Safe and reliable control provides a critical guarantee for the stable operation of quadrotor unmanned aerial vehicles (UAVs) in low-altitude scenarios. This article proposed a safety-critical control strategy for a quadrotor UAV through the integration of output constrained tracking and fault-tolerant mechanisms. First, a barrier Lyapunov function is incorporated into a backstepping control framework to rigorously enforce time-varying output constraints. Subsequently, a composite estimation module comprised of a disturbance observer and radial basis function neural networks is constructed to jointly compensate the influence of external disturbances and actuator faults. Furthermore, to address the noise amplification problem, a command filter is employed to derive a virtual control law of the proposed system, while an auxiliary system is designed to alleviate the impact of filter error. Besides, semi-globally bounded stability is proved by the Lyapunov direct method. Finally, the proposed safe and reliable control scheme is implemented on a quadrotor platform, where its effectiveness is verified by both simulations and experiments.

The White House Executive Order on Artificial Intelligence highlights the risks of large language models (LLMs) empowering malicious actors in developing biological, cyber, and chemical weapons. To measure these risks, government institutions and major AI labs are developing evaluations for hazardous capabilities in LLMs. However, current evaluations are private and restricted to a narrow range of malicious use scenarios, which limits further research into reducing malicious use. To fill these gaps, we release the Weapons of Mass Destruction Proxy (WMDP) benchmark, a dataset of 3,668 multiple-choice questions that serve as a proxy measurement of hazardous knowledge in biosecurity, cybersecurity, and chemical security. To guide progress on unlearning, we develop RMU, a state-of-the-art unlearning method based on controlling model representations. RMU reduces model performance on WMDP while maintaining general capabilities in areas such as biology and computer science, suggesting that unlearning may be a concrete path towards reducing malicious use from LLMs. We release our benchmark and code publicly at https://wmdp.ai.

This article aims to address the realistic path tracking control problem toward high-system performance for commercial autonomous ground vehicles (AGVs) with simultaneously guaranteeing the tracking accuracy, yaw and roll stability under limited vehicle network resources in global position system temporarily unavailable environments. In such conditions, the vehicle full state information and road topography might not be accessible in real time. To this end, this article proposes an effective adaptive event-trigger (AET)-based robust path tracking control strategy with introducing the reliable Takagi–Sugeno (T–S) fuzzy state observer for practical implementation. First, the vehicle yaw and roll coupled dynamics is incorporated into the vehicle-road system model, with modeling the tire cornering stiffness uncertainty by the T–S fuzzy technique and resolving the system disturbances as unknown inputs. Then, the fuzzy observer structure is established with unmeasurable premise variables which are handled by norm-bound method. Next, a well-designed AET control framework is constructed to reduce the real-time network occupation rate and economize the communication bandwidth resources. Besides, the input constraint and rollover prevention are handled using the robust set invariance. After that, the parallel distributed compensation (PDC) controller and observer are co-designed through solving the effective linear matrix inequalities (LMIs). In addition, the close-loop stability and$H\infty$performance are ensured by means of the delay dependent Lyapunov–Krasovski method. Finally, the validity and superiority of the proposed control strategy have been verified by Carsim-Simulink co-simulations in different dynamic scenarios with high-fidelity full vehicle model.

This article proposes an actuator fault detection and isolation (FDI) scheme for quadrotor unmanned aerial vehicle (UAV) with a cable-suspended load. First, a linear parameter-varying (LPV) model of quadrotor UAV is established, in which the effects of cable-suspended load are considered. Then, a state boundary-based FDI design is systemically presented. A bank of interval observers is constructed to build the preliminary upper and lower boundaries of system states under healthy conditions, where$H_{-}/H_{\infty }$performance is applied to enhance its robustness against disturbances and sensitivity to faults. Furthermore, a novel updating strategy is further proposed to periodically adjust state boundaries to cope with the effects of varying wind disturbances. Finally, based on the QDrone platform, experimental tests under random faults are carried out to verify the effectiveness and performance of the proposed scheme.

Benchmarks for language-guided embodied agents typically assume text-based instructions, but deployed agents will encounter spoken instructions.While Automatic Speech Recognition (ASR) models can bridge the input gap, erroneous ASR transcripts can hurt the agents' ability to complete tasks.We propose training a multimodal ASR model that utilizes the accompanying visual context to reduce errors in spoken instruction transcripts.We train our model on a dataset of synthetic spoken instructions, derived from the ALFRED household task dataset, where we simulate acoustic noise by systematically masking spoken words.We find that utilizing visual observations facilitates masked word recovery, with multimodal ASR models recovering up to 30% more masked words than unimodal baselines.We also find that spoken instructions transcribed by multimodal ASR models result in higher task completion success rates for a language-guided embodied agent.github.com

Distributed drive electric vehicles (DDEVs) eliminate the complex drivetrain. The independently driven in- wheel motors also endow the vehicle with more ability for improving maneuverability. To this end, this paper proposes an integrated control framework of torque vectoring (TV) and active front-wheel steering system (AFS) to ensure the vehicle lateral motion stability performance. First, the polytope method with finite vertices is employed to deal with the system uncertainties and simplify the modeling structure, based on which a distributed model predictive control is adopted to construct a dynamic interactive model between agents. Then, through introducing the game theory, a distributed parallel control scheme is developed to obtain the cooperative strategy of agents. Such a design can also satisfy the modular and scalable requirement for integrated chassis control. To ensure the system asymptotic stability, the terminal input combined with the terminal cost function are treated as the constraints in the game paradigm and then transformed as the linear matrix inequalities. Furthermore, a robust$\text{H}\infty $compensation method is used to suppress the system disturbance. Finally, the hardware-in-the-loop (HIL) tests are conducted to assess the control performance. The results verify the proposed integrated control scheme is effective to enhance the vehicle handling stability.

The information of the shaft torque and the resistance torque is crucial to develop advanced control and fault diagnosis/detection schemes for electrified powertrain systems. However, reliable physical sensors for torque measurement are not affordable for commercial vehicle applications. This article investigates the simultaneous estimation problem of the state dynamics and the resistance torque for integrated motor-transmission (IMT) systems of electric vehicles. To this end, the IMT system is first reformulated as a nonlinear switched model, where the resistance torque is considered as an unknown input (UI). This modeling reformulation allows taking into account not only the nonlinear nature of IMT dynamics but especially also the intrinsic discontinuity of the gear-shifting process. Then, we propose a nonlinear switched observer (NSO) structure to simultaneously estimate the nonlinear IMT dynamics, thus the shaft torque, and the unknown resistance torque. The observer design does not require any a priori information on the unknown resistance torque as for the classical proportional-integral observer design, nor the well-known matching condition for UI decoupling techniques. Using the Lyapunov stability theory, we derive sufficient conditions, expressed in terms of linear matrix inequality (LMI) constraints, to design an NSO with a guaranteed$\ell _{\infty }$performance to mitigate the negative effect of sensor noises and disturbances. In particular, we propose to incorporate LMI-based bumps limitation conditions in the optimization-based observer design to reduce the impacts of expressive discontinuities at switching instants. Comparative studies are performed between the related estimation methods to show the practical effectiveness of the proposed solution.

Most Relevant Explanation (MRE) is an inference task in Bayesian networks that finds the most relevant partial instantiation of target variables as an explanation for given evidence by maximizing the Generalized Bayes Factor (GBF). No exact MRE algorithm has been developed previously except exhaustive search. This paper fills the void by introducing two Breadth-First Branch-and-Bound (BFBnB) algorithms for solving MRE based on novel upper bounds of GBF. One upper bound is created by decomposing the computation of GBF using a target blanket decomposition of evidence variables. The other upper bound improves the first bound in two ways. One is to split the target blankets that are too large by converting auxiliary nodes into pseudo-targets so as to scale to large problems. The other is to perform summations instead of maximizations on some of the target variables in each target blanket. Our empirical evaluations show that the proposed BFBnB algorithms make exact MRE inference tractable in Bayesian networks that could not be solved previously.

Most Relevant Explanation (MRE) is a new inference task in Bayesian networks that finds the most relevant partial instantiation of target variables as an explanation for given evidence by maximizing the Generalized Bayes Factor (GBF). No exact algorithm has been developed for solving MRE previously. This paper fills the void and introduces a breadth-first branch-and-bound MRE algorithm based on a novel upper bound on GBF. The bound is calculated by decomposing the computation of the score to a set of Markov blankets of subsets of evidence variables. Our empirical evaluations show that the proposed algorithm scales up exact MRE inference significantly.

To develop effective learning algorithms for online cursive word recognition is still a challenge research issue. In this paper, we propose a probabilistic framework to model the inherent ambiguity of cursive handwriting by using soft target vector of each character class. In the proposed algorithm, the values of soft targets are estimated by introducing a lower bound on the log likelihood and optimizing this lower bound via an EM like algorithm. In the experiments on 207 K collected cursive words written by 1060 subjects, the proposed algorithm clearly outperforms baseline method with word error reduction up to 11.6%. Furthermore, the estimated soft target values are useful for measuring the separability between output classes.