The task of trajectory processing is one of the fundamental tasks for the implementation of unmanned vehicles. Radio monitoring systems provide information support for transport objects based on obtaining data on their coordinates. The operating conditions of such systems are determined by the presence of a complex electronic environment, chaotic pulses and noise interference affecting the sensors of primary measurement information. This is reflected in the ambiguous nature of signal and information processing at all stages, which indicates the urgent need to improve information processing algorithms. The solution to this problem is traditionally based on the construction of shaping filters, particle filters and neural networks. Since each of the approaches has advantages and disadvantages, the urgent task is to study the synergy between the neural network approach, or “black box”, and classical algorithms based on physical patterns – the “white box”, which leads to the need to study a hybrid model of the “gray box”. As a result of the analysis of various options for combining the Kalman filter with a neural network, a consecutive connection architecture was chosen, since in trajectory processing tasks the Kalman filter gives good but imperfect estimates requiring nonlinear refinement. In this paper, we consider the problem of synthesizing hybrid algorithms for information processing based on the integrated use of an adaptive Kalman filter and neural networks.