MIT Research News' Journal
[Most Recent Entries]
[Calendar View]
Monday, September 15th, 2014
| Time |
Event |
| 12:00a |
Bound for robotic glory Speed and agility are hallmarks of the cheetah: The big predator is the fastest land animal on Earth, able to accelerate to 60 mph in just a few seconds. As it ramps up to top speed, a cheetah pumps its legs in tandem, bounding until it reaches a full gallop.
Now MIT researchers have developed an algorithm for bounding that they’ve successfully implemented in a robotic cheetah — a sleek, four-legged assemblage of gears, batteries, and electric motors that weighs about as much as its feline counterpart. The team recently took the robot for a test run on MIT’s Killian Court, where it bounded across the grass at a steady clip.
In experiments on an indoor track, the robot sprinted up to 10 mph, even continuing to run after clearing a hurdle. The MIT researchers estimate that the current version of the robot may eventually reach speeds of up to 30 mph.
The key to the bounding algorithm is in programming each of the robot’s legs to exert a certain amount of force in the split second during which it hits the ground, in order to maintain a given speed: In general, the faster the desired speed, the more force must be applied to propel the robot forward. Sangbae Kim, an associate professor of mechanical engineering at MIT, hypothesizes that this force-control approach to robotic running is similar, in principle, to the way world-class sprinters race.
“Many sprinters, like Usain Bolt, don’t cycle their legs really fast,” Kim says. “They actually increase their stride length by pushing downward harder and increasing their ground force, so they can fly more while keeping the same frequency.”
Kim says that by adapting a force-based approach, the cheetah-bot is able to handle rougher terrain, such as bounding across a grassy field. In treadmill experiments, the team found that the robot handled slight bumps in its path, maintaining its speed even as it ran over a foam obstacle.
“Most robots are sluggish and heavy, and thus they cannot control force in high-speed situations,” Kim says. “That’s what makes the MIT cheetah so special: You can actually control the force profile for a very short period of time, followed by a hefty impact with the ground, which makes it more stable, agile, and dynamic.”
Kim says what makes the robot so dynamic is a custom-designed, high-torque-density electric motor, designed by Jeffrey Lang, the Vitesse Professor of Electrical Engineering at MIT. These motors are controlled by amplifiers designed by David Otten, a principal research engineer in MIT’s Research Laboratory of Electronics. The combination of such special electric motors and custom-designed, bio-inspired legs allow force control on the ground without relying on delicate force sensors on the feet.
Kim and his colleagues — research scientist Hae-Won Park and graduate student Meng Yee Chuah — will present details of the bounding algorithm this month at the IEEE/RSJ International Conference on Intelligent Robots and Systems in Chicago.
Toward the ultimate gait
The act of running can be parsed into a number of biomechanically distinct gaits, from trotting and cantering to more dynamic bounding and galloping. In bounding, an animal’s front legs hit the ground together, followed by its hind legs, similar to the way that rabbits hop — a relatively simple gait that the researchers chose to model first.
“Bounding is like an entry-level high-speed gait, and galloping is the ultimate gait,” Kim says. “Once you get bounding, you can easily split the two legs and get galloping.”
As an animal bounds, its legs touch the ground for a fraction of a second before cycling through the air again. The percentage of time a leg spends on the ground rather than in the air is referred to in biomechanics as a “duty cycle”; the faster an animal runs, the shorter its duty cycle.
Kim and his colleagues developed an algorithm that determines the amount of force a leg should exert in the short period of each cycle that it spends on the ground. That force, they reasoned, should be enough for the robot to push up against the downward force of gravity, in order to maintain forward momentum.
“Once I know how long my leg is on the ground and how long my body is in the air, I know how much force I need to apply to compensate for the gravitational force,” Kim says. “Now we’re able to control bounding at many speeds. And to jump, we can, say, triple the force, and it jumps over obstacles.”
In experiments, the team ran the robot at progressively smaller duty cycles, finding that, following the algorithm’s force prescriptions, the robot was able to run at higher speeds without falling. Kim says the team’s algorithm enables precise control over the forces a robot can exert while running.
By contrast, he says, similar quadruped robots may exert high force, but with poor efficiency. What’s more, such robots run on gasoline and are powered by a gasoline engine, in order to generate high forces.
“As a result, they’re way louder,” Kim says. “Our robot can be silent and as efficient as animals. The only things you hear are the feet hitting the ground. This is kind of a new paradigm where we’re controlling force in a highly dynamic situation. Any legged robot should be able to do this in the future.”
This work was supported by the Defense Advanced Research Projects Agency. | | 10:22a |
Reducing traffic congestion, remotely At the Intelligent Transportation Systems World Congress last week, MIT researchers received one of the best-paper awards for a new system, dubbed RoadRunner, that uses GPS-style turn-by-turn directions to route drivers around congested roadways.
In simulations using data supplied by Singapore’s Land Transit Authority, the researchers compared their system to one currently in use in Singapore, which charges drivers with dashboard-mounted transponders a toll for entering congested areas.
The Singapore system gauges drivers’ locations with radio transmitters mounted on dozens of gantries scattered around the city, like the gantries used in many U.S. wireless toll systems. RoadRunner, by contrast, uses only handheld devices clipped to cars’ dashboards. Nonetheless, in the simulations, it yielded an 8 percent increase in average car speed during periods of peak congestion.
Moreover, for purposes of comparison, the MIT researchers restricted themselves to road-access patterns dictated by Singapore’s existing toll system. Modifying those patterns — encouraging or discouraging the use of different stretches of road — could, in principle, lead to even greater efficiency gains.
“With our system, you can draw a polygon on the map and say, ‘I want this entire region to be controlled,’” says Jason Gao, a graduate student in electrical engineering and computer science who developed the new system together with his advisor, Professor of Electrical Engineering and Computer Science Li-Shiuan Peh. “You could do one thing for a month and test it out and then change it without having to dig up roads or rebuild gantries.”
Gao and Peh also tested their system on 10 cars in Cambridge, Mass. Of course, 10 cars is not enough to dramatically affect local traffic patterns. But it was enough to evaluate the efficiency of the communications system and of the vehicle-routing algorithm. It also provided reliable data about the system’s performance for use in simulations.
Max capacity
Urban toll systems like the one in Singapore designate certain regions — with gantries at every entry point — as prone to congestion. Drivers are charged a fee for entering any such region, so they have an incentive to avoid it. The fee fluctuates over the course of the day, according to historical traffic data.
RoadRunner, by contrast, assigns each such region a maximum number of cars. Any car entering the region must acquire a virtual authorization that Gao and Peh call a “token.” If no tokens are free, RoadRunner routes the car around the region using turn-by-turn voice prompts.
The version of RoadRunner used in the Cambridge tests was largely decentralized: A car leaving a region would wirelessly announce that its token was available, and a car seeking to enter the region would request it. The system used a wireless standard called 802.11p, a variation on Wi-Fi that uses a narrower slice of the electromagnetic spectrum but is licensed for higher-power transmissions, so that it has a much larger broadcast range.
It could be that the time savings promised by RoadRunner would be enough to induce commuters to use it. But it would also be possible to modify the system so that any car entering a congestion-prone region without a token would be assessed a small fine.
Reporting a car for tokenless entry would require uploading data to a central server, but it wouldn’t require specifying the car’s location at a resolution finer than that of the region. So Gao believes that, even though RoadRunner relies on GPS data, it wouldn’t compromise drivers’ privacy any more than existing urban toll systems do. In fact, he argues, it would compromise privacy less, since cars that followed the system’s routing instructions would never have their locations reported.
An app for that
In their experiments, Gao and Peh used cellphones to control commercial 802.11p radios, which are about the size of a typical electronic-toll dashboard transponder. But in the future, it may be possible to embed the radios directly into cellphones.
At the International Symposium on Low Power Electronics and Design in August, Gao, Peh, and lead author Pilsoon Choi, a postdoc in Peh’s group, together with researchers at Nanyang Technological University in Singapore, presented a paper demonstrating that an 802.11p radio built from gallium nitride and controlled by silicon electronics would consume half the power that existing radios do.
Moreover, the Singapore-MIT Alliance for Research and Technology (SMART) has developed a technique for integrating gallium nitride into existing silicon-chip manufacturing processes and is currently building a chip-fabrication facility to implement it.
“In Singapore, the government already requires every single registered vehicle to have a dash-mounted transponder,” Gao says. “That’s already there, so you might as well take advantage of it. In other places, where you don’t have that in place, it would be easier to deploy it if you said, ‘You can download this app and just leave your cellphone on your dashboard.’”
“A distributed decision process is an alternative to centralized models that has to be explored and, as far as I know, has been rarely if not ever addressed,” says Jean Bergounioux, secretary general of ATEC ITS France, a French industrial research consortium dedicated to novel transportation systems. “RoadRunner offers the possibility of decentralizing as many decisions as possible at the lower level, without excluding that global decisions be made at the upper level.”
“It's worth getting into field trial as soon as possible to test and evaluate the feasibility of its industrial development and deployment,” Bergounioux adds. | | 12:45p |
Renewing a place of faith Singular and cylindrical, iconic and graceful, the MIT Chapel (W15) has served the Institute community for close to six decades – and is now the focus of a substantial renewal effort that begins today, Monday, Sept. 15.
Work on the Chapel will continue through February, during which time the building will be closed. During the months of its closure, programs have arranged for alternative spaces for worship and other activities that would normally take place in the Chapel. Please visit the Religious Life website for relocation information, schedules, and updates.
A plan for renewal
During the renewal, the Chapel's stained glass walls will be repaired, restored, and protected; its spire and bell tower will be removed temporarily to allow the roof to be replaced. Exterior work will include brick repair and repointing, and the rebuilding of the central skylight.
The Chapel's moat will be rebuilt to incorporate a new filtration system and design elements to prevent water leakage. This work will restore the beauty of the moat, which allows soft secondary light to fill the Chapel through openings at the base of its walls.
Inside, the Chapel's travertine floors will be upgraded and renewed, brick will be cleaned, wood surfaces will be refinished as needed, and new railings will be added to the organ loft. Plumbing, fire protection, and heating, ventilation, and air conditioning systems will be replaced or upgraded as needed.
Landmark building
Designed by Finnish-American architect Eero Saarinen, the Chapel was dedicated along with Saarinen’s Kresge Auditorium in May 1955. The round, windowless structure features a domed central skylight that illuminates a cascading metal sculpture of rods and crosspieces, created by Harry Bertoia as an altarpiece screen. At the time of the dedication, some members of the MIT community found the Chapel’s modernist architecture an affront to the classical style found elsewhere on campus. James Killian, then MIT's president, responded by saying, “MIT should be forward-looking in its architecture as well as in its research and education.”
Today, the Chapel serves as a nondenominational gathering place for the MIT community, offering space for worship, meditation, and private ceremonies. The renewal project will help preserve this unique building as a welcoming place of faith for generations to come.
Navigating the site
A site fence will be installed around the work area today. To facilitate access around the site, a temporary walkway will be established on the west side of the Chapel, facing Kresge Auditorium. Amherst Street will remain open during the renovation project.
Questions about this project can be directed to Brian Healy of MIT's Department of Facilities: healyb@mit.edu. | | 3:00p |
Neuroscientists identify key role of language gene Neuroscientists have found that a gene mutation that arose more than half a million years ago may be key to humans’ unique ability to produce and understand speech.
Researchers from MIT and several European universities have shown that the human version of a gene called Foxp2 makes it easier to transform new experiences into routine procedures. When they engineered mice to express humanized Foxp2, the mice learned to run a maze much more quickly than normal mice.
The findings suggest that Foxp2 may help humans with a key component of learning language — transforming experiences, such as hearing the word “glass” when we are shown a glass of water, into a nearly automatic association of that word with objects that look and function like glasses, says Ann Graybiel, an MIT Institute Professor, member of MIT’s McGovern Institute for Brain Research, and a senior author of the study.
“This really is an important brick in the wall saying that the form of the gene that allowed us to speak may have something to do with a special kind of learning, which takes us from having to make conscious associations in order to act to a nearly automatic-pilot way of acting based on the cues around us,” Graybiel says.
Wolfgang Enard, a professor of anthropology and human genetics at Ludwig-Maximilians University in Germany, is also a senior author of the study, which appears in the Proceedings of the National Academy of Sciences this week. The paper’s lead authors are Christiane Schreiweis, a former visiting graduate student at MIT, and Ulrich Bornschein of the Max Planck Institute for Evolutionary Anthropology in Germany.
All animal species communicate with each other, but humans have a unique ability to generate and comprehend language. Foxp2 is one of several genes that scientists believe may have contributed to the development of these linguistic skills. The gene was first identified in a group of family members who had severe difficulties in speaking and understanding speech, and who were found to carry a mutated version of the Foxp2 gene.
In 2009, Svante Pääbo, director of the Max Planck Institute for Evolutionary Anthropology, and his team engineered mice to express the human form of the Foxp2 gene, which encodes a protein that differs from the mouse version by only two amino acids. His team found that these mice had longer dendrites — the slender extensions that neurons use to communicate with each other — in the striatum, a part of the brain implicated in habit formation. They were also better at forming new synapses, or connections between neurons.
Pääbo, who is also an author of the new PNAS paper, and Enard enlisted Graybiel, an expert in the striatum, to help study the behavioral effects of replacing Foxp2. They found that the mice with humanized Foxp2 were better at learning to run a T-shaped maze, in which the mice must decide whether to turn left or right at a T-shaped junction, based on the texture of the maze floor, to earn a food reward.
The first phase of this type of learning requires using declarative memory, or memory for events and places. Over time, these memory cues become embedded as habits and are encoded through procedural memory — the type of memory necessary for routine tasks, such as driving to work every day or hitting a tennis forehand after thousands of practice strokes.
Using another type of maze called a cross-maze, Schreiweis and her MIT colleagues were able to test the mice’s ability in each of type of memory alone, as well as the interaction of the two types. They found that the mice with humanized Foxp2 performed the same as normal mice when just one type of memory was needed, but their performance was superior when the learning task required them to convert declarative memories into habitual routines. The key finding was therefore that the humanized Foxp2 gene makes it easier to turn mindful actions into behavioral routines.
The protein produced by Foxp2 is a transcription factor, meaning that it turns other genes on and off. In this study, the researchers found that Foxp2 appears to turn on genes involved in the regulation of synaptic connections between neurons. They also found enhanced dopamine activity in a part of the striatum that is involved in forming procedures. In addition, the neurons of some striatal regions could be turned off for longer periods in response to prolonged activation — a phenomenon known as long-term depression, which is necessary for learning new tasks and forming memories.
Together, these changes help to “tune” the brain differently to adapt it to speech and language acquisition, the researchers believe. They are now further investigating how Foxp2 may interact with other genes to produce its effects on learning and language.
This study “provides new ways to think about the evolution of Foxp2 function in the brain,” says Genevieve Konopka, an assistant professor of neuroscience at the University of Texas Southwestern Medical Center who was not involved in the research. “It suggests that human Foxp2 facilitates learning that has been conducive for the emergence of speech and language in humans. The observed differences in dopamine levels and long-term depression in a region-specific manner are also striking and begin to provide mechanistic details of how the molecular evolution of one gene might lead to alterations in behavior.”
The research was funded by the Nancy Lurie Marks Family Foundation, the Simons Foundation Autism Research Initiative, the National Institutes of Health, the Wellcome Trust, the Fondation pour la Recherche Médicale, and the Max Planck Society. |
|