This paper addresses challenges such as semantic gaps, logical inconsistencies, and neglect of domain knowledge in task decomposition for complex systems, especially in military command scenarios. We propose the BT-D2 framework — a behavior tree-guided method that integrates large language models (LLMs) with structured domain knowledge to enable dynamic and executable task decomposition. By embedding military doctrines and equipment parameters into hierarchical behavior tree nodes, the framework creates a constraint-based template for transforming high-level instructions into tactical action sequences. Key components include an extended behavior tree architecture for dynamic parameter binding, a multi-agent collaborative workflow for strategic-to-tactical planning, and an adaptive mechanism for semantic alignment and conflict resolution. Experimental results demonstrate that BT-D2 outperforms baseline methods (CoT, ReAct, GTN) in semantic integrity, decomposition diversity, and structural rationality across 200 combat scenarios. Ablation studies validate the critical role of behavior trees in enhancing task executability and hierarchical architecture in improving planning efficiency. This work contributes a novel framework for bridging generative AI with domain-specific constraints, offering promising applications in military planning, logistics, and intelligent systems.

With the development of Synthetic Aperture Radar (SAR), there is a growing demand for rapid SAR image processing. However, traditional Graphics Processing Unit (GPU) based processing faces challenges in meeting real-time application requirements, especially in scenarios like maritime search and rescue, due to time delays caused by input/output data transmission through the Peripheral Component Interconnect Express (PCIe) bus and high power consumption. To address this issue, our research proposes a real-time SAR ship detection system based on a lightweight Field-Programmable Gate Array (FPGA). The system utilizes a trainable pseudo-color synthesis network for SAR image preprocessing and employs a generic convolutional architecture to convert the YOLO v5 model into FPGA-executable hardware language. Experimental results indicate that the FPGA achieves a processing time of 68.9 milliseconds, significantly outperforming the GPU (234.7 milliseconds), without compromising detection accuracy. This research enhances SAR image detection speed, with implications for deploying object detection. algorithms on FPGAs.

In this article, we present the circular accessible depth (CAD), a robust traversability representation for an unmanned ground vehicle (UGV) to learn traversability in various scenarios containing irregular obstacles. To predict CAD, we propose a neural network, namely CADNet, with an attention-based multiframe point cloud fusion module, stability-attention module (SAM), to encode the spatial features from point clouds captured by LiDAR. CAD is designed based on the polar coordinate system and focuses on predicting the border of traversable area. Since it encodes the spatial information of the surrounding environment, which enables a semisupervised learning for the CADNet, and thus, desirably avoids annotating a large amount of data. Extensive experiments demonstrate that CAD outperforms baselines in terms of robustness and precision. We also implement our method on a real UGV and show that it performs well in real-world scenarios.

Transfer learning from the model trained on large datasets to customized downstream tasks has been widely used as the pre-trained model can greatly boost the generalizability. However, the increasing sizes of pre-trained models also lead to a prohibitively large memory footprints for downstream transferring, making them unaffordable for personal devices. Previous work recognizes the bottleneck of the footprint to be the activation, and hence proposes various solutions such as injecting specific lite modules. In this work, we present a novel memory-efficient transfer framework called Back Razor, that can be plug-and-play applied to any pre-trained network without changing its architecture. The key idea of Back Razor is asymmetric sparsifying: pruning the activation stored for back-propagation, while keeping the forward activation dense. It is based on the observation that the stored activation, that dominates the memory footprint, is only needed for backpropagation. Such asymmetric pruning avoids affecting the precision of forward computation, thus making more aggressive pruning possible. Furthermore, we conduct the theoretical analysis for the convergence rate of Back Razor, showing that under mild conditions, our method retains the similar convergence rate as vanilla SGD. Extensive transfer learning experiments on both Convolutional Neural Networks and Vision Transformers with classification, dense prediction, and language modeling tasks show that Back Razor could yield up to 97% sparsity, saving 9.2x memory usage, without losing accuracy. The code is available at: https://github.com/VITA-Group/BackRazor_Neurips22.

In this paper, we propose a novel deep reinforcement learning (DRL) system for the autonomous navigation of mobile robots that consists of three modules: map navigation, multi-view perception and multi-branch control. Our DRL system takes as the input a routed map provided by a global planner and three RGB images captured by a multi-camera setup to gather global and local information, respectively. In particular, we present a multi-view perception module based on an attention mechanism to filter out redundant information caused by multi-camera sensing. We also replace raw RGB images with low-dimensional representations via a specifically designed network, which benefits a more robust sim2real transfer learning. Extensive experiments in both simulated and real-world scenarios demonstrate that our system outperforms state-of-the-art approaches.

In the 2011 Tohoku tsunamigenic earthquake, ionospheric anomalies generated by tsunami-induced gravity waves were observed by many types of instruments. Three digisondes and one Doppler receiver located in East Asia were applied to investigate the far-field ionospheric response to the westward-propagating gravity waves generated by the earthquake. Based on time-period spectrum analysis, oscillations between 20 and 36 min on the $fo\text{F}2$ (critical frequency of F2 layer) curves and Doppler variations of each observation location were identified. The horizontal group speed, generation time, and source location estimated with the ray-tracing method all indicated that the periodic disturbances recorded by the four radio systems were gravity waves induced by the tsunami following the 2011 Tohoku earthquake. The plasma frequency variations at five fixed altitudes in the ionospheric F2 layer over I-Cheon were used to investigate the vertical propagation of the tsunami-associated gravity waves. The measured phase progression in the vertical direction was opposite that of the energy transport, which further confirmed that the recorded waves were atmospheric gravity waves. The use of multisite remote sensing for examining tsunami-induced waves in the ionosphere may open new perspectives in oceanic monitoring and future tsunami warning systems.