Slashdot Mirror
Immobile Robots
Roland Piquepaille writes "Wade Roush wrote a long and well-documented article for the Technology Review about this new concept, the immobot, short for "immobile robot." He gives different industrial examples, from NASA to the water utility in Porto Alegre, and from Toyota cars to some new Xerox photocopiers. And he looks at the programming model behind the immobots. No "heuristic" programs here, but model-based programs instead. Check this column for details." The original article has more information.
Tossing the probe in with the rest of these "immobots" is a bad example, since it is mobile after all. The control system turned the robot, or in this case, the probe.
The problem is that the word "robot" is being misused (well, redefined perhaps) here. Ever since it was first applied in the play R.U.R., "robot" has always indicated a mobile machine, usually with some fashion of humanoid appendages (arms, head, sometimes legs). Primary parts of the goal of robotics are path planning (how to move an arm to pick up an object, or how to mow the lawn without hitting the puppy running back and forth), and environmental awareness (being aware that the puppy or the object is there to be avoided or picked up in the first place.)
However, the examples in the articles don't have direct contact to physical appendages, rather, they have a model of the appendages internally to work on. While not making for an impressive sight, these have the advantage of allowing the designer to break free of the anthropomorphism all too common in robotics. Why does a robot need arms and a head? The original article talks about controlling a water treatment system where these appendages are rivers and treatment tanks. Unlike traditional robotics, the goal isn't physical path planning, rather, its planning a course of action that solves a problem. The larger such systems become, the more complex their model will become, which will require greater environmental awareness than visual object identification.
The development in this field will surely help the "real" robots, as advancements in developing these models will continue until the robot is capable of extending these models itself, which will allow your lawnmower to decide that its also important to avoid hitting the neighbor's cat that your puppy has been chasing around your lawn, even though you forgot to tell it about the cat.
If I have been able to see further than others, it is because I bought a pair of binoculars.
It is a robot in the sense that it recieves data through its sensors and then interacts with the world via effectors. It just happens that for an immobile robot, none of the effectors are wheels or other locomotion devices.
Control theory is more of a traditional engineering discipline, studied by electrical, mechanical, and industrial engineers. It takes a strong math background: calculus, linear and nonlinear equations, tensors, Laplace and Z transforms. There isn't yet "Control Systems for Dummies", although some friends of mine are trying to change that by writing a controls curriculum, accompanied by a parts kit, for bright high-school students.
The path to low-level AI (moving around, not bumping into stuff, not falling down) may lie in the region between model-based control and machine learning. That region is now open for business, due to cheap compute power. Control systems used to be powered by computers with well under 1 MIPS; most of them still are. With cheap gigaflops available, approaches that were once far out of reach can be used. Real-time stereo vision finally works, and is about to get cheap. Stability enhancement systems for cars are quite impressive today. Self-balancing machines, from the Segway to the Asimo, are showing up as products.
Mobile robots, which have been sluggish machines for decades, typically have rather low-performance control systems. The DARPA LA to Las Vegas robot race may change that.