Abstract
The purpose of research . The purpose of this scientific work is to conduct a comprehensive theoretical and analytical review of modern machine learning algorithms used to solve the problems of dynamic route planning for mobile robots. The main focus is on a comparative assessment of the effectiveness of various learning paradigms – reinforcement learning, teacher-based learning, and hybrid approaches – in a changing and uncertain environment where rapid adaptation, learnability, and algorithm stability are important. Methods . The study is based on an analysis of more than 40 peer-reviewed scientific publications selected from leading international academic databases for the period from 2020 to 2024. A structured methodology was used, including descriptive, comparative, and analytical approaches. The main evaluation criteria were: convergence rate; computational efficiency; generalization ability; noise tolerance; adaptability to real-time and stable behavior in changing conditions. Results . It is shown that tabular algorithms provide basic navigation functionality, but they do not scale for complex tasks. Deep models have a high degree of adaptability and efficiency. Teaching with a teacher demonstrates accuracy in the presence of expert data, but is vulnerable to the accumulation of errors. Hybrid architectures combining graph neural networks and symbolic modeling achieve the best interpretability and stability in an unstable environment. Conclusion . The results obtained form a reliable theoretical basis for the selection and application of autonomous navigation algorithms. The comparative analysis highlights the value of flexible, scalable, and explicable models in intelligent robotics systems of a new generation.