Apologies for the light posting this week: the entire IEEE Spectrum team (both digital and print) was closeted away in meetings working on ways to better serve you, dear reader. Did we come up with some? Sure we did, but for now, they’re secret until we get them to work.
Leading the video news for today is research from UC Berkeley focused on teaching robots to learn tasks in ways that can be adapted to new situations, using a deep learning approach based on neural nets. The upshot is that it enables robots (like Berkeley’s PR2, named BRETT) to learn new tasks in a matter of hours and perform those tasks generally independently of their environment, all with a minimal amount of sensors.
This is stupendously important in two ways: first, it means that robots get significantly easier to teach, as opposed to requiring programming. And second, it means that robots are able to do useful stuff in useful environments, like your house as opposed to a robotics lab. Watch BRETT do his thing, and all the rest of our videos, starting right now.
“What we’re reporting on here is a new approach to empowering a robot to learn,” said Professor Pieter Abbeel of UC Berkeley’s Department of Electrical Engineering and Computer Sciences. “The key is that when a robot is faced with something new, we won’t have to reprogram it. The exact same software, which encodes how the robot can learn, was used to allow the robot to learn all the different tasks we gave it.”
“With more data, you can start learning more complex things,” [Abbeel] said. “We still have a long way to go before our robots can learn to clean a house or sort laundry, but our initial results indicate that these kinds of deep learning techniques can have a transformative effect in terms of enabling robots to learn complex tasks entirely from scratch. In the next five to 10 years, we may see significant advances in robot learning capabilities through this line of work.”
[ UC Berkeley ]
The U.S. Navy has found that it pays to listen to Rolf Mueller carry on about his bat research. From unmanned aerial systems to undersea communications, practical applications flow from the team headed by Mueller, an associate professor of mechanical engineering.
Learning how bats navigate through dense thickets without crashing into each other could also help unmanned aircraft designers create better delivery vehicles, Mueller says.
Mueller's work is supported in part by the Institute for Critical Technology and Applied Science at Virginia Tech as well as the U.S. Navy's Naval Sea Systems Command, in Newport, Rhode Island.
[ VT ]
There’s now just under two weeks until the DRC Finals, so here’s a pile of the latest video updates from teams willing to share.
[ MIT DRC ]
[ CMU DRC ]
[ IHMC DRC ]
Inspired by origami, the folding drone developed by a team at EPFL and NCCR Robotics unfurls and takes off in a third of a second. The moment it is turned on, the rotors engage, the articulated arms extend and the drone begins moving. This eager device could be quickly released in large numbers over a disaster zone in order to bring back photos and establish contact with people in need.
The arms, made of fibreglass and light inelastic polyester, fold up into the shape of a trapezoid. When not in use, they wrap horizontally around the body of the device. What sets this device apart is the ability of the arms to unfold by themselves: the force generated by the rotors causes the arms to move into place. The rotors turn in the same direction, causing the arms to rotate out the opposite way and open around two vertical folds. When the arms are fully extended, their upper section moves horizontally and locks the segment open. Small magnets hold everything in place.
For a quadrotor to maintain stability while in flight, two rotors diagonally across from each other need to turn clockwise while the other two spin anticlockwise. The drone must therefore reverse the spinning direction of two of the rotors before taking off. This action takes place automatically in under 50 milliseconds once a sensor has detected the arms have locked into place. This ingenious system has been patented and will be shown on 25 May at the International Conference on Robotics and Automation in Seattle, which will bring together some 2,000 robotics specialists.
[ EPFL ]
Want a folding drone of your own? Sprite is a very compact coaxial design that’s halfway to its goal on Kickstarter:
$800 puts you on the list for one.
[ Kickstarter ]
The New York Times has an excellent update on the testing of mind-controlled prosthetic arms:
[ NYT ]
And while we’re on the subject, from Caltech:
Neural prosthetic devices implanted in the brain's movement center, the motor cortex, can allow patients with amputations or paralysis to control the movement of a robotic limb—one that can be either connected to or separate from the patient's own limb. However, current neuroprosthetics produce motion that is delayed and jerky—not the smooth and seemingly automatic gestures associated with natural movement. Now, by implanting neuroprosthetics in a part of the brain that controls not the movement directly but rather our intent to move, Caltech researchers have developed a way to produce more natural and fluid motions.
Andersen and his colleagues wanted to improve the versatility of movement that a neuroprosthetic can offer by recording signals from a different brain region—the PPC. "The PPC is earlier in the pathway, so signals there are more related to movement planning—what you actually intend to do—rather than the details of the movement execution," he says. "We hoped that the signals from the PPC would be easier for the patients to use, ultimately making the movement process more intuitive. Our future studies will investigate ways to combine the detailed motor cortex signals with more cognitive PPC signals to take advantage of each area's specializations."
In the clinical trial, designed to test the safety and effectiveness of this new approach, the Caltech team collaborated with surgeons at Keck Medicine of USC and the rehabilitation team at Rancho Los Amigos National Rehabilitation Center. The surgeons implanted a pair of small electrode arrays in two parts of the PPC of a quadriplegic patient. Each array contains 96 active electrodes that, in turn, each record the activity of a single neuron in the PPC. The arrays were connected by a cable to a system of computers that processed the signals, decoded the intent of the subject, and controlled output devices that included a computer cursor and a robotic arm developed by collaborators at Johns Hopkins University.
After recovering from the surgery, the patient was trained to control the computer cursor and the robotic arm with his mind. Once training was complete, the researchers saw just what they were hoping for: intuitive movement of the robotic arm.
[ Caltech ]
Kuka’s Medical Assistant arm won’t perform surgery itself, but it will try to keep you from performing surgery badly:
[ Kuka LBR Med ]
Pepper is learning localization and navigation, but it doesn’t matter what it does, as long as it tells you about it in that cutesy voice:
[ Aldebaran ]
The UAE Innovation Challenge is a collaboration between Northrop Grumman and the UAE’s Higher Colleges of Technology to teach Emirati students how to design, build and fly unmanned aircraft.
After five months of working on their aircraft, more than 100 students gathered in Abu Dhabi to compete in flying competitions and oral presentations.
Teams of undergraduate and graduate students from around the country are demonstrating their excavator robots May 18-22 at the Kennedy Space Center Visitor Complex in Florida. During the Robotics Mining Competition, participating teams' custom-built, remote-controlled mining robots will traverse simulated Martian terrain features and excavate simulated regolith.
Yaskawa has a new “Motoman Mobile STEM Robotics cart,” but I have to wonder how compatible STEM education really is with that depressing industrial teach pendant:
[ Yaskawa ]
Robotdalen is a Swedish robotics initiative enabling commercial success of new ideas and research within robotics and automation. We support the development of new robotics solutions for the industry, field (autonomous vehicles) as well as the health care sector. This video demonstrates a number of products that we support with development and commercialization.
[ Robotdalen ]
Next month, DJI is hosting some sort of “dedicated aerial skill test for pilots” near Vienna, which looks not at all like most of the footage in this video:
Well, that’s a little bit over the top. And anyway, I don’t think this is the sort of flight that anyone was dreaming of since the dawn of time, since no people are actually flying in this competition: it’s just the robots. And that’s cool enough, I hope.
[ DJI ]
RoboNation TV has a spot on UMD’s RoboRaven, which I never seem to get tired of watching fly:
[ UMD Robotics ]
“The founder of Sirius XM satellite radio, Martine Rothblatt now heads up a drug company that makes life-saving medicines for rare diseases (including one drug that saved her own daughter's life). Meanwhile she is working to preserve the consciousness of the woman she loves in a digital file ... and a companion robot. In an onstage conversation with TED's Chris Anderson, Rothblatt shares her powerful story of love, identity, creativity, and limitless possibility.”
[ TED ]
Here’s the next two videos from the WeRobot 2015 conference:
Panel 2: “Robot Passports”
Panel 3: “Robotics Governance”
[ WeRobot 2015 ]
We’ll close with a pair of 10-minute videos from the Stanford-Berkeley Robotics Symposium 2014; not sure why they just popped up on YouTube this week (since they were uploaded last year), but whatevs, we can enjoy them anyway. For more videos from SBRS, check out this YouTube channel.
“Safe Learning in Robotics,” presented by Claire Tomlin:
“Planning and Control for Spacecraft and Space Robots,” presented by Marco Pavone: