×

You are using an outdated browser Internet Explorer. It does not support some functions of the site.

Recommend that you install one of the following browsers: Firefox, Opera or Chrome.

Contacts:

+7 961 270-60-01
ivdon3@bk.ru

Synthesis of an adaptive control system based on deep reinforcement learning for robotic systems

Abstract

Synthesis of an adaptive control system based on deep reinforcement learning for robotic systems

Safin M.A.

Incoming article date: 03.01.2026

The article discusses the problem of synthesizing adaptive control systems for robotic complexes operating under conditions of uncertainty and variable external influences. A methodology for building control systems based on deep reinforcement learning algorithms integrated with traditional management methods is proposed. The scientific novelty lies in the development of a hybrid architecture combining a deterministic dynamic robot model that provides basic stability and an adaptive neural network module based on Deep Deterministic Gradient Policy  that compensates for unaccounted-for disturbances and parametric uncertainties. The practical significance is confirmed by the results of computational experiments on a manipulator model with six degrees of freedom, where the proposed system showed a 67% reduction in positioning error under variable loads compared to the optimal PID controller, as well as the ability to adapt online when the weight of the load changes. The implemented approach opens up prospects for the creation of autonomous robotic systems capable of efficiently performing tasks in unstructured environments.

Keywords: deep reinforcement learning, adaptive control, robotic complexes, hybrid control systems, dynamic modeling, neural network controllers