Accurate multi-agent trajectory prediction in urban scenarios requires accurate modeling of agent interactions and map constraints under complex traffic dynamics. Existing methods often struggle with computational inefficiency when processing dense traffic scenarios, leading to suboptimal capture of essential interaction patterns. Our architecture employs dual-scale attention modules to comprehend traffic scenarios. The Localized Cross-Attention module dynamically constructs agent-centric subgraphs through spatiotemporal attention operations, resolving immediate agent-agent conflicts and ensuring precise agent-lane alignments. Meanwhile, the Global Cross-Attention module captures extended social networks and topological influences across interconnected road systems. A Temporal Cross-Attention Module further synchronizes these spatial hierarchies across historical and predictive time horizons, enabling computationally efficient multi-modal trajectory generation. The hierarchical attention paradigm uniquely preserves urban-specific motion patterns, particularly at conflict-rich intersections where trajectory distributions bifurcate based on lane connectivity and traffic rules. Extensive experiments on Argoverse1 demonstrate our method’s superior capability in urban multi-agent prediction tasks, achieving balanced performance in capturing both local interaction patterns and global traffic flow constraints.

Under the background of dual carbon, lithium batteries are widely used in the energy field. However, range anxiety limits the popularity of electric vehicles. Optimization strategies for fast charging of lithium-ion batteries have been extensively studied to solve this problem. The huge parameter space of charging protocols and the complex aging mechanism of batteries limit the application of fast charging methods. The ability of reinforcement Learning to learn from the environment and adapt to the stochastic nature of battery behavior is a significant advantage. In this paper, we propose an innovative method for fast charging lithium-ion batteries using the Two-Delay Deep Deterministic Policy Gradient (TD3) algorithm and the Single Particle Model with Electrolyte model. The trained agent dynamically adjusts the charging current according to the state every 5 seconds to optimize the trade-off between fast charging and safety limits and can charge the state of charge (SOC) of the battery from 0.2 to 0.8 in 410 seconds, while protecting against overvoltage and overheating. Meanwhile, it still works well for different initial SOC.

The low-level control problem of fixed-wing unmanned aerial vehicles(UAVs) is complex and challenging. Although reinforcement learning (RL) algorithms are widely applied in the field of robot control, their convergence speed is typically slow due to the sparsity of rewards. An imitation reinforcement learning(IRL) algorithm based on proximal policy optimization(PPO) algorithm for attitude control of fixed-wing UAVs has been proposed. The strategy of imitation learning enables the avoidance of ineffective and perilous exploration during the initial stages of training. The nonlinear adaptive distribution rejection control (NLADRC) algorithm has been devised as the imitated agent, and a smooth transition strategy has been proposed to effectuate the switch from imitation learning to reinforcement learning. A series of simulation experiments demonstrate that the proposed strategy significantly improves the quality of reinforcement learning training, and even in strong gale wind environments, it still demonstrates superior control effectiveness compared to the imitated agent.

Accurate trajectory prediction is crucial in autonomous driving to ensure safe and efficient navigation, yet effectively modeling complex interactions among multiple agents remains a significant challenge. Many existing methods still suffer from over-reliance on HD maps, high computational cost, and a lack of interpretability in interaction reasoning. In response, the proposed model innovatively incorporates counterfactual reasoning into social interaction modeling to tackle the challenges of interaction-aware multi-agent trajectory prediction, prioritizing both accuracy and efficiency. In light of the spatiotemporal interaction mechanism and the inherent human cognition governing agents’ motion, the approach simultaneously generates multi-modal trajectories for all agents in a scenario, providing a novel perspective for modeling social interactions through causal reasoning. The results demonstrate that our map-free, lightweight trajectory prediction model rivals the performance of state-of-the-art methods and shows notable improvements over various baselines on publicly available real-world datasets.

Paroxysmal sympathetic hyperactivity (PSH) is an important clinical problem of severe traumatic brain injury (TBI) which incurs approximately 90% of all TBI-related costs. However, current detection approach is hampered by no consensus clinical diagnostic criteria, paroxysmal episode feature with complex manifestations, and already overloaded clinical activities. These limitations cause delayed recognitions which result in poor clinical outcomes. In this paper, we design an integrated Medical Cyber-Physical System (Medical CPS) for early detection of paroxysmal sympathetic hyperactivity patients. First, a formal model is proposed to describe clinical diagnostic criteria. With the formalized models employed, we implement an early detector and integrate it with revised medical device adapters into Medical CPS. Our system will monitor patient conditions automatically and continuously to relieve medical staff from the heavy burden of clinical activities and provide timely decision supports. Evaluations on 107 clinical cases extracted from medical publications demonstrate the effectiveness and the efficiency of our integrated system.