YSU's ACM

Experimental AI Powers Robot Army

The U.S. Air Force is working to develop robots with navigational abilities well beyond those required to complete the DARPA Grand Challenge. Whereas those robots merely had to steer themselves over miles of desert terrain, intelligent agents for the military would have to be able to autonomously navigate into underground bunkers, map unfamiliar sites in three dimensions, and determine what is inside those sites without being detected. With those goals in mind, the Air Force Research Laboratory (AFRL) is looking well beyond the capabilities of any existing system, and staking its hopes on developing new software that would enable the robots to learn, walk, and interact in a more sophisticated way than ever before. The software is based on the concept of developing new ideas building on existing knowledge, and similar applications have already written music and designed soft drinks. The software is a form of neural network with two identifying features. One is the noise that is introduced into the network to jumble existing ideas into new forms; the other is a filter to compare the novel ideas with existing knowledge and discard what is deemed unsuitable. Self-learning and adaptability will be central to the success of the software. The research is based on Stephen Thaler's Creativity Machine, which excels at adapting to new physical features and ferreting out the most efficient way to perform a particular task. In his work for AFRL, Thaler has been designing what he calls Creative Robots, which can work together in a swarm to accomplish a common goal. “This approach has less chance of getting stuck than any other” when negotiating unfamiliar obstacles, said AFRL's Lloyd Reshard. Thaler's current project, called CSMARRT (Creative, Self-Learning, Multi-Sensory, Adaptive, Reconfigurable, Robotics Toolbox), is a software package built for designing and modeling virtual robots that can control any type of robot hardware and handle locomotion, sensors, and intelligent behavior to execute a mission... more

Algorithms to Calculate Unusual Behavior

Researchers at National ICT Australia (NICTA) are developing sophisticated surveillance applications such as algorithms that monitor "inappropriate behavior" in public places. The Smart Applications for Emergencies (SAFE) team has already proposed a specification for a warning language that can discern and communicate threat levels. The project's goal is to deliver as much information as possible to decision makers on the front lines of an incident, with much of the work already completed having focused on improving facial-recognition algorithms. "Identifying a particular person is one thing but we are looking at unusual behavior in an open environment," said Chris Scott, research director at NICTA's Queensland laboratory. Scott says that "we are working on algorithms not just to search for a person based on facial recognition but to analyze the level of threat based on their actual behavior." Existing facial-recognition algorithms rely too much on the geometry of the face to make a comparison with the faces that are stored in memory, Scott says, adding that his team is developing algorithms to handle poor lighting and producing images from the side on. The project, which is using data from the 6,000 surveillance cameras on the network of Queensland Transport and Queensland Rail, aims to move away from the dependence on humans looking at monitors... more

Size Matters, But Quantity Can Be Best Way to Supercomputing

Last week, the University of California, Berkeley, and NASA teamed up to launch the Stardust@Home project to harness the power of distributed computing. The project aims to tap the efforts of thousands of users around the world to help locate the few dozen micron-sized stardust particles brought back from space by the Stardust mission. Participants in the Stardust@Home project receive a primer on a Web-based software application that they then use to begin searching the 1.6 million sections in the stardust collector. The trend of harnessing distributed resources to accomplish tasks that otherwise would fall to a single supercomputer is taking hold throughout the technology world. An increasing number of systems on the list of the world's top 500 supercomputers are powered by thousands of the same off-the-shelf commodity processors that can be found in individual PCs. The same principle applies to chip design, where manufacturers such as Intel and AMD are beginning to roll out multicore devices. The speed of the connections between processors is a consideration for both distributed supercomputers and multicore chips. In supercomputing, the value of having thousands or even millions of nodes included in the connection can offset the slowness that comes from the added distance of distributed resources. Grid computing projects such as SETI@Home operate under the same logic. Applications like Google's search rankings and Amazon's ratings system also take advantage of the power of the masses in the online world, using the habits and opinions of their critical mass of users to generate a useful database... more

Wooing the Next-Gen Developer

Platform vendors are racing to support flexible, dynamic languages that facilitate the streamlining of the software development process. The appeal of dynamic languages--so called because programs written in them can alter their structure as they run--is growing along with the complexity of systems, and the challenge of using dynamic languages practically is being met thanks to upgrades in computer speed, chip speed, and memory capacity. In dynamic languages type-checking is performed at run-time, while static languages execute type-checking at compile time; though errors can be caught earlier with static languages, dynamic languages' looser typing scheme yields smaller and simpler code that developers often prefer. The platform vendor who offers optimal dynamic-language support stands to win the hearts and minds of developers, according to observers. The two leading competitors are Sun Microsystems and Microsoft: Sun is attempting to provide stable, secure support on its Java Virtual Machine (JVR) platform, while Microsoft is pursuing the same goal with its Common Language Runtime (CLR) platform. With the ability to use a Java-supported scripting language to develop a robust application, “you get all the benefits of the Java platform as well as the ability to develop enterprise-scalable applications using a scripting language,” said Java developer Bruce Snyder. CLR development leader Jim Hugunin explained that Microsoft is taking a multi-level approach to its own dynamic-language support initiative. “What we're going to try hard to do is, instead of doing a dynamic language specification, provide a dynamic language library and have guidance on how to use it,” he noted... more

Too Much Information: Two applications reveal the key challenges in making context-aware computing a reality

Realizing context-aware computing requires meeting a number of challenges, and IBM research is focused on capturing context and passing it on to humans to determine what course of action to follow. However, IBM's attempt to apply context awareness to communications tools used regularly by IBM employees has not yet lived up to its potential. Their efforts focus on two services: IBM's Grapevine service, which is designed to help a person communicate with another individual using an aggregated and filtered set of data rendered as a business card with real-time information; and the IBM Rendezvous Service, which is supposed to help people convene in small groups and talk on plain old telephones. The Grapevine service demonstrated that the most useful piece of information provided was a person's current or last known physical location, which was applied toward the selection of an appropriate communications method as well as the person to be communicated with. Including the user's computer application activity on the e-card did not pan out as well, because users were often uncomfortable with others knowing their goings-on, while ascertaining a person's situation from studying his or her application activity was complicated by the fact that people use applications for multiple purposes. Lessons taken from the Grapevine experience include the understanding that users will not take any additional action to provide context; users' concerns about the visibility of their context information must be eased simply, powerfully, and intuitively; instant messaging is the preferred means of obtaining real-time context; and a sizable semantic gulf exists between the information detected by low-level sensors and programs and a person's high-level ability and willingness to communicate with another person. The Rendezvous project, meanwhile, uses a conference call proxy for users who can access their laptops during a call, and which visualizes how much time is remaining in the meeting, authenticated users still on the call, the meeting host, the persons speaking, those on mute, participants who have yet to arrive, and recent actions. From these examples, it can be concluded that it is critical to gain experience with implementing real applications and services at a realistic scale... more

Robot backers turn to linux

What the world needs now, according to Japanese research group, is a low-cost programmable robot.

To spur more development of robots at the hobbyist level, Japan's National Institute of Advanced Industrial Science and Technology (AIST) is promoting a humanoid creature named HRP-2m Choromet. One problem with the current robots, AIST says, is that they tend to be little more than remote-controlled devices. Another is that getting beyond that evolutionary stage thends to take a lot of cash. ... Choromet, which bears a striking resemblance to the Transformers character Optimus Prime, comes with programmable software that runs Linux... more