Can Signal Delay Be Functional? Including Delay In Evolved Robot Controllers

Signals travel at finite speeds within the nerves of living organisms, between satellites and the Earth, or in computers and other technological devices. As a result, they incur delays in moving from one point to another, which engineers, roboticists, control-theorists and neuroscientists typically consider as a source of error. Delays mean that signals carry information that is old when it finally arrives, disrupting the feedback loops on which precision control depends. The frequent result is dynamic instability that undermines smooth functioning.
Yet biological organisms working with very slow signals often achieve exquisite control of complex dynamics, even in harsh and uncertain environments. It is known, for example, that the finite speed of signals in certain neurons in octopuses plays an essential role in the producing coordinated activity. Inspired by this fact, LML External Fellow Claire Postlethwaite and colleagues have considered the possibility that signal delay may, in some systems, actually play a positive role. In a recent study, they tried to evolve a simple robot controller to solve a basic stimulus-distinction task, while working with a delayed control signal. Modeling the dynamics with delay-differential equations, they explored the kinds of invariant sets underlying both successful and unsuccessful performances of the robot, and how changes in the nature of the task as well as in the signal delay cause bifurcations in these invariant sets.
The results indicate that the robot controller was generally more successful in achieving good outcomes with a finite signal delay than in the case of zero delay. Although the system dynamics could be quite complex – involving oscillatory or even chaotic behaviour – the controller still often achieved superior results in achieving successful control with delay. The study suggests that delays probably do serve a functional role in control architectures, and that many biological organisms, especially, have evolved to exploit this possibility.
The paper, accepted and soon to appear in the journal Artificial Life, is Egbert M. D., Keane, A., Postlethwaite C. M., Wong, N. Can Signal Delay Be Functional? Including Delay In Evolved Robot Controllers