Simulation vs. Real-time Control; with Applications to Robotics and Neural Networks

Simulations are often used to model real physical systems prior to electrical, mechanical, and computer hardware development. This allows engineers and scientists to experiment with various concepts before committing time and effort into hardware. Simulations can also be run concurrently with real-time systems to build knowledge of the environment that the real-time system is operating in, then provide feedback to the system to optimize its performance. For both of these types of simulations, the simulation must accurately model the real physical system. A comparison of simulations to real-time controlled physical systems is illustrated in this paper using several simple robotic and artificial Neural Network examples. The robotics examples show how real-time control of mobile robots and robotic-arms, and the resultant governing equations and software algorithms can provide several interesting simulation problems to overcome if the simulation is to accurately model the physical system. The Neural Network example demonstrates how computational speed and numerical precision can become an issue when comparing simulations to real-time Neural Network hardware. In general, comparison of simulations to real physical systems often enhances understanding of the underlying governing principles and equations, and results in simulations that accurately model the real world. I