EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Spatial-Temporal Graph Diffusion Policies for Safe Bimanual Robotic Manipulation

Chaewon Na, Joonho Woo

International Journal of Information and Communication Technology Trends · 01.05.2026 · 6 (1) · 258-268

DOI: 10.66525/ijictt90 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Information and Communication Technology Trends, online erschienen am 01.05.2026. Verfasst von Chaewon Na und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The integration of artificial intelligence into complex robotic systems has significantly advanced the capabilities of autonomous agents in unstructured environments. Bimanual robotic manipulation, a critical domain requiring high degrees of coordination, spatial awareness, and temporal synchronization, remains a formidable challenge due to the exponential growth of the state-action space and the strict requirements for collision avoidance. This paper presents a novel methodological framework that leverages spatial-temporal graph neural networks coupled with diffusion-based policy generation to achieve safe and robust bimanual manipulation. By modeling the dual-arm system, manipulated objects, and environmental obstacles as a dynamically evolving spatial-temporal graph, the proposed architecture captures the intricate physical dependencies and temporal correlations inherent in cooperative tasks. Diffusion models are subsequently employed to represent highly expressive, multi-modal action distributions, allowing the system to generate complex trajectories that traditional deterministic policies struggle to formulate. Furthermore, a hard-constraint safety projection mechanism is embedded directly within the iterative denoising process to guarantee zero-collision trajectories during inference. Comprehensive evaluations across a suite of simulated bimanual tasks, including deformable object manipulation and precision assembly, demonstrate that the proposed framework significantly outperforms existing behavioral cloning and reinforcement learning baselines in both task success rate and safety adherence. The findings highlight the potential of combining graph-based relational inductive biases with the generative power of diffusion models for safety-critical robotic applications

Zitieren

APA

Na, C., & Woo, J. (2026). Spatial-Temporal Graph Diffusion Policies for Safe Bimanual Robotic Manipulation. International Journal of Information and Communication Technology Trends, 6(1), 258-268. https://doi.org/10.66525/ijictt90

Vancouver

Na C, Woo J. Spatial-Temporal Graph Diffusion Policies for Safe Bimanual Robotic Manipulation. International Journal of Information and Communication Technology Trends. 2026;6(1):258-268. doi:10.66525/ijictt90

BibTeX
@article{Na2026,
  author = {Na, Chaewon and Woo, Joonho},
  title = {Spatial-Temporal Graph Diffusion Policies for Safe Bimanual Robotic Manipulation},
  journal = {International Journal of Information and Communication Technology Trends},
  year = {2026},
  volume = {6},
  number = {1},
  pages = {258--268},
  doi = {10.66525/ijictt90}
}
Kostenlos starten →