Saturday, March 13, 2010

DARPA's ARM Challenge

UPDATE #2: Dr. Robert Mandelbaum gives IEEE Erico Guizzo a progress report as of mid-October on this project and provides wonderful photos. Click to read.

For many years now, engineers have designed robotic tools that can do specific tasks better than by hand.  For example, for putting in a screw they would simply design an arm with a screwdriver on the end.

But DARPA wants a multiple-purpose hand, actually two hands working together.  DARPA is attempting to get a robotic autonomous manipulator that mimics the human hand and which can be used for multiple purposes.  They want a hand that can do one task well and then do a completely different job immediately afterwards.

UPDATE: On March 16th, DARPA chose Barrett Technology to lead the hardware development portion of the project and RE2 (RE Squared) to do the integration. DARPA has yet to select a set of teams who will be given identical Barrett hardware and will be tasked to develop software to maximize the hardware's capabilities and meet DARPA's goals. RE2 will integrate the Barrett technology with various sensing technologies and a mobile platform for the teams to utilize during the program.

Initially, DARPA wants the hand to accomplish three major tasks:
  1. Open up a duffle bag, search through the contents and find a revolver that is hiding inside.  This requires the ability to handle flexible material like clothing and have force feedback to feel for objects (rather than just relying on sensor cues).  Further, they are asking for bimanual coordination, that is, the ability to use two hands at once.  One hand to hold onto the bag and the other to open the zipper.
  2. The next task will be rubble removal for search and rescue - picking up oddly shaped objects with both arms.
  3. The Third challenge will require that the arms insert one object into another, e.g.: a shell into a mortar.
One can easily see all the other civilian and defense uses for these two hands.  For defense, explosive ordnance disposal, casualty care, search and rescue, weapons support, checkpoint and access control, operations in extreme environments like space and underwater.  For civilian operations the uses are wide open - and exciting to ponder.  Hold a jar and unscrew the lid, assemble an object from a kit of parts, insert a battery into a device, pour a glass of water, fit a connector into a receptacle, put a cap on a pen, apply duct tape, remove duct tape, tie a knot, remove an item from a wallet, etc.  As a prosthetic or as the hand on a robotic assistant, this initiative is wonderful to behold.

There are two tracks: software and hardware.  Estimated time: 3 years and 9 months.

Software: in three phases:
  1. Grasping and manipulation - 15 months, up to 6 teams
  2. Complex grasping and bimanual manipulation - 15 months, up to 4 teams
  3. Mobile bimanual manipulation in real-world tasks - 18 months, up to 2 teams
Hardware: also three phases:
  1. Hand design and critical design review - 12 months, up to 3 teams
  2. Hand development and production of multiple copies - 15 months, 1 team
  3. Hand maintenance and support - 18 months, 1 team
The challenge for the hardware track is the hand cannot be prohibitively expensive.

It will take three years and nine months to complete the project... mid-2014.  I can't wait!

Tuesday, March 9, 2010

Robots in 3-5 and 5-10 years: View of the Future from Singularity University

Silicon Valley's new Singularity University offers a 10-day executive program that exposes participants to the opportunities and disruptive influences of exponentially growing technologies and to understand how these fields will affect their future and the futures of their companies and industries.

One avid participant wrote:
I think the positive surprise takeaway (for me at least) was how far we have advanced in artificial intelligence and especially robotics. Artificial intelligence has been promised to us for decades, and has been a disappointment for so long that I have consigned it to the dustbin of my research. Ditto for robots. I mean, seriously, if the Roomba (a glorified vacuum cleaner) is the best we can do after decades of work, how are AI and robots going to change the world? It is all well and good for a single-purpose robot to be designed to make a spot weld on a car, but a general-purpose robot seemed a long way off.

The robotic sessions were led by Dan Barry, a three-time astronaut and veteran of many space station adventures (as well as appearing on Survivor!). What I saw onscreen and heard about has made me rethink my doubts about robotics. There are significant strides being made in mobility and utility in robotics. I saw robots walking on four feet through very difficult terrain, on ice, and up stairs. Robot "hands" are a lot further along than I had thought. Mobile robots on wheels, and walking - balanced on two feet - are working today.

The ability of robots to recognize their surroundings, to differentiate between a table and a glass on the table (which is a very difficult thing to program), to pick up the glass, etc. is advancing at a fairly good pace. Dan is an enthusiastic advocate, and it was easy to get infected with his vision, but I can see a robotics industry in the 2020s actually having some significance in the US and world economy. We explored all manner of potential uses for robots, some with more economic potential than others. I am often asked where the jobs of the future will come from. It may be in robotics.

I was particularly drawn to the personal assistant robot. It is actually plausible to design a robot to be the "maid" in a home, to be able to purchase groceries, to assist the elderly, etc. These are the repeatable types of tasks that can be programmed and learned. We may only be ten years away from a nascent and powerful new industry. Now, this is not the robot of iRobot. It will not have intellectual conversations with you. But it will respond to voice commands and clean up, put away toys, etc. Cooking, however, other than microwave foods, is a LOT harder. You will have to make your own omelets for few decades.

John Mauldin, author: Thoughts from the Frontline
Dan Barry categorized robots into two types: special and general purpose.  He suggested that industry will have a steady need for the former but the real breakthrough and potentially disruptive technologies will be in the area of autonomous general purpose robots.

Barry emphasizes that artificial intelligence puts the brains into the robot and that in the next 3-5 years we will see better human-robot interfaces, some with gesture and/or speech recognition.  We can also expect, within the next 3-5 years, with the costs of sensors going down dramatically, better navigation and object avoidance systems and better object recognition and intelligent grasping.

NOTE: Assistive bots are already hitting the marketplace: The French company Robosoft is presently launching their Kompai robots which are designed to assist dependent people in their homes. The robot can speak, understand instructions, find its way around the house, access internet services and act as a telepresence robot for doctor-patient communication. Fascinating video.

Down the road in 7-10 years, as AI continues to improve, we can expect nurse bots that assist with pill dispensing, physical and other rehabilitation therapies, and can read and process sensors and instruments and assist with diagnosis. Companion and household robots are likely to be a big hit too. And swarm and nano bots will be entering the medical and entertainment marketplaces.

Watch this informative 15-minute interview of ex-NASA astronaut Dan Barry:

Tuesday, March 2, 2010

An international perspective on the service robotics boom

Last week's International Expert Days 2010, a two-day conference in a little factory town (Lauffin) in the Black Forest area of Germany, focused on the emerging service robotics sector. Very pretty and scenic. Nearly 100 participants attended and 22 spoke.

View from the smoker's porch at the Schunk factory conference center.

Most of the speakers expressed optimism about the emerging service robotics sector. Henrik Christensen, who spearheaded America's robotics roadmap, described progress of the new roadmap, the robotic caucus within Congress, and National Robotics Week, a first for the industry, and a necessary step to stimulate positive public awareness of robotics. Martin Haegele from the Fraunhofer Institute for Manufacturing Engineering and Automation cited 20% annual growth rates for service robotics for government and business and 30-40% for service robots for household and personal use.

Many economic drivers are propelling that growth:
  • Defense and security spending for research and devices
  • Consumer insistence on robotic minimally invasive surgery
  • Healthcare digitization and other efficiencies to reduce healthcare costs
  • Technological breakthroughs in surgeon augmentation and autonomous navigation
  • Movement by industrial robot manufacturers into unexplored non-automotive logistics and small and medium enterprise (SME) manufacturing
  • A desire by seniors to work with robots to assist them to stay independent instead of traditional eldercare facilities and hospitals
  • A shrinking legal labor market for jobs that are dirty, dangerous or dull
[Supporting the first driver is a recent (expensive) Research and Markets study on Military Robots and Unmanned Vehicles showing sales of $831 million in 2009 and projeced annual increases to a rate of $9.7 billion by 2016.]

Two speakers from the University of Heidelberg, one a surgeon, the other representing the University's engineering/robotics group, described the interest and progress in German university hospitals. They reported that there are 457 robotic surgical devices active in Germany of which less than 10% are available commercially. This level of activity is also occurring in the U.S. and Japan and beginning to gain attention from venture capital firms as well as regulatory agencies.

Solving country-specific needs with robotic solutions is the essence of the roadmap projects for Europe, Korea, Japan and last year in the US. Each roadmap identifies key barriers to accomplishing those countries' unique needs and suggests a timetable for their solution and commercialization. Rainer Bischoff, a EUROP executive, emphasized the need for closer academia-industry cooperation and faster technology transfers in Europe because the process is unacceptably slow. This was supported by a professor from the Technical University of Munich, who represented ECHORD, an EU clearing house for open robotics development.

There was much talk about new definitions and standards for the service robotics sector and the health and medical care segment within - safety and hygiene being paramount. Thomas Bömer, from the Institute for Occupational Safety and Health, discussed safety requirements for successful growth of robotic service companies as they move from caged industrial robots to robots that regularly interact with humans. He referenced the recent FDA workshop on the same subject just a few days earlier in Washington, DC.

Mike Stilman from GA Tech showed videos of an unusual navigation scheme which led to an incidental invention in motion and control in his quest to build a humanoid personal robot.

Many of the speakers described progress in artificial intelligence as it relates to robotic projects. Marc Ronthaler of the German Center for AI described how cognitive robotics learn through interaction with their surroundings and are capable of collecting information and of independently identifying options for actions.

Many hours of presentations focused on improvements in grasping and manipulating techniques of 3-fingered and 5-fingered hands to the frustrations of not yet being able to robotically handle limp or soft materials. Fascinating to see what is being worked on in research labs around the EU, but confusing because there are a wide range of successful grasping methodologies and hands already commercialized.

UPDATE: The confusion is ended and my education is updated.  There's research activity in robotic hands and grasping because the state of the art hasn't enabled robots to do multiple tasks, one right after another.  DARPA has just initiated a challenge to develop such a hand, actually two hands working together, that, when completed mid-2014, will be able to open the zipper of a duffle bag, riffle through the contents and remove a particular item.  Professor Molfino, who talked about her efforts to robotically handle clothing and other soft materials, was laboring on the very problem DARPA is challenging.

One speaker from the University of Bremen showed progress on their RoboWalker used to help in gait rehabilitation yet many surveys have shown that the two core requirements of rehab therapists are not yet being met: low entry costs and lasting muscle retraining.These economic and therapeutic metrics appear to be missing in almost all the areas where rehabilitation robotics are making their way through university research labs. Thus far there are no breakthrough affordable robotic therapeutic inventions entering the marketplace of parents, therapists, care facilities, schools, and hospitals.

Standardizations often facilitate growth and reduce duplication yet almost every personal robot project around the world has its own proprietary (or partially proprietary) software. Eric Berger from Willow Garage described their open source ROS software and PR2 personal robot hardware platforms and made a strong case for software standardization using ROS.

Even the Care-O-Bot project - a very capable and carefully developed robot assistant - isn't available for commercialization; rather it is offered as an open source hardware platform and education tool to help speed up development and deployment of assistive applications.

It brings up a sore subject: the process of technology transfer is broken and needs revamping; certainly it is extremely inefficient in its present form. There are conflicting interests: college professors are rewarded for publishing papers and getting grants. Their resulting inventions often have over-inflated values and require much work to adapt into a business framework, if adaptable at all. Businesses, on the other hand, want their research to solve real needs... they want directed research which appears to be anathema to the college set.

As an aside but directly related to my confusion and frustration, is the best one-liner I've heard so far this year:

Michael Thali is director of the University Forensic Institute in Bern, Switzerland and developed and heads up Virtopsy, a center for the performance of virtual autopsies. There are numerous advantages to a virtopsy: it is not invasive or destructive of tissue, provides dramatic 3-D pictures of skeletal and soft tissue injury, provides long-lasting data for trials and retrials, detects internal bleeding, missile paths, bone and missile fragmentation, fracture patterns - all of which are technically difficult to demonstrate to juries (particularly when the jurors must pore over gory photos dryly described by a forensic pathologist instead of self-evident digital scans and simulations), and enables a database never before available for education, science, training and simulation. In a recent phone conversation Thali said that although they do 400-500 virtual autopsies per year, the database of the scanned and imaged cadavers is not available to scientists for further study, or even to forensic scientists for forensics study. Privacy issues.

He is finding resistance to the deployment of the Virtopsy methods because of the high costs of the various scanners (CT, MRI and others) along with other medical analysis devices that comprise the Virtopsy system. Even in America, where there are serious economic drivers to reduce healthcare costs and to digitize data, American medical examiners just don't have the up-front funds to build Virtopsy facilities even if, down the road, it is a better solution and more cost efficient than present methods.

So, in answer to my question that these two situations must be personally frustrating to him he replied: "How can I say this in a very Swiss diplomatic way?" And then, after a very long pause, he quietly said: "Yes it is."

I see many benefits to the Virtopsy project - the cadaver database for research, education and pattern recognition; the ability to consolidate coroner operations with significant cost savings; the ability to perform autopsies in areas where religion prohibits present methods, etc. Nevertheless, there are valid criticisms as well.

A knowledgeable European robotics scientist said of the Virtopsy project:
What you see here is a common problem and exactly one of the challenges we face today. The Virtopsy project was developed by university researchers. They did it because it was fun to do and an interesting challenge. They did not do it in response to or because they saw a market need. Clearly the team hadn't performed the "due diligence" study to understand the business needs of medical examiners (MEs), pathologists and forensic scientists: what is really needed to improve the autopsy process, what would governments be willing to pay for such improved services, and can the system cost less than that?

In many cases you need a visual identification or you need tissue samples for later lab analysis. So the virtopsy could be used for 'detection' of places for sampling of tissue and fluids which could be collected by laparoscopy and other techniques, so minimum impact on the body occurs. But fully virtual might be a dream. Not to mention that most MEs are used to cutting bodies. How many morgues will be able to afford MR and CT scanners? Let alone the other analytic equipment and computers? So it is a great idea, but the end-users - the MEs - don't appear to have been an integral part of the formulation of the Virtopsy system.