Recent advancements in end-to-end autonomous driving systems (ADSs) underscore their potential for perception and planning capabilities. However, challenges remain. Complex driving scenarios contain rich semantic information, yet ambiguous or noisy semantics can compromise decision reliability, while interference between multiple driving tasks may hinder optimal planning. Furthermore, prolonged inference latency slows decision-making, increasing the risk of unsafe driving behaviors. To address these challenges, we propose ExpertAD, a novel framework that enhances the performance of ADS with Mixture of Experts (MoE) architecture. We introduce a Perception Adapter (PA) to amplify task-critical features, ensuring contextually relevant scene understanding, and a Mixture of Sparse Experts (MoSE) to minimize task interference during prediction, allowing for effective and efficient planning. Our experiments show that ExpertAD reduces average collision rates by up to 20% and inference latency by 25% compared to prior methods. We further evaluate its multi-skill planning capabilities in rare scenarios (e.g., accidents, yielding to emergency vehicles) and demonstrate strong generalization to unseen urban environments. Additionally, we present a case study that illustrates its decision-making process in complex driving scenarios.

In massive multiple-input and multiple-output (mMIMO) systems, compressed sensing (CS)-based channel state information (CSI) feedback methods still face significant challenges, such as unknown channel sparsity, high computational complexity, and unavoidable channel estimation (CE) errors at the user equipment (UE). These factors degrade the accuracy of downlink CSI reconstruction at the base station (BS). To tackle these challenges, inspired by the work of perception-assisted communication, a perception-assisted CS-based CSI feedback method is proposed in this paper. In this method, an active perception scheme is developed to extract the support set of downlink CSI from the echo signals, addressing the issue of unknown channel sparsity in CS-based methods. With the perceived support set, the perception-assisted reconstruction without perception errors (PaRwoPE) method is proposed to achieve a lower bound of CSI recovery accuracy. This method develops a perception-assisted suppression scheme to reduce CE errors and transceiver noise and uses the perceived support set to avoid iterative reconstruction, thereby significantly improving the recovery accuracy of downlink CSI with markedly reduced computational complexity. To suppress the inevitable perception errors, a perception-assisted reconstruction with perception errors (PaRPE) method is developed. This method utilizes path correlation to eliminate false paths in the downlink CSI, while also developing a low-complexity iterative scheme to recover missed paths. The computational complexity analysis shows that the proposed method notably reduces computational complexity. Furthermore, simulation results demonstrate its effectiveness in improving normalized mean squared error (NMSE) performance and exhibit robustness against parameter variations.

End-to-end autonomous driving systems (ADSs), with their strong capabilities in environmental perception and generalizable driving decisions, are attracting growing attention from both academia and industry. However, once deployed on public roads, ADSs are inevitably exposed to diverse driving hazards that may compromise safety and degrade system performance. This raises a strong demand for resilience of ADSs, particularly the capability to continuously monitor driving hazards and adaptively respond to potential safety violations, which is crucial for maintaining robust driving behaviors in complex driving scenarios.To bridge this gap, we propose a resilience-oriented runtime framework, named Argus, to mitigate the driving hazards, thus preventing potential safety violations and improving the driving performance of an ADS. Argus continuously monitors the trajectories generated by the ADS for potential hazards and, whenever the EGO vehicle is deemed unsafe, seamlessly takes control via a hazard mitigator. We integrate Argus with three state-of-the-art end-to-end ADSs, i.e., TCP, UniAD and VAD. Our evaluation has demonstrated that Argus effectively and efficiently enhances the resilience of ADSs, improving the driving score of ADSs by 150.30% on average, and preventing 64.38% of the violations, with little additional time overhead.

Accurate acquisition of downlink channel state information (CSI) at the base station (BS) remains a critical challenge in frequency division duplex (FDD) millimeter wave (mmWave) massive multiple-input multiple-output (mMIMO) systems. Although compressive sensing (CS) and deep learning (DL)-based CSI feedback methods demonstrate their advantages, the recovery accuracy of downlink CSI in FDD mmWave mMIMO systems still faces severe challenges due to the facts of significant user equipment (UE) estimation errors, and typical compression requirements. To tackle these challenges, this paper proposes a novel echo sensing-aided CSI feedback framework designed to enhance downlink CSI recovery accuracy. In the proposed method, the communication echo signals observed at the BS are utilized to extract the dedicated sensing prior information for the downlink CSI recovery. With the extracted sensing prior information, a CSI denoising method is developed to suppress the non-path entries of the downlink CSI matrix in the angular-delay domain, thereby improving the recovery accuracy of downlink CSI at the BS. The framework operates as a plug-in module at the BS receiver, requiring no modifications to existing UE hardware. This work establishes a new paradigm for integrating sensing-assisted enhancements into FDD mmWave mMIMO systems without compromising compatibility with standardized UE operations. Simulations validate that the proposed method outperforms conventional CS and DL-based methods in terms of CSI reconstruction accuracy while demonstrating robustness to parameter variations.

Due to the wide bandwidth and large observation area, spaceborne synthetic aperture radar (SAR) is easily interfered by other electromagnetic signals, namely radio frequency interference (RFI), which can severely degrade SAR image quality and submerge useful information. Classic parametric and non-parametric methods are used to suppress RFI as much as possible without considering the useful information. To protect the real reflected signals, semi-parametric methods, based on low-rank and sparse recovery, are proposed to mitigate RFI, but they suffer from the singular-value over-shrinking problem when RFI is not strictly low-rank, resulting in interference residues in the recovered scene. Hence, in this paper, a robust interference self-cancellation (RISC) method is proposed to protect raw ground scenes from polluted data with better extraction accuracy of RFI. The proposed model can adaptively fit in different scenes and backgrounds by using adjacent homologous interference (HI) subregions instead of the low-rank constraints, thus better protecting SAR scenes and enhancing its robustness. Based on the alternating direction method of multipliers (ADMM), we design two different solvers for the proposed optimization model, and both are tested on four different scenes of Sentinel-1 measured data. All experiments demonstrate that the proposed method has excellent performance in RFI mitigation and SAR image recovery.

Contemporary advancements in large language models (LLMs), both proprietary and open-source, have demonstrated capabilities surpassing human performance across diverse general domains. Nevertheless, their efficacy in specialized professional fields, particularly medicine, remains suboptimal within the open-source ecosystem. To address these limitations, we introduce Domain Adaptative Medicine-LLM (DAMe-LLM). Our approach encompasses continued pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF) methodologies. In the development of DAMe-LLM, we have curated an extensive Chinese and English medical dataset for continued pre-training, complemented by a high-fidelity SFT dataset encompassing a broad spectrum of medical specialties. Our contribution provides the scientific community with valuable resources for further investigation and development in the domain of medicine artificial intelligence.