Research on Black Boxes

Shuichi Adachi (Professor, Department of Applied Physics and Physico-Informatics)

In the 1990s, the term "fuzzy control" became popular, leading to the sale of home appliances like fuzzy logic washing machines. Even when not explicitly named in such products, the concept of "control" is widely used all around us. For example, riding a bicycle without falling over is a form of control, as is an air conditioner maintaining a comfortable room environment through temperature and humidity control. Control is also necessary for a robot to walk on two legs (Photo 1). Furthermore, the cars we drive are filled with electronic control technologies, beginning with engine control. Indeed, modern automobiles would be unable to function without the power of control.

In our laboratory, we research "control engineering," which provides a theoretical system for the concept of control that we use casually in our daily lives. While conventional engineering fields have clearly defined subjects—for instance, electrical engineering deals with electricity—control engineering is applied to a wide range of subjects. These applications range from the small scale, such as hard disk head position control requiring nanometer precision, to the large scale, like attitude control for space stations, vibration control in skyscrapers, pressure and flow control in chemical systems such as oil refineries, and more recently, even system biology and quantum systems. Thus, control theory can be applied to any control object that possesses dynamics (i.e., temporal movement). For example, the Kalman filter, one of the greatest achievements in control theory, is actively applied in diverse fields, including automotive navigation systems and computer vision.

To perform control, a mathematical model of the control object is necessary. The design method for control systems based on such models is called "model-based control," and this methodology is being actively put into practice, particularly in the automotive industry. While first-principle modeling—which builds models based on the physical laws of the control object (e.g., Newton's equations of motion)—is the mainstream approach, "system identification" has recently gained attention. This method treats the object as a black box and models it primarily through statistical techniques using its input/output data. It is an attempt to investigate the inner workings (the cause) of a black box from its input/output data (the result), and is classified mathematically as a difficult type of problem known as an inverse problem. However, this research has a fascinating quality, much like deducing the culprit in a mystery novel, and our laboratory is vigorously researching modeling for control based on system identification.

Recent technological innovation has been tremendous, and the world is now filled with black boxes. Even elementary school students can operate televisions and mobile phones without understanding their mechanisms (their inner workings). I believe this is a sign of technological progress. However, if everyone in Japan becomes merely a user, the nation is doomed. In particular, I feel there is a growing number of people who can understand concrete, externally visible things but struggle to comprehend the abstract concepts within. I cannot help but think this is related to the fact that modern Japanese society, with its lack of breathing room, constantly demands immediate, visible results. I hope to see many more students and engineers who are interested in what goes on inside these black boxes.

Photo 1: Control is also incorporated into bipedal robots.

Photo 2: Professor Kalman (right) and the author (left) (in Cambridge, September 2006).

Figure 1: In the Department of Applied Physics and Physico-Informatics, we research both physics and informatics, and modeling and control serve as a bridge between "physics and informatics."