Darpa Robotics Challenge: the search for the perfect robot soldier

The Atlas robot looks something out of the post-apocalyptic future, or maybe a Will Smith blockbuster. It's a 330lb cyborg with eerily human-like hands and a head equipped with a laser. It lunges forward with a grim, deliberate clatter on curved slices of metal for feet.

This bot will be a first responder in times of crisis, says the Pentagon – it's designed to use tools and trudge through difficult terrain, heading into smoky, dangerous areas that humans can't. With its human capabilities it could also, not coincidentally, be a good soldier. This robot – named Atlas after the Greek mythological hero who supported the world on his back – is part of the Pentagon's quest to create a humanoid robot that can do everything from turning knobs in nuclear plants to driving a car.

Atlas sits on his massive hydraulic haunches at the leading edge of robotics. And the US military wants more like him. Starting today, the Atlas is the center of a $2m robot beauty pageant – the Darpa Robotics Challenge – held by the US government's Defense Advanced Research Projects Agency. Darpa is the technical-innovation arm of the US military; its purpose is to "create strategic surprise for US adversaries by maintaining the technological superiority of the US military." The internet, for instance, started out as a Darpa project.

For the Darpa Robotics Challenge, the agency is offering as much as $34m in awards money to contestants. They will include software developers and engineers vying to create the operating system that will rule future Pentagon robots as well as robot-makers themselves. If successful, Atlas and others machines in the competition could be put to work in future disaster zones deemed too dangerous for humans; their abilities could include turning a valve, connecting a hose to a spigot, and even driving a car.

Boston Dynamics – the Waltham, Mass-based company that co-created Atlas – will provide copies of the cyborg to teams that will compete to win military prizes for designing the best software to bring Atlas to life. Already Atlas, and its predecessors, can rapidly climb stairs, do more push-ups than any human, and even pass for a person while donning a chemical protection suit.

Chris Atkeson thinks the best robots for such tasks should not only have the dexterity of a human, but also can assess a situation and model a plan. As a professor in the Robotics Institute at Carnegie Mellon University, and an expert in cognitive science, he's leading one of the school's two teams in the Darpa competition.

Atkeson's mission is to create robots that base their actions on an ability to observe the changing world around them. He says his robot software designs are like investment strategies in the stock market that use models and algorithms. While other robots might make snap decision, like investors who buy stocks based on their gut feeling, Atkeson is designing robots to think about the "probabilities" of potential outcomes.

"In the old days there were a lot of people who went with their gut and would invest in General Electric or something. You ask them why, and they say: 'Oh you know it feels good,'" he said "The transition in robotics that I'm interested in is getting to the point where robot behavior is determined by the quants." The goal of a semi-autonomous robot able to perform various diverse tasks on its own is still years away, Atkeson said, but Darpa has a history of pushing technologies borrowed from the future. The internet appeared similarly improbable when that agency began to funnel money into its development, Atkeson said.

"If you had run around and said, 'we're going to have almost universal access to a massive computer network,' if you had said that in the 1960s and 70s, people would have shrugged,'' he said. ''They wouldn't have even understood the benefits of what you're talking about, and robots are the same thing." Atkeson will demonstrate the Carnegie Mellon team's bionic brain in the first heat of the contest, starting June 10, when software developers will compete to create the best program to run a robot. The winners of that first leg will get a crack at deploying their programs in Atlas himself in a later event of the competition.

Perhaps, not surprisingly, the US military is also increasing its use of robots - or unmanned systems, as it calls them. The Unmanned Systems Integrated Roadmap, a paper published by the Defense Department in 2011, outlines its strategy for expanding robotics technologies through 2036 to include war machines. Despite the ongoing controversy over drones, the Pentagon contends that robots will make military operations to be faster and more affordable.

The Darpa Robotic Challenge isn't the only effort to expand robotics research in the US. In June 2011, President Obama shared a stage at Carnegie Mellon with a boxy four-wheeled yellow SensaBot – which looks a golden sibling of Pixar's Wall-E – designed to inspect industrial facilities. At the time, Obama announced the Advanced Manufacturing Partnership, a plan to fund new research that would lead to job growth in factories. About $70m of the $500m funding is allotted for projects in "next-generation robotics."

The moral aspect of creating humanoid military machines weighs heavily on some. US army Lt Col Douglas Pryer oversaw the deployment of hand-launched reconnaissance planes while commanding a company in Iraq in 2004. In a column titled Rise of the Machines and published in the March/April edition of Military Review, he described his concern that machines, as the distance they allow troops to remain from the battlefield, will encourage troops to feel indifferent toward civilian deaths.

"Since we were physically removed from the action, maybe such an event would not affect us much. Would it look and feel, I wondered, like sitting at home, a can of Coke in hand, watching a war movie," he said. Atkeson, of Carnegie Mellon, also comes from a military family. He acknowledges the ethical issue involved with developing robots for the Department of Defense. He reasons that as long as the military exists, it might as well be an efficient one: "If you ask me a technical question of who can fly an airplane better, I think a machine can."

View the original article here

Tiny compound semiconductor transistor could challenge silicon’s dominance

Silicon’s crown is under threat: The semiconductor’s days as the king of microchips for computers and smart devices could be numbered, thanks to the development of the smallest transistor ever to be built from a rival material, indium gallium arsenide.

The compound transistor, built by a team in MIT’s Microsystems Technology Laboratories, performs well despite being just 22 nanometers (billionths of a meter) in length. This makes it a promising candidate to eventually replace silicon in computing devices, says co-developer Jesús del Alamo, the Donner Professor of Science in MIT’s Department of Electrical Engineering and Computer Science (EECS), who built the transistor with EECS graduate student Jianqian Lin and Dimitri Antoniadis, the Ray and Maria Stata Professor of Electrical Engineering.

To keep pace with our demand for ever-faster and smarter computing devices, the size of transistors is continually shrinking, allowing increasing numbers of them to be squeezed onto microchips. “The more transistors you can pack on a chip, the more powerful the chip is going to be, and the more functions the chip is going to perform,” del Alamo says.

But as silicon transistors are reduced to the nanometer scale, the amount of current that can be produced by the devices is also shrinking, limiting their speed of operation. This has led to fears that Moore’s Law — the prediction by Intel founder Gordon Moore that the number of transistors on microchips will double every two years — could be about to come to an end, del Alamo says.

To keep Moore’s Law alive, researchers have for some time been investigating alternatives to silicon, which could potentially produce a larger current even when operating at these smaller scales. One such material is the compound indium gallium arsenide, which is already used in fiber-optic communication and radar technologies, and is known to have extremely good electrical properties, del Alamo says. But despite recent advances in treating the material to allow it to be formed into a transistor in a similar way to silicon, nobody has yet been able to produce devices small enough to be packed in ever-greater numbers into tomorrow’s microchips.

Now del Alamo, Antoniadis and Lin have shown it is possible to build a nanometer-sized metal-oxide semiconductor field-effect transistor (MOSFET) — the type most commonly used in logic applications such as microprocessors — using the material. “We have shown that you can make extremely small indium gallium arsenide MOSFETs with excellent logic characteristics, which promises to take Moore’s Law beyond the reach of silicon,” del Alamo says.

Transistors consist of three electrodes: the gate, the source and the drain, with the gate controlling the flow of electrons between the other two. Since space in these tiny transistors is so tight, the three electrodes must be placed in extremely close proximity to each other, a level of precision that would be impossible for even sophisticated tools to achieve. Instead, the team allows the gate to “self-align” itself between the other two electrodes.

The researchers first grow a thin layer of the material using molecular beam epitaxy, a process widely used in the semiconductor industry in which evaporated atoms of indium, gallium and arsenic react with each other within a vacuum to form a single-crystal compound. The team then deposits a layer of molybdenum as the source and drain contact metal. They then “draw” an extremely fine pattern onto this substrate using a focused beam of electrons — another well-established fabrication technique known as electron beam lithography.

Unwanted areas of material are then etched away and the gate oxide is deposited onto the tiny gap. Finally, evaporated molybdenum is fired at the surface, where it forms the gate, tightly squeezed between the two other electrodes, del Alamo says. “Through a combination of etching and deposition we can get the gate nestled [between the electrodes] with tiny gaps around it,” he says.

Although many of the techniques applied by the team are already used in silicon fabrication, they have only rarely been used to make compound semiconductor transistors. This is partly because in applications such as fiber-optic communication, space is less of an issue. “But when you are talking about integrating billions of tiny transistors onto a chip, then we need to completely reformulate the fabrication technology of compound semiconductor transistors to look much more like that of silicon transistors,” del Alamo says.

The team presents its work this week at the International Electron Devices Meeting in San Francisco.

Their next step will be to work on further improving the electrical performance — and hence the speed — of the transistor by eliminating unwanted resistance within the device. Once they have achieved this, they will attempt to further shrink the device, with the ultimate aim of reducing the size of their transistor to below 10 nanometers in gate length.

Matthias Passlack, of Taiwanese semiconductor manufacturer TSMC, says del Alamo’s work has been a milestone in semiconductor research. “He and his team have experimentally proven that indium arsenide channels outperform silicon at small-device dimensions,” he says. “This pioneering work has stimulated and facilitated the development of CMOS-compatible, III-V-based-technology research and development worldwide.” 

View the original article here