Rethink Robotics (formerly Heartland Robotics) has come out of stealth mode with the announcement of Baxter -- a $22,000 dual-arm, human-scale robot with compliant joints. The details are available in Rethink Robotics' Baxter datasheet and brochure, but here are a few key highlights: dual 7-DoF arms with 5 lb (2.3kg) payload with a max no-load speed of 3.3 ft/sec (1m/sec). The arms are compliant owing to series-elastic actuators (SEAs) with force control and torque sensing at each joint. The robot torso sans-pedistal is 3'1" tall (94cm), and the robot has a reach of roughly 104cm. The robot weighs in at 165 lbs (75kg) and has a suite of sensors including: 5 camers (1 up top, 2 in the torso, and 2 "eye-in-hand"), a 360-deg. ring of ultrasonic range sensors in the head, and IR range sensors at the gripper, and (naturally) kinematics and torque sensing at each joint. Did I mention the starting price of $22,000 and that it starts shipping this October!?! This will be a HUGE deal for robotics. Comparable arms easily cost an order of magnitude more (~$100k each), so getting a full pair for $22k is going to completely change the game -- perhaps even more than the Kinect. It's an exciting time for robotics! Read on for pictures, an interview with Rod Brooks (CTO and co-founder of Rethink), and the press release.
Rod Brooks is the CTO and co-founder of Rethink Robotics (and co-founder of iRobot, former CSAIL director at MIT, and my academic grandfather). Rethink and Rod afforded me a great privilege: a half-hour conversation the day preceding the launch. Here's a rough summary based on my quick notes (Rod, correct me if I make any egregious mistakes):
Travis: In terms of sensing, you have the 5 cameras, some ultrasound sensors, kinematics, and joint torques. Why no 3D sensing (a la the Kinect or depth cameras)? How extensible is your hardware?
Rod: There are 5 cameras: one above the screen, two in the chest, and two "eye in hand" (wrist-mounted) cameras to help with manipulation. There's also the 360-degree sonar ring (eg. to detect people coming up behind the robot) and IR range sensors in each hand. Plus the kinematics and torques. We wanted to integrate a Kinect-like sensor, but it debuted too late in the development process.
We knew that this platform would be popular with researchers. The gripper is exchangeable. The entire system is self-contained -- no bulky external boxes [Travis: I'm looking at you, Kuka!]. We're running off a quad-core Intel machine, so there's plenty of computation. We've also included USB and ethernet onboard, so extensibility shouldn't be a problem.
Travis: On the software front...?
Rod: We plan to release a SDK to allow researchers to modify the system. It will be based on ROS.
Travis: I'm a mobile manipulation guy.... why no mobile base?
Rod: Not in this version. Two reasons. First, we already had a lot of complexity in 14-DoF worth of arms. Second, our early case studies suggested that mobility would only be useful in 3% of factory-type tasks -- such as standing stationary and doing small-object pick-and-place. [Travis: Fascinating! Wouldn't have guessed it was that low. Building a quick plug-and-play mobile base for research labs could be a nice little startup opportunity! Lots of labs will want 'em.]
Travis: OK, my one real gripe about existing hardware: it looks like its lacking a true wrist (ie. spherical joint at the end effector). Why?
Rod: Mechanical complexity and cost tradeoff. To keep the price reasonable, we have some clever gearboxes designs... the complexity of a true wrist would've been too much.
Travis: In the press release, it says Baxter can "Apply common sense to its environment." From a research perspective, that seems pretty vacuous. What does that even mean?
Rod: Indeed. We don't mean "common sense" from a Marvin Minsky-like strong AI perspective. Baxter's "execution" application consists of a series of behavior-based systems. During "training," the robot detects task-relevant features and uses it to build up the behavior based system.
For example, let's say a user is training the robot for a pick and place task. During the "pick" phase, a user places the gripper above an object and closes the gripper. The force on the gripper is detected by the robot. Our "training" application detects this sequence as "the robot is grasping an object"... so during "execution", Baxter won't proceed unless it actually detects an object in the robots gripper. Thus, if the object fell out, it would stop (or do something else). This is different from how existing industrial robots work -- they'd just merrily continue the pick-and-place without the object.
Collectively, these "behavior primitives" [Travis: my phrasing] are assigned and composed, ie. "learned", during "training" by having non-technical users directly manipulate the robot rather than programming it (which is also possible for those inclined). This gives the robot an air of common sense.
Travis: The press release also mentions "No need for cages." I know SEA's, force sensing, and the payload make the robot relatively safe (based on my experiences with Meka robots and the PR2). However, how do you handle existing safety regulations? I thought E-Stops, cages, and light curtains were required for industrial settings.
Rod: You're right. That whole space is in flux. Existing standards just aren't designed for these types of arms... they're too used to big, dangerous, position-controlled arms. Thankfully, factories can do their own safety analysis. We've already gone through this process, and the results are promising.
Furthermore, almost everyone in the industry is trying to revise the standards -- from Kuka, ABB, and Universal Robotics. It looks live everyone is aligned to make this happen, and things like ISO Standards for robot safety are moving in that direction
Travis: I have my own take on this... but why now? What is special about right now that makes this possible compared to 10-15 years ago?
Rod: The usual suspects: a market need, inexpensive manufacturing, improved sensing, and the big one: computation. The computation is such that sensing (eg. vision, 3D, tracking) and kinematics (eg. gravity comp, force control, anti-collision planning) are much more feasible. The time is ripe for disruption.
[ Travis: Heh, Rod mentioned using the screen to assist with perception, eg. using it as a uniform, known background for visual segmentation. Cool idea! You could even flash a texture on the screen to do 3D imaging sans a range camera. Sorta like Grant Schindler's (Georgia Tech alum) "trimensional" app on the iPhone. ]
Travis: The Gill Pratt / Matt Williamson SEA patent is set to expire shortly (2015? It was filed in 1995 and issued in 1997). Are you worried about competitors (eg. Redwood Robotics) or knockoffs?
Rod: We've got an exclusive on the SEA patent, so we're not worried short-term. Plus, we've got several key patents in the pipeline (filed). We think we're pretty well covered. In any case, it's an exciting time for robotics.
Travis: Do you have any advice to budding entrepreneurs? Rethink has raised something like $50 Million in VC funding and seems to be on the verge of success. What's your secret?
Rod: Here's the thing... If you had gone to any VC four years ago and pitched, say, an elder-care robot... they would've told you "you're crazy." Someone needs to prove that a super low-cost platform is possible first. Rethink is it. Now you can point to Rethink and say "see it's possible" and even build it (hopefully) using our arms. This should spawn a large number of companies using our arms in ways we never imagined -- perhaps bypassing the research labs entirely. Get out there and solve problems.
(Note: this is currently the unofficial version. I'll update it with the official one and a link when it's available.)
Rethink Robotics Revolutionizes Manufacturing with Humanoid Robot
BOSTON, Mass.,– September 18, 2012 – Ushering in a new era in robotics, Rethink Robotics today unveiled the world’s first humanoid robots capable of applying common sense automation behavior to manufacturing environments. Affordably priced, versatile and safe enough to work shoulder-to-shoulder with people, Rethink’s new line of Baxter robots redefine how small, mid-size and large domestic manufacturers use automation to compete with manufacturers in low-cost regions of the world.
Based on the vision of Rodney Brooks, co-founder of iRobot and former director of the MIT Computer Science and Artificial Intelligence Lab, Baxter breaks down the cost and safety barriers that have limited automation in American manufacturing up until now.
“Roboticists have been successful in designing robots capable of super-human speed and precision. What’s proven more difficult is inventing robots that can act as we do--in other words, able to inherently understand and adapt to their environments,” said Rodney Brooks, founder and CTO of Rethink Robotics. “We believed that if we could cross that chasm with the manufacturing environment specifically in mind, we could offer new hope to the millions of American manufacturers who are looking for innovative ways to compete in our global economy.”
Rethink’s Baxter robots are groundbreaking in six different ways. No other product on the market can:
- Apply common sense to its environment – Baxter understands and can adapt to the way the world works. For example, if Baxter drops an object, it knows to retrieve it before doing the next task.
- Safely work alongside humans – Baxter comes complete with sensors and related software to help it see and understand its environment. These design elements coupled with its ability to apply common sense, allow Baxter to work next to people – eliminating the need for cages to keep robots and people separate. This is unheard of in manufacturing robotics.
- Be trained with no expertise required – Unlike traditional robots that require sophisticated software programming, Baxter can be trained just as you would teach a person. Low-skilled workers can interact with the robot directly to train it to do a task in less than 30 minutes.
- Offer broad task flexibility – Because of its versatility and the short amount of time it takes to retrain, Baxter can be moved to different and varying tasks over the course of a day, week and month. This flexibility can make the difference for contract manufacturers who need to quickly make adjustments to meet demand.
- Bring automation to manufacturers of all sizes – Baxter was specifically designed to be affordable for midsize and small manufacturers, companies that have never been able to afford robotics before. The list price for the Baxter robot is $22,000 and with zero integration required, the solution is a fraction of the cost of a traditional industrial robot.
- Go from packaging to factory floor in one hour – Baxter is a complete and self-contained system. Unlike traditional robots that require manufacturers to make additional capital investments and develop custom software, Baxter comes ready to be used on the factory floor in less than one hour.
Manufacturers like Nypro have already seen the potential of Rethink’s Baxter. “Our customers come to us because they trust us to bring the highest quality and the best value,” said Michael McGee, Director of Technology, Nypro Inc.. “Our business has been built on that trust one customer at a time for over half a century, around the world. Rethink and their advanced robotics platform Baxter, helps us to sustain this offering and enables us to continue to bring solutions to today’s complex manufacturing challenges.”
Baxter is a complete system with application software that will be enhanced through regular software updates. Rethink will also make a software development kit (SDK) available in early 2013, opening up Baxter-centered innovation to programmers, robot enthusiasts, inventors and anyone else with a passion for cutting edge technology.
“Our guiding vision was to create a robot that would break the mold in two distinct ways. First, the robot had to inherently function in a totally new way,” said Scott Eckert, CEO of Rethink Robotics. “We also set out to create a platform that would form the foundation for future software updates. As we release new software building on Baxter’s abilities, our customers can take advantage by simply updating the software. And with the SDK, the opportunity to think of new and novel ways to use Baxter is open to the world at large with our customers as the primary beneficiaries.”
From innovation to production, Baxter is uniquely a product of the U.S. “It was important to us that we remain true to the vision of the company and Baxter is proof positive that you can manufacture a superior product cost effectively with American manufacturing,” Eckert concluded.
Already on pre-order, the first product shipments of Baxter will begin in October 2012.
About Rethink Robotics
Rethink Robotics, Inc. developed the world’s first robot with common sense to increase productivity and efficiency in manufacturing environments. Introducing robotics into places that have never been automated before, Rethink’s Baxter robot is making manufacturers more efficient, their workers more productive and keeping jobs from migrating to low-cost regions. Based in the Innovation District of Boston, Massachusetts, the company is funded by Charles River Ventures, Highland Capital Partners, Sigma Partners, Draper Fisher Jurvetson and Bezos Expeditions, the personal investment company of Jeff Bezos. For more information about Rethink Robotics, please visit www.rethinkrobotics.com .