Understanding the fine-grained trajectories of metro passengers, especially at the train and route levels, is essential for analyzing system-level dynamics and individual behavior. However, existing approaches often rely on strong behavioral priors or simplified boarding assumptions, limiting their generality and realism. This study proposes a fully data-driven framework for passenger trajectory inference that explicitly incorporates physical capacity constraints and crowding effects. Entry, transfer, and egress walking durations are modeled using non-parametric Kernel Density Estimation (KDE) at the platform level. Based on these distributions, we construct a confidence-based model to estimate the probability of each feasible itinerary. A congestion-aware penalty function is introduced to reduce the confidence of infeasible itineraries involving overloaded in-vehicle links. To balance inference accuracy and computational efficiency, we develop a dynamic batch-size adjustment algorithm that iteratively updates train loads and refines probabilities. The framework is validated using large-scale AFC and timetable data from Chengdu Metro. Results demonstrate that the proposed method effectively suppresses violations of physical capacity constraints, improves behavioral plausibility, and provides reliable inputs for downstream applications such as resilience analysis and passenger behavior modeling.

Reinforcement learning (RL) has demonstrated remarkable capability in acquiring robot skills, but learning each new skill still requires substantial data collection for training. The pretrain-and-finetune paradigm offers a promising approach for efficiently adapting to new robot entities and tasks. Inspired by the idea that acquired knowledge can accelerate learning new tasks with the same robot and help a new robot master a trained task, we propose a latent training framework where a transferable latent-to-latent locomotion policy is pretrained alongside diverse task-specific observation encoders and action decoders. This policy in latent space processes encoded latent observations to generate latent actions to be decoded, with the potential to learn general abstract motion skills. To retain essential information for decision-making and control, we introduce a diffusion recovery module that minimizes information reconstruction loss during pretrain stage. During fine-tune stage, the pretrained latent-to-latent locomotion policy remains fixed, while only the lightweight task-specific encoder and decoder are optimized for efficient adaptation. Our method allows a robot to leverage its own prior experience across different tasks as well as the experience of other morphologically diverse robots to accelerate adaptation. We validate our approach through extensive simulations and real-world experiments, demonstrating that the pretrained latent-to-latent locomotion policy effectively generalizes to new robot entities and tasks with improved efficiency.

Motion control of autonomous vehicles (AVs) that considers nonlinear dynamics and multidimensional motion coupling characteristics represents a critical research direction, particularly for vehicles operating under extreme conditions. However, the nonlinearity of model and the time-varying characteristics of reference states make control law design challenging, often resorting to computationally expensive approaches such as model predictive control (MPC). This paper proposes an explicit nonlinear control design framework for optimal trajectory tracking of AVs. Specifically, taking the three-degree-of-freedom vehicle dynamics model as an example, we first augment the system state to yield an affine representation, followed by input-output feedback linearization. The discrete-time linearized system is then augmented with reference states from the preview horizon, and a linear quadratic regulator is designed to provide an analytical solution for the virtual control inputs. Finally, the original control inputs are computed using the feedback linearization control law. We performed simulations to evaluate the effectiveness of our proposed approach and compared it against MPC. Results indicate that our proposed approach achieves tracking accuracy, smoothness, and robustness comparable to MPC, while significantly reducing computational requirements, with a computing time of nearly 0 ms.

Reinforcement learning demonstrates strong capabilities in handling complex control tasks, especially in the field of autonomous driving where vehicles cope with uncertain environments. Existing reinforcement learning methods attempt to model the value distribution using unimodal, but in this modeling process, a significant amount of the complete distribution information is lost. In response to this problem, we propose the DACER++, an online multimodal distributional RL algorithm. DACER++ characterize the value distribution as multimodal will enhance the accuracy of characterizing the value distribution and improve algorithm performance. We construct the quantiles value network and use quantile regression to approximate the full quantile function of the state-action return distribution. This method allows for the precise modeling of multi-modal distributions, and formulates risk-sensitive policies adaptable to different environment. Then, We integrate quantiles value network with the actor-critic architecture algorithm DACER. Experiments on multi-goal tasks and MuJoCo benchmarks show that DACER++ not only has multimodal policy representation capability, but also achieves state-of-the-art performance. In stochastic vehicle meeting environments, DACER++ can learn different multimodal value distributions and multimodal trajectories according to various risk preferences, including the conservative and aggressive driving style.

Reinforcement learning (RL), known for its self-evolution capability, offers a promising approach to training high-level autonomous driving systems. However, handling constraints remains a significant challenge for existing RL algorithms, particularly in real-world applications. In this paper, we propose a new safety-oriented training technique called harmonic policy iteration (HPI). At each RL iteration, it first calculates two policy gradients associated with efficient driving and safety constraints, respectively. Then, a harmonic gradient is derived for policy updating, minimizing conflicts between the two gradients and consequently enabling a more balanced and stable training process. Furthermore, we adopt the state-of-the-art DSAC algorithm as the backbone and integrate it with our HPI to develop a new safe RL algorithm, DSAC-H. Extensive simulations in multi-lane scenarios demonstrate that DSAC-H achieves efficient driving performance with near-zero safety constraint violations.

Learning multimodal policies is crucial for enhancing exploration in online reinforcement learning (RL), especially in tasks with continuous action spaces and non-convex reward landscapes. While recent diffusion policies show promise, they often suffer from low computational efficiency in online settings. A more training-efficient paradigm involves modeling the policy as a Boltzmann distribution and guiding the action sampling directly with the gradient of the Q-value with respect to the action (proportional to the score function of the policy), such as via Langevin dynamics. However, analysis in this paper reveals that this gradient-guided approach suffers from two critical challenges: sampling instability caused by the widely varying magnitude of action gradients; and mode imbalance, where the sampling process inaccurately represents the weights of different high-value action modes. To address these challenges, this paper introduces three targeted techniques: score normalization and reshaping to stabilize the sampling process, and value-based resampling to correct mode imbalance. These techniques are then integrated into an actor-critic framework, resulting in the Score-Enhanced Actor-Critic (SEAC) algorithm. Simulation and real-world experiments demonstrate that SEAC not only effectively learns multimodal behaviors but also achieves state-of-the-art performance and high computational efficiency compared to prior multimodal RL methods. The code of this paper is available at https://github.com/THUzouwenjun/SEAC.

This paper examines the structural and reliability aspects of multilayer transportation networks. Using Space L, we construct these networks and develop a topology index system including node overlap degree, node activity degree, average path length, network diameter, clustering coefficient, betweenness, betweenness centrality, multilayer participation coefficient, edge overlap coefficient, and edge crossing coefficient. We evaluate reliability by analyzing node and edge failures. A case study demonstrates our method effectively captures node/edge heterogeneity and topological features in integrated transportation systems. Results show the metro-urban bus multilayer network exhibits small-world and scale-free properties. Nodes with high overlap/activity degree, betweenness, or participation coefficient significantly impact the network. Paths with larger edge overlap coefficients offer redundancy, faster recovery, and higher reliability post-failure. Networks with greater edge crossing coefficients exhibit enhanced redundancy and risk resistance.

Real-time safety-critical control is essential for high-level autonomous driving. Existing methods usually formulate safety-critical control as a constrained optimal control problem (COCP), and suffer from high computational complexity of the underlying iterative optimization processes. To address the issue of complexity, this paper presents an explicit safety-critical control method called the Control Safety Function (CSF) approach, which can replace online optimization with an analytical control law, dramatically enhancing real-time control capabilities. The CSF is formulated as the weighted sum of Control Lyapunov Function (CLF) and Control Barrier Functions (CBFs), with the value of CSF increasing to infinity as the state approaches the boundary of a safe set. The explicit control law is then derived from the gradient of CSF and system dynamics. Different from existing explicit controllers that can only apply to systems of relative degree one, the CSF method provides an approach to enforce safety constraints to systems with high relative degree, making CSF especially suitable for autonomous driving. The CSF approach is evaluated in a vehicle path-tracking scenario with multiple obstacles, accompanied by a comparative analysis against the Model Predictive Control (MPC) method. Simulation results indicate that CSF achieves control accuracy comparable to MPC, with significant reduction in computation time - approximately 3.23 ms per step, which is about 94.0% faster. These results suggest that CSF is a promising approach for real-time safety-critical control of high-level autonomous driving.

For assisted driving or unmanned driving, various information acquisition and comprehensive and effective information utilization will make the driving assistance system and the driving measures more reliable. With the support of advanced information acquisition, information interaction and other technologies, measuring the risk threat capability of each traffic element, and using potential field theory and risk pulse theory can effectively describe the risk distribution in the road traffic environment, which is conducive to ensuring driving safety and the implementation of driving control. In this paper, we use the risk pulse energy to measure the threat ability of each traffic element, establish the corresponding driving risk pulse field based on the risk analysis of each traffic element, measure the overall risk performance and overall level of the road traffic environment from the perspectives of vector superposition and quantity superposition, and establish a unified driving risk pulse field model. The characteristics of the established driving risk pulse field model are simulated and described, including basic risk pulse energy, random risk pulse energy and relevant parameters. Taking GM model as an example, a car following model based on driving risk pulse field is established by combining driving risk pulse field with GM model. Finally, the simulation analysis of car following model considering driving risk pulse field is carried out by taking a typical car following scene as an example. The results show that the car following model considering the impact of driving risk pulse field has the following advantages: (i) It can take into consideration the impact of the front and rear vehicle motion state changes on the current vehicle driving. (ii) The headway can be adjusted according to the change of the risk pulse energy of the front and rear vehicles. (iii) It is able to select appropriate driving strategies according to the risk pulse energy level of the front and rear vehicles. The whole process aims to minimize the risk threat to the vehicles. (iv) Based on the description of the driving risk pulse field, the visualization of the driving environment risk can be realized, and we can directly see and understand the driving risk situation and facilitate the appropriate and reasonable driving measures.

Vehicle control is one of the most critical challenges in autonomous vehicles (AVs) and connected and automated vehicles (CAVs), and it is paramount in vehicle safety, passenger comfort, transportation efficiency, and energy saving. This survey attempts to provide a comprehensive and thorough overview of the current state of vehicle control technology, focusing on the evolution from vehicle state estimation and trajectory tracking control in AVs at the microscopic level to collaborative control in CAVs at the macroscopic level. First, this review starts with vehicle key state estimation, specifically vehicle sideslip angle, which is the most pivotal state for vehicle trajectory control, to discuss representative approaches. Then, we present symbolic vehicle trajectory tracking control approaches for AVs. On top of that, we further review the collaborative control frameworks for CAVs and corresponding applications. Finally, this survey concludes with a discussion of future research directions and the challenges. This survey aims to provide a contextualized and in-depth look at the state of the art in vehicle control for AVs and CAVs, identifying critical areas of focus and pointing out the potential areas for further exploration.

The mobile edge computing (MEC) paradigm places traditional digital infrastructure next to mobile networks and thus drives substantial improvements in performance and latency for mobile computing cases like gaming, video streaming, and IoT. However, it remains a great challenge to provide a effictive and performance guaranteed strategies for services offloading and migration in the MEC environment. Most existing solutions in this direction tend to consider task offloading as a offline decision making process by employing transient positions of users as model inputs. In this work instead, we consider a predictive-trajectory-aware task offloading strategy called PreMig. Simulations clearly demonstrate that our proposed strategy outperforms traditional ones in terms of effective service rate and migration overhead.

Essential decision-making tasks such as power management in future vehicles will benefit from the development of artificial intelligence technology for safe and energy-efficient operations. To develop the technique of using neural network and deep learning in energy management of the plug-in hybrid vehicle and evaluate its advantage, this article proposes a new adaptive learning network that incorporates a deep deterministic policy gradient (DDPG) network with an adaptive neuro-fuzzy inference system (ANFIS) network. First, the ANFIS network is built using a new global K-fold fuzzy learning (GKFL) method for real-time implementation of the offline dynamic programming result. Then, the DDPG network is developed to regulate the input of the ANFIS network with the real-world reinforcement signal. The ANFIS and DDPG networks are integrated to maximize the control utility (CU), which is a function of the vehicle's energy efficiency and the battery state-of-charge. Experimental studies are conducted to testify the performance and robustness of the DDPG-ANFIS network. It has shown that the studied vehicle with the DDPG-ANFIS network achieves 8% higher CU than using the MATLAB ANFIS toolbox on the studied vehicle. In five simulated real-world driving conditions, the DDPG-ANFIS network increased the maximum mean CU value by 138% over the ANFIS-only network and 5% over the DDPG-only network.

Risk is an objective quantity applied to describe the degree of harm to a specific system of many activities and technologies. In order to manage risk successfully and ensure the normal operation of a system, an explicit system risk definition and a system risk analysis approach are essential. After analyzing the system risk change curve, a new definition of system risk is given, which is formulated as the risk energy change in a system. A Gaussian pulse energy function is introduced to calculate the changed risk energy, and a new unit H is given to describe the system risk, where H formulates each unit energy change in the system or subsystem caused by the risk pulse. Based on the Gaussian pulse energy function, an improved risk field is modeled and described from two perspectives: risk potential and risk field strength. The characteristics of the improved risk field are modeled and simulated, including the superposition law of the risk potential, superposition law of the risk field strength and risk force, the flux and divergence of the risk field, and the ring flux and rotation of the risk field. The risk potential superposition law and the risk field strength superposition law provide a new system risk analysis approach to explain the formation process of system coupling risk and the interaction laws of the risk factors in a system. Finally, a real-world case study is conducted by taking a railway dangerous goods transportation accident in 2001 as a background. The simulation results of risk pulse energy show that the system risk energy value of the risk-accident critical state is between 1.712e-04H and 1.922e-04H. When the fourth risk pulse appeared, the system risk change curve crossed the risk-accident critical state line, and the accident happened. The simulation results of the risk potential and risk field strength show that risk source 1 has the biggest risk potential value and risk field strength value, followed by risk sources 1, 3, and 4, which means that risk source 1 has the biggest contribution to the accident, and it has the biggest risk impact strength to other matters.