Title: Adaptive obstacle avoidance technology for mobile robots in indoor environments

Authors: Xin Zhang; Wei Li; Jinbao Sun; Kunjian Liu; Zhe Zhang

Addresses: School of Information Science and Engineering, Shenyang University of Technology, Shenyang, 110870, China ' School of Information Science and Engineering, Shenyang University of Technology, Shenyang, 110870, China ' Neusoft Group, Shenyang, 110179, China ' School of Information Science and Engineering, Shenyang University of Technology, Shenyang, 110870, China ' School of Information Science and Engineering, Shenyang University of Technology, Shenyang, 110870, China

Abstract: This study presents a cost-effective and efficient path obstacle avoidance method tailored for complex indoor environments. It integrates a lightweight MobileNet v4 for visual processing, leveraging depthwise separable convolutions and a hybrid attention mechanism to enhance image feature extraction and perception. Skip connections and bilinear interpolation further mitigate depth information loss. For path planning, an improved deep Q-network (DQN) with linear and angular velocity constraints is employed, and the reward function dynamically balances obstacle avoidance and path following. The proposed model achieves a 98.3% success rate, a Q-value of 4.6, an average reward of 94, and a runtime of 3.1 seconds at 0.75 m/s and π/5 rad/s. Results confirm significant improvements in navigation accuracy, stability, and computational efficiency, offering a practical solution for autonomous indoor robot navigation.

Keywords: indoor; path; obstacle avoidance; deep reinforcement learning; image.

DOI: 10.1504/IJMMS.2025.148936

International Journal of Mechatronics and Manufacturing Systems, 2025 Vol.18 No.1, pp.21 - 46

Received: 28 Apr 2025
Accepted: 05 Jun 2025
Published online: 04 Oct 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article