Over the past two decades, Unmanned Aerial Vehicles (UAVs), or drones, have gained significant attention and adoption across diverse domains, including logistics, disaster response, and surveillance. Drone design is inherently multidisciplinary, requiring the integration of aerodynamics, power management, embedded software, and fault tolerance. However, current simulation frameworks typically isolate these concerns, making it difficult to analyze them and to design effective fault management strategies. This paper proposes a unified co-simulation framework that enables concurrent modeling and analysis across mechanical, electrical, software, and environmental domains. The framework integrates: (1) SystemC AMS for modeling analog/mixed-signal, electromechanical, and power subsystems; (2) an Instruction Set Simulator (ISS) for software execution and profiling; and (3) Unreal Engine for high-fidelity visualization, environmental modeling, and user interaction. The proposed integration supports holistic design space exploration and early fault injection for a preliminary evaluation of fault analysis and for the design of fault recovery mechanisms. The platform promotes a white-box approach to UAV design, providing insight into how faults propagate across subsystems and performing efficient power design space exploration.
Autonomous cyber-physical systems increasingly use end-to-end vision-based controllers, with deep neural networks processing rich sensory inputs to make real-world decisions. However, existing safety verification tools do not scale to high-dimensional controllers whose inputs are thousands of dimensions, such as pixel-based images. This article addresses this limitation by approximating high-dimensional vision-based neural controllers with multiple low-dimensional neural controllers by leveraging verification-aware knowledge distillation. We then inflate the low-dimensional reachability and statistical approximation analyses, providing the high-confidence reachability guarantees for the original high-dimensional controller. Three reachable tube inflation techniques are investigated based on the state and action discrepancies between trajectories and individual time steps. All of them show promising performance in three OpenAI gym benchmarks.
This work presents testing of a Control Barrier Function (CBF) supervised Automated Vehicle (AV) in live freeway traffic. The CBF is designed and implemented using a common dynamical model for AV longitudinal control and a time-gap based safety property combined with CAN bus injection software/hardware. 1.17 hours of car-following data is collected from driving in congested freeway traffic. We analyze the extent to which the CBF controlled AV satisfied three properties: 1) forward-invariance of the safety property, 2) recovery of the safety property, and 3) collision avoidance. Our main findings are as follows. Forward-invariance was not strictly satisfied across all states. When trajectories begin satisfying the safety property it was violated by a maximum of \(3.4[m]\) , and \(90\%\) of violations were less than \(2[m]\) . Recovery was also not strictly satisfied across all states. For trajectories which begin outside of the safe set due to merge-in events, if the violation to the safety property was by more than \(5[m]\) the AV was recovering back to the safe set in more than \(98\%\) of the time. Finally, the minimum spacing-gap was \(11.7[m]\) . Across the tests the AV remained far from any collisions. We additionally analyze errors between control inputs and achieved outputs and hypothesize that unaccounted for modeling errors may lead to under-braking compared to what the CBF logic specifies, which may contribute to the observed property violations. Using the collected experimental data we consider the role that control input delay may have played as an unmodeled error source. We perform model calibration considering both a traditional goodness of fit metric as well as capacity to model observed safety violations. Finally we propose two new approaches for handling delay in CBF AV control both based on lead vehicle acceleration estimation and compare this to existing approaches. We use collected experimental data to quantify this. We find that our proposed approach requires less added robustness to achieve forward-invariance in safety compared to existing approaches.
Predicting spatiotemporal patterns is critical for traffic flow forecasting in Intelligent Transportation Systems (ITS), as accurate predictions can significantly enhance traffic management and decision-making. Recent data-driven approaches, particularly Graph Neural Networks (GNNs) integrated with physics-informed partial differential equations (PDEs), have demonstrated strong capability in modeling complex traffic dynamics. However, these models may still produce overconfident yet incorrect predictions, which can be particularly problematic in safety-critical scenarios. Uncertainty quantification (UQ) enhances the reliability of predictions and the practical application of NNs by estimating prediction confidence, not just accuracy. Despite its importance, existing UQ methods in traffic forecasting are primarily developed for purely data-driven models, and the role of physics-informed modeling in uncertainty estimation remains insufficiently understood. In this paper, we bridge this gap by integrating multiple UQ baseline methods with physics-informed modeling to systematically investigate how physical constraints influence uncertainty estimation in traffic forecasting. We further propose a physics-informed loss function that enhances the model's ability to capture physically consistent dynamics and improves the calibration and reliability of uncertainty estimates. In addition, we evaluate robustness under both noisy and adversarial perturbations, showing that our approach yields more stable predictions and uncertainty estimates under distribution shifts. Extensive experiments on real-world traffic datasets demonstrate that our approach improves both prediction accuracy and uncertainty quality, achieving up to 14.8% improvement in short-term and 8.7% in long-term traffic speed prediction errors, while providing better-calibrated and more robust uncertainty estimates.
The newly proposed multimodal transformer architecture offers a new paradigm for UAV detection and aerial object recognition. It introduces an innovative way of feeding multiple data streams, such as audio, infrared video, RGB video, and radar, into the architecture for processing, using independent modalities. The unique features of each modality are attached and processed together in the architecture, where the features are then exposed to the multimodal transformer for classification. Thus, all complementary information can be pooled within the integration framework to allow the model discrimination of any drone target under outdoor conditions from other aerial objects such as birds, helicopters, and airplanes. These methodologies are expected to outperform traditional single-modality systems by improving detection accuracy through class balancing and addressing modality-specific limitations. The proposed model has been further tested through various experiments to evaluate its robustness under conditions such as missing entries, corrupted data, and synthetic inputs. The results suggest that it has strong potential to serve as a benchmark in UAV detection. Thus, this work takes part of an emerging body of sensor fusion and deep learning-related research, demonstrating the potential of multimodal data in real-world detection problems.
