Security-Aware Sensor Fusion with MATE: the Multi-Agent Trust Estimator
Sensor fusion in multi-agent systems, including smart cities, often lacks security awareness and is vulnerable to attacks. We propose a security-aware sensor fusion framework that estimates and incorporates probabilistic trust with uncertainty to defend against compromised insider agents. Trust is modeled as a hidden Markov process and updated via Bayesian inference using novel trust pseudomeasurements (PSMs), which map discrepancies between expected and observed sensor data into trust evidence. Trust estimates reweight agent contributions during fusion and identify corrupted information, mitigating the influence of compromised nodes. Our system includes a dynamic field-of-view estimator using LiDAR ray tracing, novel logic for PSM generation, and efficient Bayesian updates with conjugate priors. Evaluated in adversarial scenarios, our method significantly reduces fusion error and accurately detects compromised agents. These results show trust-guided fusion enables resilient situational awareness under attack, making it suitable for adversarial cyber-physical environments.
