Year: 2026
Type: article
Abstract: In recent years, multi-unmanned aerial vehicle (UAV) systems are widely adopted in missions where individual vehicles are required to collaborate with each other under dynamic communication, physical degradation, sensor failures and hostile attacks. One of the challenges is that a mistaken and compromised UAV can provide false information and they need to be treated in different ways. Unless untimely, a suspected compromise should be quarantined and progressively reduced in trust, while the identified physical fault should be diagnosed, taken care of and progressively returned to production. This paper introduces TASHA (Trust-Aware Self-Healing Agentic AI), a layered AI system that is centered on this split. The proposed architecture delivers integration of a probabilistic trust engine; two-channel evidence made available by reported information and physical health telemetry on the target task; fault-attack diagnosis; a recovery state machine; hierarchical task replanning and bounded deterministic safety shield by a mission planner acting as an agent. The framework is tested by an abstract multi-UAV simulation with benign faults, constant attacks, combined fault, attack with low speed of and low altitude to the target, collusive attackers and adaptive sleepers’ attackers. The reported evaluation is based on 20 UAVs, 400 simulation steps, 150 random runs per scenario and the same random numbers were used. The results showed that TASHA contributes to minimizing the latency in detecting UAVs, keeps the repair and reintegration rate of the fleet high and caps the impact of rogue drones. In the constant-attack scenario, for instance, the detection latency for TASHA is 9.7 steps. For the 30% colluding scenario, the mean position error reported drops from 1.61 m without defence to 0.49 m with TASHA. The study is presented as a proof-of-concept for a simulation-based study. Future work includes an empirical validation of the bounded language-model planner, high-fidelity flight simulation and real-radio experiments and hardware validation.
Authors: Devender Singh, Piyush Verma, Madhav Sharma
Language: English
Topic: Access Control and Trust
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7235829323
