Cyber-physical systems (CPS) are increasingly deployed in safety-critical applications, making them prime targets for adversarial attacks. Timely detection and mitigation of such attacks are imperative for the safe operation of CPS. This paper proposes a novel residual generator design method for enhanced detection of bias injection attacks (BIAs) in linear CPS driven by white Gaussian noise. Specifically, we define a flexible attack impact metric based on the weighted norm of the injected bias and a detectability metric based on the Kullback-Leibler divergence. Using these two metrics, we characterize the worst-case BIAs as those that minimize detectability while maintaining a specified minimum impact. For residual generation filter synthesis, we formulate two optimization problems: one for maximizing the detectability of worst-case BIAs at the attack onset and the other at steady state. Since these two problems are inherently conflicting, we employ the ϵ-constraint method to obtain Pareto-optimal solutions that balance transient and steady-state detectability. The effectiveness of the proposed filter design method is demonstrated through numerical simulations, with a comparison against two state-of-the-art benchmarks: the Kalman filter and the H_ / H2 filter.

Efficient trajectory generation is essential for applications such as environmental monitoring using autonomous systems, where navigating complex, non-convex environments poses significant challenges. In this paper, we introduce a novel hybrid continuous optimization method that extends gradient techniques to continuous-time frameworks for trajectory planning in non-convex landscapes. The method combines adaptive step-size adjustments with momentum-based gradient techniques, enabling robust navigation through plateau regions and accurate convergence to optimized trajectories. We apply the method to drone trajectory optimization in environments modeled by Gaussian Mixture Models (GMMs), specifically focusing on wildfire monitoring in northern Mexico. The proposed approach generates smooth trajectories and achieves faster convergence when compared to traditional discrete and continuous methods without the hybrid adaptation. Simulation results showcase the method’s potential for broader applications in control systems and other domains requiring optimization in high-dimensional non-convex environments.