A review of deep learning–based action recognition and pose estimation in industrial and robotic applications

Amidst the major transformation initiated by the Fourth Industrial Revolution, also known as Industry 4.0, advancements in robotics and artificial intelligence profoundly influence manufacturing. While the objective has always been to improve productivity and efficiency, human intervention remains essential for process validation and action verification. Yet this can be problematic in high-risk, unstructured, or isolated manufacturing environments. This paradox underscores the growing importance of research areas like visual tracking and event understanding that can guide robotic systems toward full autonomy. Advanced optical sensor technologies, including RGB and RGB-D cameras, enable AI-based models to extract valuable insights from frame sequences for highly accurate predictive analytics and complex decision-making. However, a significant gap persists in the literature in applying these technologies to robotic systems themselves. Existing research on Pose Estimation (PE) in manufacturing has largely concentrated on localizing assembly parts, while overlooking the critical need for robotic self-perception through joint estimation. Similarly, Action Recognition studies have predominantly focused on human-centric tasks, leaving the validation of robot-executed actions underexplored. Addressing these deficiencies, this integrated review delivers a systematic analysis of the state of the art in Action Recognition and PE, examining them as foundational technologies for enabling event understanding and visual tracking in robotic systems. The review first provides an overview of optical sensor technologies, then structures the analysis of existing research in Action Recognition around three primary data modalities: Skeletal, Optical, and Sensor. A similar framework is applied to the analysis of PE, focusing on 2D and 3D approaches using both RGB and RGB-D inputs. The review focuses primarily on deep learning techniques, discussing the architectures, datasets, and learning paradigms that have driven recent advances, while acknowledging classical computer-vision approaches. The review systematically structures the field and highlights the underexplored area of robot-centric analysis. It concludes by outlining prospective avenues for advancing next-generation autonomous manufacturing systems capable of operating with greater intelligence, safety, and efficiency.

Authors

Institutions

Publication Details

Journal
Robotics and Computer-Integrated Manufacturing
Published
2026-11-14
DOI
https://doi.org/10.1016/j.rcim.2026.103465
Primary Topic
Human Pose and Action Recognition
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A review of deep learning–based action recognition and pose estimation in industrial and robotic applications

Ramy Harik, Thorsten Wuest, Jad Samaha, Ibrahim Yousif
Robotics and Computer-Integrated Manufacturing
Human Pose and Action Recognition
article

A review of deep learning–based action recognition and pose estimation in industrial and robotic applications

Ramy Harik, Thorsten Wuest, Jad Samaha, Ibrahim Yousif
article en

Abstract

Amidst the major transformation initiated by the Fourth Industrial Revolution, also known as Industry 4.0, advancements in robotics and artificial intelligence profoundly influence manufacturing. While the objective has always been to improve productivity and efficiency, human intervention remains essential for process validation and action verification. Yet this can be problematic in high-risk, unstructured, or isolated manufacturing environments. This paradox underscores the growing importance of research areas like visual tracking and event understanding that can guide robotic systems toward full autonomy. Advanced optical sensor technologies, including RGB and RGB-D cameras, enable AI-based models to extract valuable insights from frame sequences for highly accurate predictive analytics and complex decision-making. However, a significant gap persists in the literature in applying these technologies to robotic systems themselves. Existing research on Pose Estimation (PE) in manufacturing has largely concentrated on localizing assembly parts, while overlooking the critical need for robotic self-perception through joint estimation. Similarly, Action Recognition studies have predominantly focused on human-centric tasks, leaving the validation of robot-executed actions underexplored. Addressing these deficiencies, this integrated review delivers a systematic analysis of the state of the art in Action Recognition and PE, examining them as foundational technologies for enabling event understanding and visual tracking in robotic systems. The review first provides an overview of optical sensor technologies, then structures the analysis of existing research in Action Recognition around three primary data modalities: Skeletal, Optical, and Sensor. A similar framework is applied to the analysis of PE, focusing on 2D and 3D approaches using both RGB and RGB-D inputs. The review focuses primarily on deep learning techniques, discussing the architectures, datasets, and learning paradigms that have driven recent advances, while acknowledging classical computer-vision approaches. The review systematically structures the field and highlights the underexplored area of robot-centric analysis. It concludes by outlining prospective avenues for advancing next-generation autonomous manufacturing systems capable of operating with greater intelligence, safety, and efficiency.

Robotics and Computer-Integrated ManufacturingVol. 104
University of South Carolina (US), Clemson University (US)
Human Pose and Action Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon