Draft: June 24, 1998.
Robotics can be a hobby, a science fiction genre, a scientific/engineering discipline, or an industrial technology. As a sometimes controversial subject, it is often misrepresented in the popular media, by advocates and opponents. No single definition is going to satisfy such a variety of perspectives and interests.
One very popular and useful definition was first suggested by Mike Brady: the intelligent connection of perception to action. This definition identifies characteristics of machines that enthusiasts are the most enthusiastic about, and it identifies the three processes of greatest interest. Some drawbacks are that "intelligence" is itself difficult to define, and it encourages snobbery. (That's not a robot, it isn't intelligent.)
Another idea that is deeply embedded in robotics, although usually not offered as a definition, is the idea of autonomous general-purpose machines. Most machines seem to be either autonomous or general-purpose, but not both. For example, a dishwasher works all by itself, so it is autonomous, but it really only does one thing: wash dishes. (Never mind the recipes for poaching fish in aluminum foil in the dishwasher.) On the other hand we have machines like a power drill, which can drill, drive screws, strip paint, stir paint, and a number of other things. It has many uses, but it is not autonomous. The human user is providing the generality. Now consider a programmable mobile platform, with some simple bump sensors, light detectors, and other simple sensors. In principle it could do a number of things: deliver mail, patrol a building looking for intruders, spy on the employees, pick up trash, and so on. A single robot might be capable of doing all of these things, changing its behavior according to the time of day or other changing conditions.
As a matter of fact, mobile robots have been programmed to do all of the tasks mentioned above, but each such task is a challenging engineering problem. As we shall see, turning masses of sensory data into useful information is difficult, and choosing effective actions is equally difficult. Most robots are not really capable of doing very much, at least not compared to animals. On the other hand robots certainly are more interesting than dishwashers and power drills, and they are rapidly progressing to more complex behaviors.
Another definition of robotics, perhaps never offered before, is mind-body engineering. Robotics has interesting intellectual roots in a classic philosophical problem: the mind-body problem. The question is how mental events, such as thoughts, ideas, desires, can be connected to physical events, such as bullets and love letters. The more we know about the physical world, the more difficult the question becomes. It would seem that events in the physical world can be explained entirely in terms of forces, voltages and currents, and other physical stuff. You can trace the processes that led to a flying bullet, without ever referring to hatred, greed, jealousy, or any of that mental stuff. Nonetheless, everybody seems to agree that hatred really does sometimes lead to flying bullets. What is the nature of that connection?
The mind-body problem doesn't seem to trouble roboticists very much. After all, when we program we think in terms of computational processes that are analogous to mental processes. For example, a robot might infer the location of an obstacle from a sensor and change its path to avoid it. We think of these processes in the same terms that we think of mental processes: the robot saw something green, decided it might be the file cabinet, and didn't want to hit it. Somehow a desire turned into a physical action. It is the mind-body problem again, but in a new context where we have some hope of understanding the connection. The desires, beliefs, etc., are implemented by encoding them as minute electrical charges in RAM, which can be decoded and amplified to spin electrical motors. The key words seem to be "implemented" and "encoding", which can still mystify, but for some people the problem is essentially solved. Many people, especially roboticists and Artificial Intelligence researchers, would say that the brain is a computer, and the mind is a computational process. Once that principle is accepted, it can even be applied to much simpler systems. In a famous debate, in response to the statement that a thermostat does not have ideas, John McCarthy replied that a thermostat can have either of two ideas: it's too cold in here, or it's too hot in here. How intriguing that a deep understanding of thermostat design principles might provide a solution of the mind-body problem.
One great thing about robotics is that you get to trample the boundaries between Electrical Engineering, Mechanical Engineering, and Computer Science. Effective design of robots depends on ideas and techniques developed in all of these disciplines. This course draws from each discipline. A fundamental premise of this course, and of Andrew's Leap from which this course springs, is that the fundamental ideas can be made accessible to high school students. As an example, Systems Theory and Dynamic Systems are traditionally taught in the third or fourth year of undergraduate studies, and require preparation in linear algebra and differential equations. However, some of the key ideas, such as signals and systems, state transfer functions, and Markov processes, can be introduced much earlier, and provide a framework for the design and programming of robots.
Despite the discussion of philosophy, principles, and theories, the main focus of the course is very practical: robotic technology and techniques for designing and programming robots. In order to solve a chicken-and-egg problem in the design of the curriculum, we start you off with a very simple robot running a very simple program. As the course progresses you will learn the design issues for each different component of the robot, so that by the end of the course the robot will have evolved to a system of your own design.