Журнал «Современная Наука»

Russian (CIS)English (United Kingdom)
 MOSCOW +7(495)-142-86-81

ROBOTIC LEARNING BASED ON Q-LEARNING AND ANN USING DEMONSTRATIONS (ADQN)

Tianci Gao   (Ph.D. Candidate Bauman Moscow State Technical University )

Bo Yang   (Ph.D. Candidate Bauman Moscow State Technical University )

Shengren Rao   (Ph.D. Candidate Bauman Moscow State Technical University )

In the field of robotic systems training, the classical approach based on pure Q-Learning requires a large volume of trial-and-error interactions. This not only reduces efficiency but can also be unsafe when working with real robots. In this paper, we propose an ADQN (Augmented Deep Q-Network) method that combines Q-Learning, an artificial neural network (ANN), and demonstration data. In the first phase, the Q-network is trained offline on expert trajectories. Then, during the online phase, TD updates and Margin-based supervision on demonstration actions are used simultaneously. This approach accelerates the convergence of the algorithm and increases overall success rates. We compare ADQN with two baseline methods: (1) pure DQN (no demonstrations) and (2) pure imitation (ANN). Experiments in the MATLAB/Simulink environment and on a real Kinova Gen3 robot show that ADQN achieves higher performance and reaches target results faster. We also analyze the impact of prioritized replay and various modules of the algorithm. The results confirm that the proposed approach effectively combines the advantages of reinforcement learning and demonstration-based training.

Keywords:reinforcement learning, Q-Learning, artificial neural networks, learning from demonstrations, robotic manipulator.
 

Read the full article …

Citation link:
Tianci G. , Bo Y. , Shengren R. ROBOTIC LEARNING BASED ON Q-LEARNING AND ANN USING DEMONSTRATIONS (ADQN) // Современная наука: актуальные проблемы теории и практики. Серия: Естественные и Технические Науки. -2025. -№04. -С. 48-54 DOI 10.37882/2223-2966.2025.04.06
LEGAL INFORMATION:
Reproduction of materials is permitted only for non-commercial purposes with reference to the original publication. Protected by the laws of the Russian Federation. Any violations of the law are prosecuted.
© ООО "Научные технологии"

Terms Publications

Issues of the journals (Archive)

ECONOMICS AND LAW
HUMANITIES
NATURAL AND TECHNICAL SCIENCES
COGNITION

Серия - Natural and Technical Sciences