Fast Execution of Black-Box Algorithms Through a Piece-Wise Linear Interpolation Technique

Intricate engineering problems commonly make use of complex algorithms whose implementation requires high-end digital devices which are expensive, bulky, and computationally demanding. More often than not, the greater the expected outcomes are, the higher the trade-offs will be between hardware capabilities and the algorithm complexity, which, in the case of small embedded systems, tend to favor the algorithms’ simplification. Hence, an implementation methodology that enables the usage of complex algorithms in restricted hardware is highly desirable. Thereby, this work proposes a piece-wise, n-dimensional interpolation technique to execute a given algorithm in a black-box fashion, i.e., disregarding its conceptual or computational technicalities and building a numerical replica, thus trading processing burden for memory usage. This approach is tested for Artificial Neural Networks and Fuzzy Logic Control (FLC), commonly simplified for attaining implementation, and compared against standardized tools. Similarly, the implementation of an FLC over a LEGO MINDSTORMS\(^{\texttt {TM}}\) robot is achieved in real-time by the proposed technique. The proposed method has shown to conclusively outperform standardized platforms in terms of execution time and, in many cases, memory usage.