Cognitive Control in the Real World

Many articles have appeared on this site regarding robots that mimic some natural entity. These include robots that are modeled after insects, worms, and tuna.

Since many of these naturally occurring forms have adapted to particular situations over time, this seems like a good way to go. While these systems adapt the shape and movement of natural systems, there is another area in which reference to the natural world is of use. That area is control systems.

Control systems have evolved over time from fixed value controllers to adaptive control systems. A recent article in the Proceedings of the IEEE December 2012, entitled "Cognitive Control" by Haykin, Fatemi, Setoodeh, and Xue, details the methods of cognitive control and gives examples. As with many of the systems modeled on the natural world, they were studied first in areas other than engineering.

For cognitive control that includes neuroscience and psychology, control systems have evolved from open-loop and PID controllers to more adaptive systems. One of the drivers, of course, is the availability of the computing power to realize these systems. Early control systems were implemented in analog devices. I have seen, as a previous generation, spacecraft and simulator controllers that were based on analog technology. While they could be fast, they were not adaptable or easily changed. These weaknesses helped in the push to digital control systems.

As the technology has progressed, providing more computing power and memory, the ability to implement, in a very cost effective way, adaptive and other sophisticated control approaches is now feasible. Adaptive control lets the systems parameters evolve over time, based on actual interaction with the real world. In 1995, I had a car that had an adaptive shifter mechanism. It adjusted the shift points over time based on the driver's style of driving. I never did figure out how to reset it.

With the development of microcontrollers that have extensive processing power and memory, one of the primary requirements of cognitive control can be realized. That is memory. While adaptive control adjusts parameters over time, cognitive control uses memory to implement a reinforcement-learning approach to adapting.

Many microcontrollers today have the memory capabilities to implement this along with signal processing functions implemented in hardware to efficiently implement the learning algorithms. Thus, implementing learning as a part of the control loop is feasible. This allows us to close the information gap, as the article calls it, in a low-power, low-cost controller. This type of capability is becoming possible in areas like the automotive industry yielding much more efficient engines primarily by providing a more sophisticated engine management system.

Related posts: