Digital twin driven GAT and PPO for semantic fidelity preserving robotic embroidery path optimization of Mosuo costume patterns

Abstract In electrical automation systems, when using robots to process complex vector patterns, a short tool path may interrupt critical contours or compromise structural continuity, while unsuitable motion speeds can increase tracking error and process risk. Therefore, the processing sequence and robot speed need to be optimize d without violating the structural requirements of the pattern. DT-GAT-PPO is a digital twin-based framework that integrates a Graph Attention Network (GAT) with Proximal Policy optimization(PPO). Processing points and their contour connections, feasible transitions, and expert-defined structural relationships are presented in a semantic attributed graph. A multi-head GAT encodes the graph along with the current process state, and a hybrid-action PPO policy selects the next processing node and its corresponding motion velocity. Reward balances semantic-structure preservation and non-productive travel with processing time and drive energy, trajectory deviation, jump operations, and failure risk. Training is performed in a calibrated digital twin, and then domain randomization and residual adaptation are employed to reduce the simulation-to-reality gap. A total of 1,260 digitized Mosuo costume vector patterns from 18 pattern families and physical robotic embroidery experiments were conducted. The calibrated twin had a position root-mean-square error of 0.34 mm and prediction errors no more than 6.0% for processing time, energy consumption and thread tension. DT-GAT-PPO had a semantic-fidelity score of 0.947 and a constraint-satisfaction rate of 96.8% on the simulation test set. Relative to conventional GAT-PPO baseline (a GAT-PPO model without the digital twin, domain randomization and residual adaptation, and without the semantic constraints in the reward), non-productive distance, processing time, and energy consumption decreased by 9.0%, 8.2%, and 8.3%, respectively. The mean position error and completion rate of the physical experiment were 0.41 mm and 97.3%. Based on the above results, joint path-and-velocity optimizationcan be used for robotic embroidery to help preserve contour continuity and maintain the expert-defined structural relationship.

Authors

Institutions

Publication Details

Journal
Journal of Engineering and Applied Science
Published
2026-11-13
DOI
https://doi.org/10.1186/s44147-026-01317-8
Primary Topic
Digital Transformation in Industry
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Digital twin driven GAT and PPO for semantic fidelity preserving robotic embroidery path optimization of Mosuo costume patterns

Zhenghan Chen, Yuchao Zhang, Jing Jin
3 links
Journal of Engineering and Applied Science
Digital Transformation in Industry
article

Digital twin driven GAT and PPO for semantic fidelity preserving robotic embroidery path optimization of Mosuo costume patterns

Zhenghan Chen, Yuchao Zhang, Jing Jin
article en

Abstract

Abstract In electrical automation systems, when using robots to process complex vector patterns, a short tool path may interrupt critical contours or compromise structural continuity, while unsuitable motion speeds can increase tracking error and process risk. Therefore, the processing sequence and robot speed need to be optimize d without violating the structural requirements of the pattern. DT-GAT-PPO is a digital twin-based framework that integrates a Graph Attention Network (GAT) with Proximal Policy optimization(PPO). Processing points and their contour connections, feasible transitions, and expert-defined structural relationships are presented in a semantic attributed graph. A multi-head GAT encodes the graph along with the current process state, and a hybrid-action PPO policy selects the next processing node and its corresponding motion velocity. Reward balances semantic-structure preservation and non-productive travel with processing time and drive energy, trajectory deviation, jump operations, and failure risk. Training is performed in a calibrated digital twin, and then domain randomization and residual adaptation are employed to reduce the simulation-to-reality gap. A total of 1,260 digitized Mosuo costume vector patterns from 18 pattern families and physical robotic embroidery experiments were conducted. The calibrated twin had a position root-mean-square error of 0.34 mm and prediction errors no more than 6.0% for processing time, energy consumption and thread tension. DT-GAT-PPO had a semantic-fidelity score of 0.947 and a constraint-satisfaction rate of 96.8% on the simulation test set. Relative to conventional GAT-PPO baseline (a GAT-PPO model without the digital twin, domain randomization and residual adaptation, and without the semantic constraints in the reward), non-productive distance, processing time, and energy consumption decreased by 9.0%, 8.2%, and 8.3%, respectively. The mean position error and completion rate of the physical experiment were 0.41 mm and 97.3%. Based on the above results, joint path-and-velocity optimizationcan be used for robotic embroidery to help preserve contour continuity and maintain the expert-defined structural relationship.

Journal of Engineering and Applied ScienceVol. 73(1)
Guangdong University of Science and Technology (CN)
Industry, innovation and infrastructure
Digital Transformation in Industry
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