Language-guided long-horizon mobile manipulation has long been a grand challenge in embodied semantic reasoning, generalizable manipulation, and adaptive locomotion. Three fundamental limitations hinder progress: First, although large language models have shown promise in enhancing spatial reasoning and task planning through learned semantic priors, existing implementations remain confined to tabletop scenarios, failing to address the constrained perception and limited actuation ranges characteristic of mobile platforms. Second, current manipulation strategies exhibit insufficient generalization when confronted with the diverse object configurations encountered in open-world environments. Third, while crucial for practical deployment, the dual requirement of maintaining high platform maneuverability alongside precise end-effector control in unstructured settings remains understudied in the literature. In this work, we present ODYSSEY, a unified mobile manipulation framework for agile quadruped robots equipped with manipulators, which seamlessly integrates high-level task planning with low-level whole-body control. To address the challenge of egocentric perception in language-conditioned tasks, we introduce a hierarchical planner powered by a vision-language model, enabling long-horizon instruction decomposition and precise action execution. At the control level, our novel whole-body policy achieves robust coordination of locomotion and manipulation across challenging terrains. We further present the first comprehensive benchmark for long-horizon mobile manipulation, evaluating diverse indoor and outdoor scenarios. Through successful sim-to-real transfer, we demonstrate the system’s generalization and robustness in real-world deployments, underscoring the practicality of legged manipulators in unstructured environments. Our work advances the feasibility of generalized robotic assistants capable of complex, dynamic tasks.

Most of the existing generation models encounter notable challenges in complex scenes, particularly with inaccuracies in facial organs and textures that do not align with actual conditions. Traditional face restoration methods, heavily dependent on facial geometry and reference priors, often generate incorrect facial images that contribute misleading prior information. In this study, a Misinformation-Flitered GuideDenoising Diffusion Probabilistic Models (MF-GDDPM) is proposed to address these issue. Specifically, MF-GDDPM employs low-pass filtering to remove high-frequency details that contain misleading prior information. This process results in filtered low-dimensional facial contours that guide the diffusion model in generating high-quality facial images. To further enhance the fidelity of the generated results, a dualstream encoder within the Denoising Unet is constructed to process facial contours and high-dimensional details separately, while the Attention Feature Fusion (AFF) attention mechanism ensures the fidelity of image restoration. We have also incorporated the Natural Image Quality Evaluator (NIQE), a deep learning-based image quality assessment tool, into our framework as a novel loss function to crucially ensure the naturalness of restored images. Overall, the proposed method marks a significant improvement in generating accurate and clear facial images using diffusion models.

Cloud Removal (CR) utilizing Deep Learning (DL) has been widely employed to enhance the downstream applications of Remote Sensing (RS) satellite imagery affected by cloud coverage. Segmenting thin and thick cloud images into separate training sets and utilizing the CR model with a targeted learning strategy will result in improved performance. In this work, we propose a one-stop automatic cloud processing scheme for CR, including cloud classification and effective CR. To address the varying visibility of thin and thick cloudy images, we train a cloud classification network to distinguish between these two types of cloudy images, subsequently feeding them into two distinct CR networks. Furthermore, we propose a Semantic Distortion-Aware Network (SDAN) designed for similar cloud images after classification. Within SDAN, the Distortion Swin Transformer Block (DSTB) enhances the capability to extract contextual semantic information by incorporating global feature extraction and expression. This enhancement allows for targeted learning for thin or thick cloud images. Experiments conducted on the CR dataset named RICE demonstrate the enhanced performance of our model compared to various existing CR methods.