Performance Evaluation of Deep Deterministic Policy Gradient and Deep Q-Network Algorithms in Quadrotor Stability Control under Stochastic Disturbances

Toloei, Alireza; Mousavi, SayedMostafa

doi:10.22034/jast.2026.552476.1239

Performance Evaluation of Deep Deterministic Policy Gradient and Deep Q-Network Algorithms in Quadrotor Stability Control under Stochastic Disturbances

Articles in Press

Document Type : Original Article

Authors

Alireza Toloei

sayedMostafa Mousavi

Shahid Beheshti University, Tehran, Evin, Shahid Shahriari Square

10.22034/jast.2026.552476.1239

Abstract

Stability control of a quadrotor in hovering mode under random disturbances is a fundamental challenge in aerial robotics due to its highly nonlinear dynamics. In this paper, to enhance the robustness of the system against disturbances, two deep reinforcement learning algorithms — Deep Deterministic Policy Gradient (DDPG) and Deep Q-Network (DQN) — are employed to design intelligent controllers. In the proposed approach, the quadrotor dynamic model is implemented in the MATLAB simulation environment, and the performance of both intelligent agents in maintaining stability is compared considering a discrete action space. The main innovation of this study lies in analyzing the effect of action-space discretization on the performance of the DDPG algorithm and its comparison with DQN under random disturbances, a topic that has been rarely addressed in previous research. Simulation results show that both algorithms can maintain system stability; however, DDPG achieves better performance in terms of angular error reduction and smoother control response compared to DQN. The findings demonstrate the effectiveness of the proposed approach in improving the stability and adaptability of quadrotors under uncertain and disturbed conditions.

Keywords

Reinforcement learning

quadrotor

actor-critic

Deep Deterministic Policy Gradient

Deep Q-Network

Subjects