Simpler Is Better: Revisiting Doppler Velocity for Enhanced Moving Object Tracking with FMCW LiDAR
Yubin Zeng, Youjin Yu
CodeObject TrackingPoint Cloud
🎯 What it does: Proposes a learning-free tracking method called DopplerTrack based on FMCW LiDAR Doppler velocity, utilizing Doppler information for point cloud preprocessing, target detection, and velocity vector reconstruction to achieve efficient tracking of moving objects.
SimWorld: A Unified Benchmark for Simulator-Conditioned Scene Generation via World Model
Xinqing Li, Yunfeng Ai
CodeGenerationData SynthesisWorld ModelBenchmark
🎯 What it does: Proposed a simulator-conditioned scene generation engine based on a world model, and constructed a corresponding benchmark for scaling virtual and real data ratios.
Single-Microphone-Based Sound Source Localization for Mobile Robots in Reverberant Environments
Jiang Wang, Kazuhiro Nakadai
CodeRobotic IntelligenceAudio
🎯 What it does: Developed an online sound source localization method based on a single microphone, applicable for mobile robots in reverberant environments.
SLOOP: Aligned Coordinate System-aided LiDAR LOOP Closure Detection based on Semantic Node Graph Matching
Yujie Tang, Yufeng Yue
CodeAutonomous DrivingGraph Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposes SLOOP, a LiDAR loop closure detection method that first aligns semantic maps and then performs efficient similarity comparison.
🎯 What it does: Proposes a spatiotemporally efficient anomaly detection method called STEAD, focusing on rapidly detecting anomalies in time- and computation-constrained automated systems
STG-Avatar: Animatable Human Avatars via Spacetime Gaussian
Guangan Jiang, Hongyu Wang
CodeGenerationGaussian SplattingOptical FlowVideo
🎯 What it does: Proposed the STG-Avatar framework for reconstructing high-fidelity animatable human avatars from monocular videos, achieving more precise pose control and detail representation by coupling Spacetime Gaussians (STG) with Linear Blend Skinning (LBS), combined with optical flow-driven dynamic region adaptive Gaussian densification.
TACO: General Acrobatic Flight Control via Target-and-Command-Oriented Reinforcement Learning
Zikang Yin, Shiyu Zhao
CodeRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposes the TACO framework, which employs goal- and instruction-oriented reinforcement learning to achieve generalizable acrobatic flight control, and enhances the spatiotemporal smoothness and symmetry of the policy through spectral normalization and input-output re-calibration, overcoming the sim-to-real gap.
🎯 What it does: Propose the TEM3-Learning framework, which jointly optimizes driver emotion recognition, driver behavior recognition, traffic context recognition, and vehicle behavior recognition through a two-phase architecture.
🎯 What it does: Proposes the TextInPlace framework for indoor visual place recognition, addressing visual ambiguity in repetitive structures by combining scene text detection and verification.
TopoLiDM: Topology-Aware LiDAR Diffusion Models for Interpretable and Realistic LiDAR Point Cloud Generation
Jiuming Liu, Hesheng Wang
CodeGenerationData SynthesisAutonomous DrivingExplainability and InterpretabilityGraph Neural NetworkDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: Designed and implemented the TopoLiDM framework for generating high-fidelity LiDAR point clouds, combining graph neural networks with diffusion models and incorporating topological regularization.
Towards Efficient Image-goal Navigation: A Self-supervised Transformer-based Reinforcement Learning Approach
Qizhen Weng, Xiangwei Zhu
CodeTransformerReinforcement LearningImage
🎯 What it does: Proposes a self-supervised Transformer-based reinforcement learning method for image goal navigation, which utilizes a dual attention shared Transformer to predict masked visual-action embeddings and generate policies, thereby fully leveraging spatiotemporal relationships in visual-action history.
Towards Surgical Task Automation: Actor-Critic Models Meet Self-Supervised Imitation Learning*
Jingshuai Liu, S. Tsaftaris
CodeRobotic IntelligenceReinforcement LearningBiomedical Data
🎯 What it does: Propose an RL framework AC-SSIL based on Actor-Critic, which integrates self-supervised imitation learning (SSIL) to introduce expert demonstrations containing only states into RL.
Tracking-Aware Deformation Field Estimation for Non-rigid 3D Reconstruction in Robotic Surgeries
Zeqing Wang, Yutong Ban
CodeRobotic IntelligenceNeural Radiance FieldOptical FlowVideoMeshBiomedical Data
🎯 What it does: Proposed and implemented a Tracking-Aware Deformation Field (TADF) framework for non-rigid 3D reconstruction and deformation estimation of soft tissues in robotic surgery.
Triple-S: A Collaborative Multi-LLM Framework for Solving Long-Horizon Implicative Tasks in Robotics
Zixi Jia, Qinghua Liu
CodeRobotic IntelligenceLarge Language Model
🎯 What it does: Proposes a multi-LLM collaborative Triple-S framework to address robot long-horizon implication tasks, improving success rates through a closed-loop Simplification-Solution-Summary process.
UAV-DETR: Efficient End-to-End Object Detection for Unmanned Aerial Vehicle Imagery
Huaxiang Zhang, Guo-Niu Zhu
CodeObject DetectionTransformerImage
🎯 What it does: Proposed an efficient end-to-end detection transformer framework for UAV imagery, UAV-DETR, which includes multi-scale feature fusion and frequency enhancement modules, frequency-focused downsampling modules, and semantic alignment and calibration modules.
Understanding and Imitating Human-Robot Motion with Restricted Visual Fields
Maulik Bhatt, Negar Mehr
CodeRobotic IntelligenceReinforcement Learning from Human Feedback
🎯 What it does: Studied the agent's perception capability under limited field of view and range, decoupling perception models from motion strategies, and leveraging human perception modeling to better predict behavior; validated human navigation strategies within limited observation spaces through user experiments, enabled robots to learn this strategy for real-time navigation with minimal collisions, and successfully demonstrated it on physical hardware vehicles.
VAPO: Visibility-Aware Keypoint Localization for Efficient 6DoF Object Pose Estimation
Ruyi Lian, Haibin Ling
CodePose EstimationComputational EfficiencyImage
🎯 What it does: Proposed a 3D keypoint localization method based on visibility-driven importance weights, and constructed the VAPO framework to enhance the accuracy of 6DoF pose estimation.
VDTF-ACT: ACT-based Multimodal Space Fine Manipulation Method with Visual Depth Tactile Fusion
S. Lang, Zhiqiang Ma
CodeRobotic IntelligenceTransformerMultimodality
🎯 What it does: Studied a multimodal perception-based space micro-manipulation method to enhance the precise operation performance of satellite robots in low-gravity environments.
🎯 What it does: Constructed the VERAGMIL framework, integrating a high-fidelity simulator and VR interface, to record human demonstrations and train a robot-assisted feeding system for handling tasks involving granular foods such as rice and beans; evaluated imitation learning models including BC, BC-RNN, and BCQ;
View-aware Decomposition and Unification for Fast Ground-to-Aerial Person Search
Qifei Wang, Yongsheng Gao
CodeRetrievalContrastive LearningImage
🎯 What it does: Proposes a perspective-aware decomposition and unification (VADU) framework to model perspective differences in ground-to-air person search.
🎯 What it does: Propose ViewActive, a lightweight active viewpoint optimization method based on a single image, which guides robots to acquire the optimal observation angle through the 3D Viewpoint Quality Field (VQF).
VINS-MLD2: Monocular Visual-Inertial SLAM With Multi-level Detector and Descriptor
X. Nian, Yong Chen
CodeConvolutional Neural NetworkSimultaneous Localization and MappingOptical FlowImageMultimodality
🎯 What it does: Proposes a visual-inertial SLAM system called VINS-MLD2 based on multi-level detectors and descriptors, designs an efficient deep feature extraction network, and introduces matching fusion and adaptive matching strategies.
VISO-Grasp: Vision-Language Informed Spatial Object-centric 6-DoF Active View Planning and Grasping in Clutter and Invisibility
Yitian Shi, R. Rayyes
CodePose EstimationRepresentation LearningRobotic IntelligenceTransformerVision Language ModelMultimodality
🎯 What it does: Designed a vision-language system called VISO-Grasp, which utilizes foundation models for spatial reasoning and active view planning, constructs instance-centric spatial relationship representations, and achieves 6-DoF grasping in heavily occluded environments through a multi-view uncertainty-driven grasping fusion mechanism;
Visual Localization with Offline Google Satellite Map-Assisted for Ground Vehicles in GNSS-Denied Environment
Jibo Wang, Zheng Fang
CodePose EstimationAutonomous DrivingSimultaneous Localization and MappingImage
🎯 What it does: Propose a visual localization framework that utilizes offline Google satellite maps to address vehicle localization challenges in environments where GNSS signals are weak or unavailable; introduce a learning-based ground-satellite map feature matching method, propose a cross-perspective pose selection approach to construct two pose uncertainty models, and integrate them with classical SLAM methods;
VLIN-RL: A Unified Vision-Language Interpreter and Reinforcement Learning Motion Planner Framework for Robot Dynamic Tasks
Zewu Jiang, Chenyi Si
CodeRobotic IntelligenceReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Proposed the VLIN-RL framework, integrating the Vision-Language Interpreter (VLIN) with a reinforcement learning-based motion planner to enable real-time adjustment of subtasks based on visual feedback during task execution, thereby enhancing the real-time performance and robustness of robotic dynamic tasks.
WFDA: Wavelet-Based Frequency Decomposition and Aggregation for Underwater Object Detection
Xueting Liu, Shuxiang Guo
CodeObject DetectionImage
🎯 What it does: Propose a Wavelet-Based Frequency Decomposition and Aggregation Network (WFDA) for underwater target detection, utilizing wavelet transforms for feature frequency decomposition and aggregation.
🎯 What it does: Proposed a large-scale V2X dataset called WHALES and designed a scheduling algorithm named CAHS based on historical perspective coverage
YOLO-MARL: You Only LLM Once for Multi-Agent Reinforcement Learning
Zhuang Yuan, Fei Miao
CodeLarge Language ModelReinforcement Learning
🎯 What it does: Propose the YOLO-MARL framework, which utilizes a large language model (LLM) to generate strategies, state explanations, and planning functions in one interaction, followed by training decentralized multi-agent policies.
🎯 What it does: This paper proposes a zero-shot domain adaptation semantic segmentation method called SDGPA, which generates synthetic training images in the target domain style using text descriptions and performs progressive adaptation.