Developmental Robots
|
Developmental Robots |
 |
Researchers are struggling to make intelligent machines that can learn on their own. Do we really need intelligent computers that may become superior to biological systems?
|
|
|
|
|
|
|
|
Elsewhere on the Web
|
• Reinforcement Learning
• Artificial Brains
• Like a child, "smart" robot learns gradually
|
|
|
| |
|
According to Stephen Hawking, a highly acclaimed British scientist, genetic engineering should be used to prevent human intelligence being overtaken by that of computers. The famous Moore's law states that computers double their performance every 18 months. Hawking claims that targeted genetic changes could increase the complexity of our DNA and literally "improve" human beings. In his view, this is the road we should take if we want biological systems to remain superior to electronic ones. Other researchers are also concerned about the future of the humankind. Bill Joy of Sun Microsystems recently wrote an online essay for the Wired magazine, titled "Why The Future Doesn't Need Us", prompting discussion about how far robotics, genetic engineering and nanotechnology might go in supplanting and overpowering humanity. How real is the danger that our computers might develop intelligence and take over the world? Do we really want to build machines that will have intellectual advantages over their creators?
An earlier article gives a brief overview of projects aiming to build intelligent machines, capable of communicating in natural language and acquiring new knowledge. In reality, researchers are still struggling to make intelligent machines that can learn on their own. Artificial Intelligence NV (Ai), an international company split between Boston and Tel Aviv, claims a breakthrough with a computer that is learning to talk like a toddler. Contrary to the traditional approach of using a set of hardwired rules and a vocabulary database to approximate human conversation, researchers at Ai follow the Turing's idea of the child machine. Such "developmental" approach becomes especially interesting when applied to hardware: an autonomous robot cannot hold all the details of the environment in which it is going to act. Following this idea, professor Juyang Weng of Michigan State University recently introduced a new kind of robots: machines that can develop their "mental skills" automatically through real-time interactions with the environment.
Professor Weng and his team have already built a prototype developmental robot. SAIL (for Self-organizing Autonomous Incremental Learner) is a human-size robot that wanders the halls of Michigan State's Engineering Building, responding to touch, voices and vision input.
Each of its two eyes is controlled by a fast pan-tilt head. Its torso has 4 pressure sensors to sense push actions and force. It has 28 touch sensors on its arm, neck, head, and bumper to allow human to teach how to act by direct touch. Its drive-base is adapted from a wheelchair, allowing SAIL to operate both indoor and outdoor. The main computer is a high-end dual-processor dual-bus PC workstation with 512MB RAM and an internal 27GB three-drive disk array for real-time sensory information processing, real-time memory recall and update as well as real-time effector controls.
Apart from the hardware configuration, the most interesting aspect of the Sail's architecture is the control algorithm and the philosophy behind it. According to this paradigm, real-world problems are beyond predefined representations and hand programming. Robots should be designed to go through a full, autonomous mental development process motivated by human cognitive and behavioral development, from infancy to adulthood. Essentially, the key to success is to enable robots to autonomously live in our environments and to become "smart" on their own, although some degree of human supervision is always required. In Sail's case, this means that it is allowed to explore the world for itself, accompanied with the human supervisor that reinforces and modifies behavior patterns by pressing "good" or "bad" buttons. In author's own words, "... AI tasks require capabilities which have proved to be too muddy to program effectively. Although a developmental algorithm is by no means simple, the new developmental approach does not require human programmers to understand the domain of tasks nor to predict them. Therefore, this approach does not only drastically reduce the programming burden, but also enables machines to develop capabilities or skills that the programmer does not have or are too muddy to be adequately understood by the programmer. "
More technical details about this project can be found at the main Sail page. You'll even find the code in Visual C++ and detailed information on all aspects of robot programming and control. Bit it is by no means the only learning robot out there. Sony's AIBO also employs a sort of developmental learning, but on the much more primitive level. An interesting list of robot learning Web sites will provide you with an excellent introduction to this field and pointers for the further research.
One of the most brilliant minds in this field, Marvin Minsky, dismissed fears about AI in his famous article for the Scientific American magazine, "Will Robots Inherit the Earth" "Yes, but they will be our children" is the answer: he is more concerned with the fact that there are only around 10 "significant" people in the world who are tackling the "hard" problems of AI. Read more about his views and ideas, including early drafts of his forthcoming book, at his site.
Do you agree?