Article: Adaptive deep reinforcement learning-based error analysis model for strength training Journal: International Journal of Ad Hoc and Ubiquitous Computing (IJAHUC) 2026 Vol.52 No.5 pp.63 - 75 Abstract: To tackle limitations of conventional strength training movement evaluation, such as inadequate movement stage segmentation and insufficient fine-grained error perception, this study proposes ARLE-Net. This framework integrates improved YOLOv8 with CBAM attention mechanism to enhance local movement feature detection, HR-Net for high-resolution human keypoint extraction, and adaptive reinforcement learning for dynamic action segmentation and standardised score prediction. Experiments on Fitness-AQA and ARFit datasets show it outperforms state-of-the-art methods in action segmentation accuracy and score prediction mean square error, with strong stability. Ablation studies confirm component effectiveness, providing support for intelligent fitness guidance and fitting the journal's enhanced ubiquitous computing focus. Inderscience Publishers - linking academia, business and industry through research

Title: Adaptive deep reinforcement learning-based error analysis model for strength training

Authors: Guobao Zhang; Haotian Li; Xurui Liu

Addresses: School of Police Law Enforcement Abilities Training, People's Public Security University of China, Beijing 100038, China ' School of Police Law Enforcement Abilities Training, People's Public Security University of China, Beijing 100038, China ' Hainan International College, Beijing Sport University, Beijing 100084, China

Abstract: To tackle limitations of conventional strength training movement evaluation, such as inadequate movement stage segmentation and insufficient fine-grained error perception, this study proposes ARLE-Net. This framework integrates improved YOLOv8 with CBAM attention mechanism to enhance local movement feature detection, HR-Net for high-resolution human keypoint extraction, and adaptive reinforcement learning for dynamic action segmentation and standardised score prediction. Experiments on Fitness-AQA and ARFit datasets show it outperforms state-of-the-art methods in action segmentation accuracy and score prediction mean square error, with strong stability. Ablation studies confirm component effectiveness, providing support for intelligent fitness guidance and fitting the journal's enhanced ubiquitous computing focus.

Keywords: strength training; pose estimation; reinforcement learning; action quality assessment.

DOI: 10.1504/IJAHUC.2026.154094

International Journal of Ad Hoc and Ubiquitous Computing, 2026 Vol.52 No.5, pp.63 - 75

Received: 17 Oct 2025
Accepted: 22 Dec 2025
Published online: 12 Jun 2026 *

Title: Adaptive deep reinforcement learning-based error analysis model for strength training

Keep up-to-date