Technical Diagnostics and Non-Destructive Testing, 2025, №03

2025 №03 (02)

DOI of Article 10.37434/tdnk2025.03.03

2025 №03 (04)

---

"Tekhnichna Diahnostyka ta Neruinivnyi Kontrol" (Technical Diagnostics and Non-Destructive Testing) #3, 2025, pp.24-31

Analysis of the effectiveness of reinforcement learning algorithms for increasing the mobile robots autonomy

D.V. Petrenko, A.G. Protasov

National Technical University of Ukraine «Igor Sikorsky Kyiv Polytechnic Institute». 37 Beresteysky Ave., 03056, Kyiv, Ukraine. E-mail: petrenkod555@gmail.com, a.g.protasov@gmail.com

The paper is devoted to the problem of increasing the autonomy of mobile robots, which are widely used in various spheres of human activity today. Improving the means of controlling the movement of robots in real conditions, through the introduction of intelligent control systems, will allow them to adapt to changes in the environment, adequately respond to unforeseen situations and more effectively interact with other participants in the technological process. The intelligent system of controlling the movement of a mobile robot combines both hardware and software components. The software components of robot movement control systems include machine learning methods, which are based on methods of constructing algorithms capable of learning. The paper considers the most popular machine learning algorithms with reinforcement (Reinforcement Learning, RL), which are used in intelligent control systems. In this method, the main components are the agent and the environment. The environment is a dynamic world in which the agent operates and with which it constantly interacts. RL machine learning algorithms are conventionally divided into two groups - algorithms that use a model and algorithms without a model. From the results of the analysis it is obvious that to increase the autonomy of mobile robot movement in complex dynamic conditions, it is necessary to apply hybrid approaches that combine model-free learning, as in the PPO, SAC or TD3 algorithms, with model components, as in the PlaNet or MuZero algorithms. Also, an effective strategy can be the automatic adaptation of hyperparameters during training, for example, the entropy coefficient in the SAC algorithm or the policy constraint coefficient in the PPO algorithm, which provides increased resistance to changes in the environment and the observation state, reducing the need for a large number of interactions with the environment, and flexibility of adaptation to new tasks or changes in target behavior. 8 Ref., 1 Tabl., 2 Fig.
Keywords: machine learning, learning algorithms, mobile robots, control systems, robot autonomy

Received in revised form: 23.05.25
Received in revised form: 19.06.25
Accepted: 01.09.25

References

1. Petrenko, D.V., Protasov, A.G. (2024) Overview of modern technologies for increasing the autonomy of mobile wheeled robot. Vcheni Zapysky TNU, Seriya: Texnichni nauky, 35(74), 2, 122-128. [in Ukrainian]. https://doi.org/10.32782/2663-5941/2024.2/17
2. Akalin N., Loutfi A. (2021) Reinforcement Learning Approaches in Social Robotics. Sensors, 21(4), 1292. https://doi.org/10.3390/s21041292
3. Liu, Y. et al. (2023) Mobile robot path planning based on kinematically constrained a-star algorithm and DWA fusion. Algorithm Mathematics, 11(21), 4552. https://doi.org/10.3390/math11214552
4. Faseeh, M. et al. (2024) Deep Learning assisted real-time object recognition and depth estimation for enhancing emergency response in adaptive environment. Results in Engineering, 24, 103482. https://doi.org/10.1016/j.rineng.2024.103482
5. Zhang, T., Mo, H. (2021) Reinforcement learning for robot research: A comprehensive review and open issues. International J. of Advanced Robotic Systems, 18(3). https://doi.org/10.1177/17298814211007305
6. Rybczak, M., Popowniak, N., Lazarowska, A. (2024) A survey of machine learning approaches for mobile robot control. Robotics, 13(1), 12. https://doi.org/10.3390/robotics13010012
7. Lee, M.-F.R., Yusuf, S.H. (2022) Mobile robot navigation using deep reinforcement learning. Processes, 10(12), 2748. https://doi.org/10.3390/pr10122748
8. Yang, L., Bi, J., Yuan, H. (2022) Dynamic path planning for mobile robots with deep reinforcement learning. IFAC-PapersOnLine, 55(11), 19-24. https://doi.org/10.1016/j.ifacol.2022.08.042

---

Advertising in this issue:

---