Predicting robotic platform missions using a kernel activation network with an asymmetric kernel

This study considers those processes predicting the functional efficiency of robotic platforms that affect the optimization of their mission planning. Given the growing demand for autonomous mobile systems, a critical task is to ensure high efficiency of their dynamics under different loads, terrains, and speeds, which requires reliable tools for decision-making even before physical launch. To solve the task, a method based on a customized Kernel Activation Network (KAN) was devised and programmatically implemented to predict the functional efficiency of the platform. The results demonstrate a significant increase in accuracy: KAN achieves an MSE of 0.00055727 on synthetic data and 0.00041720 on the experimental sample, while other architectures demonstrate 0.00105989 and higher. The key innovation of KAN is the use of an asymmetric chi-square kernel in parallel with the Gaussian kernel, as well as the integration of input estimates that take into account the triple interaction of factors. This explains the network's ability to effectively capture complex nonlinear dependences between numerous platform parameters (rolling resistance, aerodynamic drag, climbing force, etc.) and environmental conditions. The use of an asymmetric kernel significantly simplifies the network architecture, allowing for high accuracy at lower computational complexity. In practice, the results serve as an additional tool for optimizing mission planning of robotic platforms. This makes it possible to optimize equipment selection, construct strategic logistics routes, and increase the safety and reliability of autonomous systems under actual conditions. The achieved Technology Readiness Level is 4