this robot cost?
A basic model with two hands, a head, and a body will cost $22,000. The customer will be able to configure it with added functionality, more parts, extended warranties, etc. This is much cheaper than any two armed industrial robot, and there are no integration costs as all the sensors, safety systems, and software are bundled. So it is an order of magnitude lower in total ownership cost. And on the research side when one compares the cost of this with a PR2 robot, which costs hundreds of thousands of dollars each, the value of our robot becomes apparent. That is why we are confident that hobbyists and research labs will buy these.
Question: Your robot currently is immobile. Are there plans to add mobility in the future?
That is definitely an option for a future model. We calculated that mobility would only be used 2% of the time in factories, so it didn’t make sense to add it in this robot. But adding mobility to this robot is definitely feasible – all one would need to do is bolt it to a mobile platform.
Question: What CPU does this robot have? Can it take advantage of cloud computing?
It has an Intel quad-core processor, and 32-bit embedded processors throughout its arms. It has an ethernet port, but the vast majority of small manufacturers have no network capabilities on the manufacturing area whatsoever. But once network capabilities become commonplace, we will definitely add internet connectivity.
Question: If there are no AI advances in the next decade, how will that stifle the robotics industry?
In everything from smartphones to autos, AI is already commonplace in society, although most people don’t appreciate that fact. Although I don’t foresee any major AI breakthroughs in the next decade, I am confident that steady progress will occur, and that it will be sufficient to meet the needs of the robots we will design.
Question: What field besides manufacturing will be transformed by robotics during the next decade?
There will be a tremendous push for robots in healthcare, particularly elderly care. One of my elderly relatives requires two people to get into bed at night, and that is bad for her and society. As baby boomers get old, they will embrace robots as a way to provide both practical and psychological benefits.
Question: How would you like to see robots reshape the economy in the coming decades?
I would like to see Rethink Robotics enable a renaissance in domestic manufacturing. I would also like to see robots infiltrate into an ever larger number of fields. We tend to overestimate technological progress in the short term, and underestimate it in the long term. But given the continuing decline in computing costs and the incessant increase in labor costs, the rise of robotics is inevitable.