Researchers are developing smarter robots, computer systems, software.
Researchers at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory (CSAIL) are developing a variety of robotics applications. One effort is the Molecule reconfigurable robot. Consisting of two types of core machines in female (top) and male (middle) configurations, the Molecule robots can combine to form a variety of larger robots to accomplish specific tasks (bottom).
The purpose of artificial intelligence is to create systems that are capable of human-level reasoning. One practical outcome of this research is the capability to make technology invisible to the user. Increasing automation and autonomy are beginning to appear in applications ranging from robots capable of independently navigating and mapping terrain to interfaces that can understand spoken commands to wireless communications systems that automatically configure themselves.
One institution that is actively expanding the horizons of this technology is the Massachusetts Institute of Technology (MIT) Computer Science and Artificial Intelligence Laboratory (CSAIL),
CSAIL’s mission is to research both computation and artificial intelligence. The laboratory is organized into four research areas: systems, which refers to building hardware and software computational systems; language, learning, vision and graphics; physical, biological and social systems; and theory. The largest of MIT’s laboratories, CSAIL has 92 faculty members and more than 450 graduate students, explains Rodney Brooks, CSAIL director.
The laboratory is heavily involved in robotics research, with 12 faculty members devoted to the subject. Brooks notes that it has a team competing in the Defense Advanced Research Project Agency’s (DARPA’s) upcoming Urban Challenge (SIGNAL Magazine, August 2006), which promotes the development of robotic vehicles that can move in autonomous convoys through a busy urban center. CSAIL researchers also have worked on related topics such as simultaneous localization and mapping. This capability allows a robotic vehicle to use its sensors to build a map and determine its location as it moves.
Brooks observes that from its predecessor laboratories, CSAIL inherited a long history of researching and developing walking robots. Its researchers are creating efficient walking algorithms based on how humans move. The laboratory also is working with DARPA on four-legged locomotion for robots. He explains that the best-known walking robots are Japanese-designed machines such as Honda’s Asimo but adds that these machines rely on a walking algorithm that is 20 times less efficient than human locomotion. “Those sorts of bipedal walkers are never going to be effective unless we get—it’s a joke—nuclear fusion in a backpack,” he quips. “They’re just too energy inefficient.”
The laboratory also is working on a variety of underwater robot projects. These efforts involve submerged sensor networks that permit sensors and robots to communicate underwater. Related programs focus on how data is collected and disseminated. One project features a small robot moving from sensor to sensor gathering and sharing information as a “data mule.”
Another research area is reconfigurable robotics. This work consists of small individual machines that can assemble themselves into larger forms. Although this effort is in its early stages, Brooks believes that in the long-term, the capability will permit small robots to be inserted into a restricted area, such as a nuclear reactor vessel, where they then can assemble themselves into a larger robot for a specific task.
In addition, CSAIL research focuses on autonomous robotics. Brooks notes that most robots deployed today are essentially navigation machines. He explains that robots still have difficulty identifying and autonomously manipulating objects. For example, a bomb disposal robot in
Brooks sees autonomy as a growing trend in robotics. Whether they are in space or on the battlefield, robots will sometimes be cut off from communications, and they will have to operate on their own. He notes that even home-cleaning robots are autonomous. “Otherwise, what’s the point of having one if you have to control it all the time?” he asks.
Another trend is the growing availability of robotic components and sensors. Brooks points out that when he began studying robotics 30 years ago, all of a robot’s parts had to be specially built. “We have a lot more components, so we can put together a robot a lot easier. That’s all well and good. But I think where we are lacking is in other types of sensors to build and use,” he says.
In the next few years, Brooks predicts that a great deal of creativity will occur in sensor development for robots. Another growth area is nanotechnology, which allows researchers to create vast arrays of sensors, such as sensitive touch and pressure sensors, that could not be made before.
Researchers also are studying artificial intelligence applications for human-machine interfaces. CSAIL has devoted a great deal of effort on language research such as speech identification. Brooks says that spoken language is a natural interface for humans and machines. The laboratory’s speech research concentrates on dialogue that will help the system understand the context of human instructions. The program’s objective is to create a simple and intuitive interaction between people and machines, unlike the confusing systems currently in operation. Brooks observes that many people are frustrated when they try to navigate a telephone-based voice identification system. “You’re a captive of some horrible menu that you can’t see. In our dialogue systems, we try to make it where people always have the initiative,” he says.
Such a dialogue-based system would allow people to correct themselves or to back up to a previous subject, as they would in a normal conversation. Besides English-based systems, CSAIL researchers are exploring the possibilities of multilingual systems in Japanese, Chinese, Spanish and Arabic.
CSAIL scientists are examining methods for machines to extract semantic content from Web pages. Brooks notes that when a person searches for information with an Internet search engine such as Google, the tool generates a list of Web pages that users must search through for specific content. “That works pretty well on the computer because Google is very good at giving you a few good choices. But it isn’t very good when you’re driving. If you want to know something, you’d like to be able to say the thing and have the right information spoken back to you—not read from a Web page,” he says.
Brooks explains that a system designed to answer user questions simply is very useful, especially in military situations. “You want that interface to be about the appropriate information at the appropriate time, so people don’t have to put their heads down to look at a screen,” he shares.
Such systems will not only recognize speech but also understand the information that is being retrieved from Web pages. The artificial intelligence applications would be able to determine what information the user wanted, extract it from multiple Web pages and dynamically synthesize it.
Semantic understanding is a related area of interest to CSAIL researchers. This field involves linguistics and statistical mining methods that will eventually enable systems to be trained to search for specific data based on the user’s needs. Ongoing theoretical work is underway on statistical machine learning, which applies to the processing of speech, text and information. Another application related to this research is machine vision. “It would be nice if our robots, or even our laptops, knew what we were doing,” Brooks observes.
For uses such as machine vision and robot locomotion, Brooks relates that many systems, including computer processing and software, are finally mature. He explains that this has enabled CSAIL scientists to make progress with machine vision. “We’re no longer able to use the excuse that we don’t have enough computer power. That’s not the limiting factor,” he says.
Growing computer power allows scientists to use brute force methods to overcome obstacles in robot navigation and locomotion. Smaller and more powerful processors enable interconnected sensors and systems to maintain a robot’s balance and navigation. This numerical capability previously required the use of large computers that could not be installed in robot platforms.
Another example of a high-power computational application is some robots’ ability to track moving objects and identify individual human faces. “But we can’t do what a two-year-old can do,” he states, elaborating that a child can come into an office where he or she has never been before, point to a chair and identify it as such even though the child has not seen that particular chair before. “We can’t yet get our robots to identify an object range,” he says.
CSAIL researchers have worked with NASA on robots for space exploration. These machines will precede human personnel to the moon or to Mars to prepare bases. They will assist explorers and maintain facilities when humans are not present. As a subcontractor with an Australian team, CSAIL addressed robotic digging and mining capabilities. Brooks says that burying habitation facilities is necessary to protect the crews from radiation.
The laboratory is involved in other subtasks such as connecting systems on the International Space Station. Brooks describes how recent space shuttle missions to the station have focused on astronauts wiring connectors on the platform. The joint NASA research examines the development of robots to undertake the assembly work before personnel arrive. He emphasizes that CSAIL is not building an entire system for NASA but focusing on subtasks such as cleaning moon dust from a piece of equipment before it is inserted or moving items such as connectors.
CSAIL also is developing a variety of intelligent wireless systems. Brooks notes that the laboratory hosts the World Wide Web Consortium, which establishes international standards for Web use. In addition, CSAIL shares a laboratory with the Finnish wireless telephone and systems manufacturer Nokia.
CSAIL and Nokia are examining a number of technologies, not all of them artificial intelligence based, that will help to introduce lower powered microchips and core processors into cell phones. Work also is underway on methods to integrate personal electronic devices. Brooks explains that people carry many electronic devices, each with its own idiosyncrasies. “We’ve all become system administrators at some level. We’re trying to figure out how to make all of that [work] seamlessly. The goal is for people to just talk to the system and, when there is a screen available, to point to things on the screen and make that whole interaction as smooth as if you had a very intelligent [human] assistant to whom you could just tell what you need,” he says.
The program with Nokia builds on a previous CSAIL project called Oxygen, which sought to develop a seamless wireless environment for users. Brooks explains that the partnership with Nokia already has developed text and voice systems and intelligent agents. The agents manage text and voice messages and servers to maximize efficiency in a network environment. For example, if a user’s cell phone connection is cut off, the agent will reconnect it using a compatible wireless connection.
Brooks is excited about the progress of the research because of the dramatic growth of wireless systems. He notes that more than 950 million cell phones were sold worldwide in 2006. This proliferation means that most of the people in the world now use wireless devices. Brooks adds that the goal of work such as Oxygen and the collaboration with Nokia is to move the technology into the background for the user. “Good technology should be invisible,” he says.
MIT Computer Science and Artificial Intelligence Laboratory (CSAIL): www.csail.mit.edu