O.V. Nepomnyashchy1, S.Y. Pichkovskaya2, M.K. Tikhonov3

1-3 Institute of Space and Information Technologies, Siberian Federal University (Krasnoyarsk, Russia)

1 ONepomnuashy@sfu-kras.ru, 2 SLipunova@sfu-kras.ru, 3 samualgame@gmail.com

The problem of reducing the computational requirements for the control hardware of unmanned vehicles while simultaneously improving the quality of decision-making by the control module is considered. Well-known control models that include the use of machine learning technologies for analyzing environmental data are reviewed. The goal is to create a method for intelligent control and path planning for ground-based, unmanned vehicles based on the principle of integrating deep learning technologies and a heuristic algorithm, as well as new data processing methods based on multimodal information perception.

An approach is proposed, the key aspects of which are the pre-processing of data by semantic lidar, as well as the use of puring and quantization methods for compression of models. For path planning, it is proposed to use the waypoint method and perform calculations using a convolutional neural network in combination with the A* algorithm. It is shown that this provides the required dynamics of the control device's functioning, including in the modes of route recalculation. The developed neural network models for reinforcement learning algorithms are presented, allowing for the construction and updating of the waypoint map and the creation of routes. The results of the comparative analysis demonstrate increased spatial orientation accuracy compared to known approaches.

The application of the proposed method in the creation of control models for unmanned vehicles provides significant compression with a slight loss of accuracy, which makes the proposed method competitive and suitable for practical implementation on onboard control systems with lower computational costs.

Nepomnyashchy O.V., Pichkovskaya S.Y., Tikhonov M.K. An intelligent control method of ground unmanned vehicles. Nonl inear World. 2026. V. 24. № 1. P. 36–45. DOI: https:// doi.org/10.18127/ j20700970-202601-03 (In Russian)

1. Badue C. et al. Self-drivingcars: A survey Expert systems with applications. 2021. V. 165. P. 113816.
2. Grigorescu S. et al. A survey of deeplearning techniques for autonomous driving Journal of fieldrobotics. 2020. V. 37. № 3. P. 362–386.
3. Schulman J. et al. Proximal policy optimization algorithms arXiv preprint arXiv:1707.06347. 2017.
4. Haarnoja T. et al. Soft actor-critic: Off-policy maximum entropy deep rein forcement learning with astochastic actor International conference on machine learning. PMLR. 2018. С. 1861–1870.
5. Al-Nuaimi M. et al. Hybrid verification technique for decision-making of se lf-drivingvehicles Journal of Sensor and Actuator Networks. 2021. V. 10. № 3. P. 42.
6. Yin J., Fu W. A hybridpath planning algorithm based on simulated annealing particleswarm for the self-driving car 2018 International Computers, Signals and Systems Conference (ICOMSSC). IEEE. 2018. P. 696–700.
7. Shao H. et al. Safety-enhancedautonomous driving using interpretable sensor fusiontransformer Conference on Robot Learning. PMLR. 2023. P. 726–737.
8. Wu P. et al. Trajectory-guidedcontrol prediction for end-to-end aut onomousdriving: A simple yet strongbaseline Advances in Neural Information Processing Systems. 2022. V. 35. P. 6119–6132.
9. Schulman J. et al. High-dimensional continuous control using generalized advantage estimation arXiv preprint arXiv:1506.02438. 2015. 1
10. Zhao X. et al. Fusion of 3D LIDAR and cameradata for objectdetection in autonomous vehicle applications IEEE Sensors Journal. 2020. V. 20. № 9. P. 4901–4913. 1
11. Koppula H. et al. Semantic labeling of 3d pointclouds for indoorscenes Advances in neuralinformation processing systems. 2011. V. 24. 1
12. Sellat Q. et al. Intelligent Semantic Segmentation for Self‐Driving Vehicles Using Deep Learning Computational Intelligence and Neuroscience. 2022. V. 2022. № 1. P. 6390260. 1
13. Engelmann F. et al. Know whatyourneighbors do: 3D semanticsegmentation ofpointclouds Proceedings of the European Conference on Computer Vision (ECCV) Workshops. 2018. P. 0–0. 1
14. Passino K.M., Antsaklis P.J. A metricspace approach to the specification of the heuristic function for the A* algorithm IEEE transactions on systems, man, and cybernetics. 1994. V. 24. № 1. P. 159–166. 1
15. Park D. et al. Robust datapruning under labelnoise via maximizing re-labelingaccuracy Advances in Neural Information Processing Systems. 2023. V. 36. P. 74501-74514. 1
16. Pushkareva M.M., Karandashev Ya.M. Sokrashchenie nejronnoj seti ResNet-32 sprimeneniem strukturno go pruninga inapravlennogo dropauta. Trudy NIISI RAN. 2021. S. 20 (In Russian). 1
17. Chen C. et al. Deep learning on computational‐resource‐limitedplatforms: A survey Mobile Information Systems. 2020. V. 2020. № 1. P. 8454327. 1
18. Jacinto E., Martinez F., Martinez F. Navigation of autonomous vehicles using reinforcement learning withgeneralizedad vantageestimation International Journal of Advanced Computer Science and Applications. 2023. V. 14. № 1. 1
19. Emuna R., Borowsky A., Biess A. Deep reinforcement learning for human-likedriving policies in collision avoidance tasks of self-drivingcars. arXiv 2020 arXiv preprint arXiv:2006.04218.