Low earth orbit Internet of things (LEO-IoT) networks offer global coverage and low-latency communication, making them attractive for large-scale IoT deployments. However, rapidly varying satellite connectivity and uneven, burst ground traffic lead to unstable routing performance, resulting in fluctuating delays and increased packet loss. To address these challenges, we propose a delay-aware routing optimization (DARO) algorithm that integrates traffic prediction and distributed control. A CNN-BiLSTM-Attention model is developed to capture spatial-temporal traffic patterns, enhancing the accuracy of dynamic traffic prediction. A closed-form end-to-end delay model is derived to characterize the effect of routing decisions on network latency. The routing problem is then formulated as a delay and packet loss minimization task and solved using multi-agent proximal policy optimization (MAPPO), enabling each satellite to adapt its routing strategy based on local observations and a shared critic. Simulation results show that DARO improves prediction accuracy by 19.05% to 76.39%, reduces packet loss by 27.35% to 90.76%, and lowers end-to-end delay by 2.43% to 56.04%, demonstrating its effectiveness in dynamic LEO-IoT environments.

With the growing demand for low-altitude transportation, the application of the Internet of drones (IoD) in urban logistics has become increasingly significant. However, the complex obstacles present in urban environments, such as tall buildings and no-fly zones, pose numerous challenges for the IoD, including low data-driven efficiency and difficulties in representing implicit knowledge. To address these challenges, this paper proposes an IoD swarm collaborative scheduling method based on distributed knowledge-enhanced multi-agent reinforcement learning (DKEMARL). The method leverages prior environmental knowledge to design a knowledge embedding and expansion module, which provides comprehensive and detailed environmental observation data during training and introduces a reward mechanism that balances task timeliness with flight safety. Within a centralized training and decentralized execution framework, the multi-agent deep deterministic policy gradient (MADDPG) approach is employed to facilitate efficient cooperation among the IoD. Specifically, we develop a scenario model for low-altitude transportation tasks involving an IoD swarm, considering factors such as task timeliness, flight distance cost, and safety constraints, and propose an optimization objective to balance timely task completion with flight safety. Simulation results demonstrate that the proposed DKEMARL algorithm can significantly enhance task completion efficiency compared to baseline methods that do not incorporate knowledge enhancement.

The increased deployment of multi-robot systems (MRS) in various fields has led to the need to analyze systemlevel performance. However, creating consistent metrics for MRS is challenging due to the wide range of team and task parameters, such as the number of robots and the size of the environment. This paper presents a new analytical framework for MRS based on dimensionless variable analysis that effectively condenses the complex relationships between the team and task parameters that influence MRS performance into a manageable set of dimensionless variables. Then we use these dimensionless variables to fit a predictive parametric model of team performance. We apply our methodology to two MRS applications: Multi-Robot Multi-Target Tracking (MR-MTT) and Multi-Agent Path Finding (MAPF). The application of dimensionless variable analysis to MRS offers a promising method for MRS analysis that effectively reduces complexity, improves understanding of system behavior, and can inform the design and management of future MRS deployments.

This paper proposes a novel neural network-based control policy to enable a mobile robot to navigate safety through environments filled with both static obstacles, such as tables and chairs, and dense crowds of pedestrians. The network architecture uses early fusion to combine a short history of lidar data with kinematic data about nearby pedestrians. This kinematic data is key to enable safe robot navigation in these uncontrolled, human-filled environments. The network is trained in a supervised setting, using expert demonstrations to learn safe navigation behaviors. A series of experiments in detailed simulated environments demonstrate the efficacy of this policy, which is able to achieve a higher success rate than either standard model-based planners or state-of-the-art neural network control policies that use only raw sensor data.