Developing non-collaborative dialogue agents traditionally requires the manual, unscalable codification of expert strategies.We propose METRO, a method that leverages large language models to autonomously induce both strategy actions and planning logic directly from raw transcripts.METRO formalizes expert knowledge into a Strategy Forest, a hierarchical structure that captures both shortterm responses (nodes) and long-term strategic foresight (branches).Experimental results across two benchmarks show that METRO demonstrates promising performance, outperforming existing methods by an average of 9%-10%.Our further analysis not only reveals the success behind METRO (strategic behavioral diversity and foresight), but also demonstrates its robust cross-task transferability.This offers new insights into building non-collaborative agents in a cost-effective and scalable way.Our code is available at

Unmanned aerial vehicles (UAVs) enable flexible network deployment but are vulnerable to detection due to the open-access nature of wireless signals. Visible light communication (VLC) emerges as a compelling technology for this context, offering inherent covertness through its reliance on line-of-sight links, while its wide spectrum supports high covert transmission rates. However, the high mobility of UAVs creates dynamic channel conditions that challenge traditional multiple access schemes. This motivates the utilization of rate-splitting multiple access (RSMA), a more robust and flexible framework for managing multi-user interference. Therefore, we investigate a cooperative, RSMA-based VLC system where a source UAV provides covert data transmission, assisted by a jamming UAV that creates power randomness to confuse wardens. By jointly optimizing the three-dimensional (3D) trajectory design and resource allocation of UAVs, we aim to maximize the minimum average covert transmission rate of users. To achieve this goal, the original problem is decomposed into three subproblems, namely horizontal coordinate arrangement, altitude adjustment, along with joint power assignment and rate-splitting, which are solved by the successive convex approximation method, geometric programming-based approach, and majorization-minimization algorithm, respectively. Numerical results validate the effectiveness of the proposed approach through extensive simulations under various parameters and comparative analyses against baselines, highlighting the potential of UAV-assisted visible light covert communications.

Low Earth orbit (LEO) satellite communications have received significant attention due to the global coverage and low-latency characteristics. Due to the limited size, weight, and power of LEO satellites, meeting the demand of satellite-terrestrial communications with minimal energy consumption while suppressing interference to co-existing systems remains a significant challenge. In this paper, a LEO satellite architecture with beam hopping (BH) is proposed, where the dynamic resource management and real-time load balancing are jointly implemented. Specifically, an optimization problem for the BH pattern, resource management, and load balancing with multi-satellite collaboration is formulated to achieve the long-term average energy minimization, while suppressing the interference to the co-existing system. Subsequently, by introducing the Lyapunov optimization method, the long-term energy minimization is transformed into a series of dynamic optimization problems for each time slot, thereby ensuring both the energy minimization and queue stability, where the block coordinate descent algorithm is employed to alternately address these issues. In particular, the BH pattern is determined by the multiplier penalty method combined with the majorization-minimization algorithm. Additionally, the resource management is addressed through the successive convex approximation technique, while real-time load balancing is achieved via quadratic programming. Overall, the proposed method, which can be effectively implemented within polynomial time, outperforms other benchmarks in simulations by achieving interference suppression in co-existing systems and reducing the energy consumption, while satisfying the communication demands.

The mega low-Earth-orbit (LEO) satellite constellation can achieve high throughput and low latency satellite communications, which is highlighted for next-generation communications. In this paper, the LEO satellite communications with coordinated beam hopping (BH) are investigated for throughput maximization and load balancing. Generally, the geographically non-uniform and time-varying packet arriving, along with the interference caused by dense beams, can present significant challenges. Hence, the proper BH pattern selection and dynamic power allocation are necessitated to enhance the throughput of satellite-terrestrial communications. Additionally, load balancing via inter-satellite links (ISLs) adjusts the queue length among satellites, thus indirectly altering the throughput by mitigating congestion. Therefore, an optimization problem is formulated to maximize long-term throughput by determining the BH pattern and dynamic power, while the problem for load balancing via ISLs is also proposed for stabilizing queues in a low congestion state to enhance the throughput. As both problems are inherently associated with each other through the queue length, they are solved in cycles. Specifically, the problem for long-term throughput is converted into decisions for each time slot with the Lyapunov drift-plus-penalty method, and then the BH pattern and power allocation are optimized alternately through successive convex approximation. Moreover, load balancing via ISLs is determined according to the queue length by minimizing the upper bound of Lyapunov drift, thereby decreasing the congestion of queues. Simulation results indicate that the proposed method can effectively improve the throughput and reduce the packet loss compared with the existing baselines, demonstrating the effectiveness of the Lyapunov optimization based method.

With the rising sophistication of cyber threats and the growing reliance on communication infrastructure, ensuring secure transmission has become a critical challenge. This work investigates the visible light covert communication for unmanned aerial vehicle (UAV) systems employing rate-splitting multiple access, aiming to conceal the very presence of transmission. Besides, the optimal detection threshold and detection error probability are derived to assess the covertness quantitatively. Through jointly optimizing the trajectory design and resource allocation, this work is devoted to maximizing the minimum average covert transmission rate (MACTR) among multiple users. Simulation results verify that the proposed algorithm substantially improves the MACTR compared to the baseline and conventional multiple access schemes, highlighting its effectiveness in enhancing the security for UAV systems.

Uncrewed aerial vehicles (UAVs) have gained significant attention due to their low cost and efficient deployment. Specifically, UAV-based communications hold promise to supplement the coverage limitations of terrestrial networks and achieve superior connectivity for the Internet of Things (IoT). Moreover, UAVs with visible light communication (VLC) capability do not interfere with existing radio frequency (RF) communication infrastructure, thereby providing flexible, high-throughput on-demand wireless communications. In this article, considering the limited battery capacity of UAVs, we aim to maximize the energy efficiency (EE) of a multi-UAV downlink VLC system for IoT. To this end, the system model with co-channel interference is first established, followed by the formulation of the joint spatial deployment and resource allocation problem. Besides, the original problem is decomposed into three subproblems, focusing on spatial deployment, subchannel assignment, and power allocation. By employing the block coordinate descent (BCD) method, the three subproblems are solved alternately. Moreover, simulation results demonstrate that the proposed method achieves superior improvements in EE compared to other baselines. In addition, the convergence and computational complexity of the proposed algorithm, as well as the impacts of key parameters on the overall system performance are also investigated.

Abstract Multi-object pose estimation in surveillance scenes is challenging and inaccurate due to object motion blur and pose occlusion in video data. Targeting at the temporal dependence and coherence among video frames, this paper reconstructs a multi-object pose estimation model that integrates spatiotemporal semantics for different scales and poses of video multi-objects. The model firstly, with an end-to-end detection framework, detects multiple targets in the video. Secondly, it enhances the positioning of key points of human body using the temporal cues among video frames and designs modular components to enrich the pose information, effectively refining the pose estimation. Finally, the improved human skeleton behavior recognition model based on pose estimation is employed to recognize the classroom behaviors of students oriented to video streams. Comparison with multiple classifiers through experiments reveals that the human skeleton behavior recognition model for multi-object pose estimation combined with spatiotemporal semantics exhibits an effectively improved accuracy.