Academic Journal

Generalizable Context-Aware Deep Assignment Network for the Perimeter Defense Problem.

Λεπτομέρειες βιβλιογραφικής εγγραφής
Τίτλος: Generalizable Context-Aware Deep Assignment Network for the Perimeter Defense Problem.
Συγγραφείς: Gurumurthy, Vignesh1 (AUTHOR) vigneshg@iisc.ac.in, Velhal, Shridhar1 (AUTHOR), S A, Samahith1 (AUTHOR), Sundaram, Suresh1 (AUTHOR)
Πηγή: Unmanned Systems. Jul2026, p1-13. 13p.
Θεματικοί όροι: *Assignment problems (Programming), *Deep learning, *Context-aware computing, *Convolutional neural networks
Περίληψη: This paper presents a generalizable decentralized Context-aware Deep Assignment Network (CDAN) tailored for addressing Perimeter Defense Problems (PDPs). In PDP scenarios, a group of defenders operates along a segmented convex closed perimeter, aiming to intercept intruders attempting to breach it. The PDP is framed as an assignment learning problem for the defenders to sequentially capture intruders. A context map is generated to depict the intruders’ trajectories, serving as input for the CDAN. Utilizing 3D Convolutional Neural Networks (3D-CNNs), the CDAN assigns multiple intruders for sequential capture by the defenders in a decentralized manner. Training CDAN in a decentralized fashion using ground-truth data from a centralized PDP solution enables the model to be reused across the entire defender team. Comparative performance analysis indicates that the CDAN approach outperforms existing decentralized strategies, capturing 6% more intruders. The CDAN’s context-awareness, facilitated by the context-map representation, contributes to its generalizability in accommodating variations in the number of defenders, intruder velocities, and perimeter shape and length. [ABSTRACT FROM AUTHOR]
Βάση Δεδομένων: Academic Search Index
Περιγραφή
ISSN:23013850
DOI:10.1142/s2301385028500112