Artificial neural networks offer an attractive paradigm for the design of behavior and control systems in robots and autonomous agents for a variety of reasons, including: ability to adapt and learn, potential for resistance to noise, faults and component failures, potential for real-time performance in dynamic environments (through massive parallelism and suitable hardware realization) etc. However, designing a good neurocontroller for a given robotic application is an instance of a difficult multi-criterion optimization problem, requiring complicated tradeoffs among different, often competing measures of the network, like performance, cost, complexity etc., which is further compounded by competing objectives in the realization of behavior (e.g., move quickly versus avoid obstacles). Evolutionary Algorithms (EAs), simulated models of natural evolution, have been shown to be effective in searching several vast, complex, multi-modal, and deceptive search spaces. They are therefore viable candidates to employ in the design of neurocontrollers (Bal-akrishnan & Honavar 1995). Although this synergy of approaches is not new (see (Balakrishnan & Honavar 1995) for a bibliography), this field still offers many exciting avenues of research.
Jan-10-2006, 03:48:33 GMT