arXivSub Start free trial

ICRA 2025 Papers with Code β€” Page 2

IEEE International Conference on Robotics and Automation Β· 182 papers

Monocular Visual Place Recognition in LiDAR Maps via Cross-Modal State Space Model and Multi-View Matching

Gongxin Yao, Yu Pan

CodeRetrievalAutonomous DrivingComputational EfficiencyContrastive LearningSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Propose an efficient framework that utilizes a monocular camera to achieve cross-modal visual-point cloud retrieval in a pre-built LiDAR map, enabling monocular localization and reducing the computational burden of visual SLAM.

MonoDiff9D: Monocular Category-Level 9D Object Pose Estimation via Diffusion Model

Jian Liu, A. Mian

CodePose EstimationDepth EstimationTransformerDiffusion modelContrastive LearningImagePoint Cloud

🎯 What it does: Proposes a monocular category-level 9D object pose estimation method called MonoDiff9D based on diffusion models, utilizing zero-shot depth estimation and point cloud feature fusion, and restoring the pose through a transformer denoiser.

MonoLDP: LED Assisted Indoor Mobile Bot Monocular Depth Prediction and Pose Estimation System

Chenxin Liang, Wenbo Ding

CodePose EstimationDepth EstimationRobotic IntelligenceImage

🎯 What it does: Built and evaluated a monocular camera-based indoor mobile robot system called MonoLDP, integrating depth estimation, pose estimation, and visible light communication functions

Motion-Guided Dual-Camera Tracker for Endoscope Tracking and Motion Analysis in a Mechanical Gastric Simulator

Yuelin Zhang, S. Cheng

CodeObject TrackingTransformerBiomedical Data

🎯 What it does: Propose a dual-camera motion-guided visual tracker for 3D localization and motion analysis of the flexible endoscope tip in a mechanical stomach simulator.

Multi-Goal Motion Memory

Yuanjie Lu, Xuesu Xiao

CodeOptimizationRobotic Intelligence

🎯 What it does: Developed a multi-object motion memory technology to accelerate multi-target path planning for autonomous mobile robots in dynamic environments

Multi-Layered Safety of Redundant Robot Manipulators Via Task-Oriented Planning and Control

Xinyu Jia, Haoyong Yu

CodeRobotic Intelligence

🎯 What it does: Proposes a task-oriented planning and control framework for redundant robotic manipulators to achieve multi-layer safety while maintaining task execution efficiency;

Multi-Type Preference Learning: Empowering Preference-Based Reinforcement Learning with Equal Preferences

Ziang Liu, Liang He

CodeReinforcement Learning from Human FeedbackReinforcement Learning

🎯 What it does: Proposes a multi-type preference learning (MTPL) method that simultaneously utilizes equal preference and explicit preference for training in preference reinforcement learning.

NaviDiffusor: Cost-Guided Diffusion Model for Visual Navigation

Yiming Zeng, Hui Cheng

CodeAutonomous DrivingDiffusion modelImage

🎯 What it does: Propose an RGB visual navigation framework that integrates learning and classical methods, first training a conditional diffusion model and then using differentiable cost gradients to guide path generation.

Neural Lyapunov Function Approximation with Self-Supervised Reinforcement Learning

Lucy McCutcheon, Saber Fallah

CodeOptimizationRobotic IntelligenceReinforcement LearningWorld ModelSequential

🎯 What it does: Proposes a method for neural network approximation of nonlinear Lyapunov functions using self-supervised reinforcement learning and data-driven world models.

Neural Surface Reconstruction and Rendering for LiDAR-Visual Systems

Jianheng Liu, Fu Zhang

CodeGenerationNeural Radiance FieldImagePoint Cloud

🎯 What it does: Propose a unified surface reconstruction and rendering framework that integrates NeRF and NDF to recover appearance and structural information from pose images and point clouds.

OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy Prediction

Severin Heidrich, Lutz Eckstein

CodeAutonomous Driving

🎯 What it does: An efficient uncertainty quantification method is proposed for the 3D occupancy prediction task, evaluated under different camera failure scenarios.

Open3DTrack: Towards Open-Vocabulary 3D Multi-Object Tracking

Ayesha Ishaq, R. Anwer

CodeObject Tracking

🎯 What it does: Propose an open vocabulary 3D multi-object tracking framework

OPRNet: Object-Centric Point Reconstruction Network for Multimodal 3D Object Detection in Adverse Weathers

Jae Hyun Yoon, S. Yoo

CodeObject DetectionMultimodalityPoint Cloud

🎯 What it does: Propose a point reconstruction network based on isometric projection for multi-modal 3D object detection, which includes a range-constrained noise filter, an object-oriented point generator, and a dual 2D auxiliary module.

Optimizing Underwater Robot Navigation: A Study of DRL Algorithms and Multi-Modal Sensor Fusion

Md. Ether Deowan, Ricard Marxer

CodeDepth EstimationRobotic IntelligenceReinforcement LearningMultimodality

🎯 What it does: This paper studies the integration of multiple reinforcement learning algorithms (PPO, TRPO, SAC, TD3, A2C) with multi-modal sensor fusion to enhance the autonomous navigation capabilities of low-cost underwater robots in complex environments.

Physics-Aware Robotic Palletization With Online Masking Inference

Tianqi Zhang, Wei Zhan

CodeRobotic IntelligenceReinforcement LearningPhysics Related

🎯 What it does: Proposed a box stacking planning method based on reinforcement learning, using online action space mask inference to guide the policy in selecting effective actions.

PoLaRIS Dataset: A Maritime Object Detection and Tracking Dataset in Pohang Canal

Jiwon Choi, Younggun Cho

CodeObject DetectionObject TrackingMultimodalityBenchmark

🎯 What it does: Proposed and constructed a new multimodal maritime obstacle detection and tracking dataset called PoLaRIS, and evaluated it using various state-of-the-art methods.

Polyp-Gen: Realistic and Diverse Polyp Image Generation for Endoscopic Dataset Expansion

Shengyuan Liu, Yixuan Yuan

CodeGenerationData SynthesisDiffusion modelImageBiomedical Data

🎯 What it does: Propose anε…¨θ‡ͺ动 diffusion-based generation framework named Polyp-Gen to generate realistic and diverse polyp endoscopic images for dataset expansion.

Promi: an Efficient Prototype-Mixture Baseline for Few-Shot Segmentation with Bounding-Box Annotations

Florent Chiaroni, Ola Ahmad

CodeSegmentationMixture of ExpertsImage

🎯 What it does: Propose a few-shot binary segmentation method ProMi based on bounding box annotations, utilizing a prototype-mixture model to handle background classes

ProxFly: Robust Control for Close Proximity Quadcopter Flight Via Residual Reinforcement Learning

Ruiqi Zhang, Mark W. Mueller

CodeRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed the ProxFly residual deep reinforcement learning controller to compensate for downwash effects and enhance quadrotor flight stability during close-range flights.

PTQ4RIS: Post-Training Quantization for Referring Image Segmentation

Xiaoyan Jiang, Sifan Zhou

CodeSegmentationVision Language ModelMultimodality

🎯 What it does: This paper proposes a post-training quantization framework called PTQ4RIS specifically for referential image segmentation tasks and analyzes the root causes of performance degradation during the quantization process.

PTZ-Calib: Robust Pan-Tilt-Zoom Camera Calibration

Jinhui Guo, Jieping Ye

CodePose EstimationComputational Efficiency

🎯 What it does: Proposes a two-stage PTZ camera calibration method called PTZ-Calib, which can efficiently and accurately estimate camera parameters from any perspective.

PUGS: Zero-Shot Physical Understanding with Gaussian Splatting

Yinghao Shuai, Hao Zhao

CodeRepresentation LearningContrastive LearningGaussian SplattingBenchmarkPhysics Related

🎯 What it does: Reconstruct 3D objects using Gaussian splatting and predict their physical properties such as mass and friction in a zero-shot scenario.

Radar Teach and Repeat: Architecture and Initial Field Testing

Xinyuan Qiao, Tim D. Barfoot

CodeAutonomous DrivingRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Designed and verified a full-stack radar system, Radar Teach and Repeat (RT&R), achieving GPS-free offline long-distance autonomous robot path tracking;

Radar4VoxMap: Accurate Odometry from Blurred Radar Observations

Ji-Kwon Seok, Kichun Jo

CodeAutonomous DrivingOptimizationSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed the Radar4VoxMap method to enhance the accuracy of odometry using only radar.

RadarMask: A Novel End-to-End Sparse Millimeter-Wave Radar Sequence Panoptic Segmentation and Tracking Method

Yubo Guo, Qiang Gao

CodeObject TrackingSegmentationPoint Cloud

🎯 What it does: Proposed an end-to-end method for panoptic segmentation and tracking of sparse millimeter-wave radar sequences, RadarMask, which can achieve multi-level semantic and instance descriptions in the radar data domain;

Rapid Dynamic Obstacle Avoidance for UAVs Enhanced by DVS and Neuromorphic Computing

Siyang Wang, Pengju Ren

CodeComputational EfficiencyRobotic IntelligenceSpiking Neural Network

🎯 What it does: Developed an end-to-end obstacle avoidance algorithm for UAVs based on monocular dynamic visual sensors

Real-time Deformation-aware Control for Autonomous Robotic Subretinal Injection under iOCT Guidance

Demir Arikan, I. Iordachita

CodeRobotic IntelligenceBiomedical Data

🎯 What it does: Proposed an autonomous subretinal injection method under iOCT guidance with real-time deformation awareness.

Real-Time LiDAR Point Cloud Compression and Transmission for Resource-Constrained Robots

Yuhao Cao, Haoyao Chen

CodeCompressionPoint Cloud

🎯 What it does: Proposed a real-time LiDAR point cloud compression and transmission framework named RCPCC.

Realm: Real-Time Line-of-Sight Maintenance in Multi-Robot Navigation with Unknown Obstacles

Ruofei Bai, Lihua Xie

CodeRobotic IntelligencePoint Cloud

🎯 What it does: Proposes a multi-robot navigation method based on real-time LiDAR scans with Line-of-Sight (LoS) constraints, capable of deriving LoS constraints in real-time through point cloud visibility analysis in unknown environments;

RLCNet: A Novel Deep Feature-Matching-Based Method for Online Target-Free Radar-LiDAR Calibration

Kai-Rui Luan, Huimin Lu

CodePose EstimationAutonomous DrivingPoint Cloud

🎯 What it does: Propose an online target-free radar-LiDAR extrinsic calibration method based on deep feature matching, formulating the calibration problem as a cross-modal point cloud registration task, first performing keypoint matching and then refining with dense matching.

RM-Planner: Integrating Reinforcement Learning with Whole-Body Model Predictive Control for Mobile Manipulation

Zixuan Zhuang, Hui Cheng

CodeRobotic IntelligenceReinforcement LearningPoint Cloud

🎯 What it does: Propose a planning method for mobile manipulation in unknown complex environments, integrating reinforcement learning with full-body MPC.

RMP-YOLO: A Robust Motion Predictor for Partially Observable Scenarios Even if You Only Look Once

Jiawei Sun, Marcelo H. Ang

CodeAutonomous DrivingTransformerBenchmark

🎯 What it does: Propose the RMP-YOLO unified framework, which achieves robust motion prediction in partially observable scenarios by first reconstructing complete historical trajectories.

RMSeg-UDA: Unsupervised Domain Adaptation for Road Marking Segmentation Under Adverse Conditions

Yi-Chang Cai, Chiao-Tung Chan

CodeSegmentationDomain AdaptationAutonomous DrivingGenerative Adversarial NetworkImage

🎯 What it does: Proposed an unsupervised domain adaptation framework for road sign segmentation called RMSeg-UDA, combining schedule self-training and class-conditioned adversarial training, leveraging labeled normal weather data and unlabeled data from other domains to train the model.

Robo-DM: Data Management for Large Robot Datasets

Kai-Peng Chen, Kenneth Y. Goldberg

CodeCompressionData-Centric LearningRobotic IntelligenceTransformerVideo

🎯 What it does: Proposed Robo-DM, a cloud-based, efficient open-source data management tool for collecting, sharing, and learning robot data, which stores data in EBML format and significantly reduces data volume, transmission cost, and loading time through compression and memory-mapped decoding cache.

Robotic 3D Flower Pose Estimation for Small-Scale Urban Farms

Harsh Muriki, Ai-Ping Hu

CodePose EstimationRobotic IntelligenceConvolutional Neural NetworkImagePoint CloudAgriculture Related

🎯 What it does: Using FarmBot and a custom camera to acquire 3D point clouds and estimate the pose of strawberry plant flowers; by translating the occupied grid along orthogonal axes to generate six perspectives of 2D images, utilizing 2D object detection to extract flower point clouds, and then fitting three shapes (superellipse, paraboloid, and plane) to determine the pose.

Robust Scene Change Detection Using Visual Foundation Models and Cross-Attention Mechanisms

Chun-Jung Lin, Feras Dayoub

CodeSegmentationRepresentation LearningTransformerImage

🎯 What it does: Propose a scene change detection method based on the visual foundation model DINOv2 and the full-image cross-attention mechanism.

Robust Self-Reconfiguration for Fault-Tolerant Control of Modular Aerial Robot Systems

Rui Huang, Lin Zhao

CodeOptimizationRobotic Intelligence

🎯 What it does: A robust and efficient self-reconfiguration algorithm for modular aerial robotic systems (MARS) is proposed, which maximizes the controllability margin in each intermediate phase and calculates the optimal disassembly/assembly sequence, verifying its effectiveness in various fault-tolerant self-reconfiguration scenarios.

ROD: RGB-Only Fast and Efficient Off-Road Freespace Detection

Tong Sun, Yu Hu

CodeSegmentationAutonomous DrivingTransformerImage

🎯 What it does: Proposed a novel method called ROD for detecting free space in off-road scenarios using only RGB images.

SANDRO: A Robust Solver with a Splitting Strategy for Point Cloud Registration

Michael Adlerstein, Claudio Semini

CodePose EstimationOptimizationPoint Cloud

🎯 What it does: Proposed a point cloud registration algorithm named SANDRO, combining the IRLS framework with a progressive non-convex robust loss function, and incorporating a splitting strategy to handle high outlier rates and skewed distributions of outliers.

SCA3D: Enhancing Cross-Modal 3D Retrieval via 3D Shape and Caption Paired Data Augmentation

Junlong Ren, Hao Wang

CodeData SynthesisRetrievalLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningTextMultimodalityPoint CloudMesh

🎯 What it does: Proposes the SCA3D method, which uses LLaVA to generate captions for segmented 3D shape components and performs data augmentation to create new 3D-text pairs, thereby enhancing cross-modal 3D retrieval performance.

Scenario-Based Curriculum Generation for Multi-Agent Driving

Axel Brunnbauer, R. Grosu

CodeAutonomous DrivingReinforcement Learning

🎯 What it does: Created the MATS-Gym framework, which generates variable-number multi-agent traffic scenarios using partial scene specifications and executes them in CARLA. It integrates Scenic and ScenarioRunner with a multi-agent training framework, models interactions using partially observable stochastic games, and combines unsupervised environment design to achieve adaptive auto-curriculum.

Scene-Aware Explainable Multimodal Trajectory Prediction

Pei Liu, Jun Ma

CodeAutonomous DrivingExplainability and InterpretabilityDiffusion modelMultimodality

🎯 What it does: Proposed an explainable conditional diffusion multi-modal trajectory prediction model DMTP to reveal environmental factors and internal mechanisms affecting predictions.

Self-Mixing Laser Interferometry for Robotic Tactile Sensing

R. Proesmans, Francis Wyffels

CodeRobotic IntelligencePhysics Related

🎯 What it does: Designed and verified a robotic fingertip using self-mixing interference (SMI) technology for detecting object sliding and external contact.

SLABIM: A SLAM-BIM Coupled Dataset in HKUST Main Building

Haoming Huang, Huan Yin

CodeSimultaneous Localization and MappingMultimodalityBenchmark

🎯 What it does: This paper designs and constructs the first dataset combining SLAM and BIM, named SLABIM, collects multi-sensor and multi-scenario data and builds models at the main building of the Hong Kong University of Science and Technology, followed by experimental verification on three tasks: registration, localization, and semantic mapping.

Slamspoof: Practical Lidar Spoofing Attacks on Localization Systems Guided by Scan Matching Vulnerability Analysis

Rokuto Nagata, Kentaro Yoshioka

CodeAutonomous DrivingAdversarial AttackSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Designed and implemented SLAMSpoof, demonstrating LiDAR spoofing attacks against autonomous driving positioning systems, and evaluated their impact on positioning accuracy.

Space-Aware Instruction Tuning: Dataset and Benchmark for Guide Dog Robots Assisting the Visually Impaired

ByungOk Han, Jaehong Kim

CodeRobotic IntelligenceSupervised Fine-TuningVision Language ModelMultimodalityBenchmark

🎯 What it does: Proposed the Space-Aware Instruction Tuning (SAIT) dataset and the Space-Aware Benchmark (SA-Bench), and performed spatial-aware instruction tuning on Vision-Language Models (VLMs) to improve navigation guidance for visually impaired individuals by blind navigation robots.

SpatialBot: Precise Spatial Understanding with Vision Language Models

Wenxiao Cai, Bo Zhao

CodeVision Language ModelMultimodalityBenchmark

🎯 What it does: Designed and trained the SpatialBot model, leveraging RGB and depth images to enhance the spatial understanding capabilities of Vision-Language Models (VLMs).

SSF: Sparse Long-Range Scene Flow for Autonomous Driving

Ajinkya Khoche, P. Jensfelt

CodeAutonomous DrivingConvolutional Neural NetworkPoint Cloud

🎯 What it does: Propose the Sparse Scene Flow (SSF) approach, utilizing a sparse convolution backbone to achieve long-range scene flow estimation.

Stage-Wise Reward Shaping for Acrobatic Robots: A Constrained Multi-Objective Reinforcement Learning Approach

Dohyeong Kim (Seoul National University), Songhwai Oh (Seoul National University)

CodeRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose a reward and cost function divided into stages, shaping rewards and controlling acrobatic robots under a constrained multi-objective reinforcement learning framework, implementing a corresponding practical algorithm, and verifying its effectiveness in both simulated and real environments.

Steering Prediction via a Multi-Sensor System for Autonomous Racing

Zhuyun Zhou, Tobi Delbruck

CodeAutonomous DrivingMultimodalityPoint Cloud

🎯 What it does: Integrate 2D LiDAR and event camera data within an end-to-end learning framework for steering prediction, create a multi-sensor dataset, benchmark SOTA fusion methods, and propose a low-rank efficient fusion design along with a novel fusion learning strategy.

Stonefish: Supporting Machine Learning Research in Marine Robotics

Michele Grimaldi, Nuno Gracias

CodeRobotic Intelligence

🎯 What it does: Multiple new sensors (event camera, thermal camera, optical flow camera) as well as visual optical communication, towed equipment support, improved thruster modeling, more flexible fluid dynamics models, and more accurate sonar simulations were added to the open-source simulation platform Stonefish, along with an automatic annotation tool.

Structure-Aware Radar-Camera Depth Estimation

Fuyi Zhang, Hui-liang Shen

CodeDepth EstimationAutonomous DrivingImageMultimodalityPoint Cloud

🎯 What it does: Proposed a structure-based radar depth enhancement strategy and a multi-scale structure-guided network, and constructed the structure-aware radar-camera depth estimation framework SA-RCD

Talk2Radar: Bridging Natural Language with 4D mmWave Radar for 3D Referring Expression Comprehension

Runwei Guan, Hui Xiong

CodeRecognitionConvolutional Neural NetworkGraph Neural NetworkTextMultimodalityPoint Cloud

🎯 What it does: Constructed the first Talk2Radar dataset, integrating 4D mmWave radar point clouds with natural language prompts for 3D gesture expression understanding, and proposed the T-RadarNet model.

TANGO: Traversability-Aware Navigation with Local Metric Control for Topological Goals

Stefan Podgorski, Ian Reid

CodeDepth EstimationRobotic IntelligenceImage

🎯 What it does: Propose a topology-metric navigation pipeline based on RGB and object-level targets, achieving zero-shot long-term robot navigation without requiring 3D maps or pre-trained controllers.

TCAFF: Temporal Consistency for Robot Frame Alignment

Mason B. Peterson, Jonathan P. How

CodeRobotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: A multi-hypothesis algorithm named TCAFF was developed for aligning coordinate frames of neighboring robots, and its effectiveness was verified in an experiment involving four robots collaboratively tracking six pedestrians.

ThermoStereoRT: Thermal Stereo Matching in Real Time via Knowledge Distillation and Attention-Based Refinement

Anning Hu, Danping Zou

CodeDepth EstimationKnowledge DistillationConvolutional Neural NetworkImage

🎯 What it does: Designed and implemented ThermoStereoRT, a real-time thermal imaging stereo matching method that constructs a 3D cost volume using a lightweight backbone network, generates initial disparity maps through a multi-scale attention mechanism, and refines them using channel and spatial attention modules; improves performance on sparsely annotated thermal imaging data via knowledge distillation while maintaining real-time speed; applicable to all-weather scenarios such as nighttime drone surveillance or bed-under cleaning robots.

Topology-Based Visual Active Room Segmentation

Chenyu Bao, T. Lam

CodeSegmentationSimultaneous Localization and MappingImage

🎯 What it does: This paper proposes a proactive room segmentation framework that can incrementally and autonomously complete room segmentation in cluttered indoor environments.

Toward Zero-Shot Learning for Visual Dehazing of Urological Surgical Robots

Renkai Wu, Hao Tang

CodeRestorationRobotic IntelligenceImageBiomedical DataBenchmark

🎯 What it does: Proposed an unsupervised zero-shot dehazing method named RSF-Dehaze, and created the USRobot-Dehaze dataset for urological robotic vision.

Towards Autonomous Data Annotation and System-Agnostic Robotic Grasping Benchmarking with 3D-Printed Fixtures

W. Boerdijk, Rudolph Triebel

CodePose EstimationRobotic IntelligenceSupervised Fine-TuningImageBenchmark

🎯 What it does: This paper proposes the use of 3D-printed fixtures designed for any rigid object to temporarily fix objects in the scene, extract their poses, and then remove the fixtures to maintain a natural scene and achieve automated data annotation.

Towards Robust Autonomous Driving: Conditional Multimodal Large Language Models for Fine-Grained Perception

Fengzhao Sun, Fang Gao

CodeAutonomous DrivingLarge Language ModelVision Language ModelMultimodality

🎯 What it does: Proposes Percept-DriveLM, a multimodal large language model designed for fine-grained perception tasks in autonomous driving, and develops a core visual fusion module;

Towards Survivability in Complex Motion Scenarios: RGB-Event Object Tracking via Historical Trajectory Prompting

Wenhao Xia, Xu Jia

CodeObject TrackingPrompt EngineeringMultimodalityBenchmark

🎯 What it does: Propose the EventTPT framework, leveraging key hints from historical trajectories to enhance RGB-Event object tracking performance.

Tracking Everything in Robotic-Assisted Surgery

Bohan Zhan, Baoru Huang

CodeObject TrackingRobotic IntelligenceVideoBiomedical DataBenchmark

🎯 What it does: A specialized video tracking dataset for robotic-assisted surgery was constructed, existing Tracking Any Point (TAP) methods were evaluated on this dataset, and a new tracking algorithm named SurgMotion was proposed to enhance tracking performance.

TrackOcc: Camera-Based 4D Panoptic Occupancy Tracking

Zhuoguang Chen, Hang Zhao

CodeObject TrackingSegmentationAutonomous DrivingVideo

🎯 What it does: Proposed a 4D panoramic occupancy tracking task based on cameras, and introduced the end-to-end TrackOcc method;

Transferring Visual Knowledge: Semi-Supervised Instance Segmentation for Object Navigation Across Varying Height Viewpoints

Qiu Zheng, T. Lam

CodeSegmentationDomain AdaptationContrastive LearningImage

🎯 What it does: Propose a semi-supervised instance segmentation method for target navigation under different camera height perspectives, reducing the need for additional annotations by transferring knowledge from the source height to the target height.

TranSplat: Surface Embedding-Guided 3D Gaussian Splatting for Transparent Object Manipulation

Jeong-Man Kim, Ayoung Kim

CodeDepth EstimationDiffusion modelGaussian SplattingImageBenchmark

🎯 What it does: Propose a surface embedding guided 3D Gaussian splat (TranSplat) method for transparent objects to achieve accurate and dense depth completion.

TriHRCBot: A Robotic Architecture for Triadic Human-Robot Collaboration Through Mediated Object Alignment

Francesco Semeraro, Angelo Cangelosi

CodeRobotic Intelligence

🎯 What it does: This paper proposes the TriHRCBot architecture, enabling two parallel users to simultaneously manipulate a target object through robot perception and pose adjustment.

TS-DETR: Traffic Sign Detection Based on Positive and Negative Sample Augmentation

Ching-Lung Lin, Chieh-Chih Wang

CodeObject DetectionTransformerImage

🎯 What it does: Proposes an end-to-end traffic sign detection framework based on DETR, and introduces positive and negative sample enhancement along with the UASPP module to improve recognition performance.

TSCLIP: Robust CLIP Fine-Tuning for Worldwide Cross-Regional Traffic Sign Recognition

Guoyang Zhao, Jun Ma

CodeRecognitionSupervised Fine-TuningPrompt EngineeringVision Language ModelImageBenchmark

🎯 What it does: Propose TSCLIP, a robust fine-tuning method based on CLIP for global cross-regional traffic sign recognition, and construct a cross-regional traffic sign benchmark dataset, adopting prompt engineering schemes tailored to traffic sign characteristics and Adaptive Dynamic Weight Ensembling (ADWE) technology.

UA-PnP: Uncertainty-Aware End-to-End Bird's Eye View Visual Perception and Prediction for Autonomous Driving

Zijian Huang, Qi Hao

CodeAutonomous DrivingTransformer

🎯 What it does: Propose an uncertainty-aware, end-to-end visual perception and prediction framework based on bird's eye view (BEV), comprising a feature distribution estimation network, an uncertainty-based Transformer, and a generator for future instance segmentation with uncertainty evidence decoder.

ULOC: Learning to Localize in Complex Large-Scale Environments with Ultra-Wideband Ranges

Thien-Minh Nguyen, Lihua Xie

CodeAutonomous DrivingTransformerSimultaneous Localization and Mapping

🎯 What it does: Proposed the ULOC learning framework for positioning in large-scale, complex environments using UWB. First, deploy anchor points with unknown positions, collect UWB observations during vehicle movement, and use VIO, LIO, and other SLAM methods to align with a prior map, generating map-consistent pose estimates as training labels. Subsequently, a MAMBA-based network learns UWB ranging patterns.

UltraFastCrackSeg: A Lightweight Real-Time Crack Segmentation Model with Task-Oriented Pretraining

Weiqing Qi, Yang Yang

CodeSegmentationComputational EfficiencyConvolutional Neural NetworkImage

🎯 What it does: Proposed the UltraFastCrackSeg lightweight real-time crack segmentation model

UncAD: Towards Safe End-to-end Autonomous Driving via Online Map Uncertainty

Pengxuan Yang, Dongbin Zhao

CodeAutonomous DrivingSimultaneous Localization and MappingMultimodalityPoint Cloud

🎯 What it does: Developed the UncAD system to estimate online map uncertainty and use it for motion prediction and planning, proposing a trajectory selection strategy based on uncertainty-aware collision perception.

Uncertainty-Aware Probabilistic Risk Quantification of SOTIF for Autonomous Vehicles

Botao Yao, Shaoming Duan

CodeAutonomous Driving

🎯 What it does: Proposed an uncertainty-aware probabilistic risk assessment framework for safety-of-the-intended-functionality (SOTIF) in autonomous vehicles (AVs), to quantify the risk of AVs violating safety constraints and calculate the expected average severity of such violations.

Unified Human Localization and Trajectory Prediction with Monocular Vision

Po-Chien Luan, Alexandre Alahi

CodePose EstimationAutonomous DrivingTransformerSimultaneous Localization and MappingImage

🎯 What it does: Propose MonoTransmotion (MT), a Transformer-based framework that utilizes a monocular camera to simultaneously achieve human BEV localization and trajectory prediction.

Universal Online Temporal Calibration for Optimization-Based Visual-Inertial Navigation Systems

Yunfei Fan, Feng Zhou

CodeAutonomous DrivingOptimizationSimultaneous Localization and MappingVideoMultimodalityTime Series

🎯 What it does: Proposed a general online time calibration strategy that incorporates time offset as a state parameter into an optimized residual model to achieve time synchronization in visual inertial navigation systems.

Unleashing HyDRa: Hybrid Fusion, Depth Consistency and Radar for Unified 3D Perception

Philipp Wolters, Gerhard Rigoll

CodeDepth EstimationAutonomous DrivingTransformerImagePoint CloudBenchmark

🎯 What it does: Proposes the HyDRa camera-radar fusion architecture for low-cost, vision-oriented 3D perception tasks.

UR-MVO: Robust Monocular Visual Odometry for Underwater Scenarios

Zein Alabedeen Barhoum, S. Kolyubin

CodeConvolutional Neural NetworkTransformerSupervised Fine-TuningSimultaneous Localization and MappingImage

🎯 What it does: Proposed a robust monocular visual odometry (UR-MVO) pipeline for underwater scenarios, using SuperPoint for feature extraction and SuperGlue for matching, while employing few-shot unsupervised learning to fine-tune SuperPoint for scene-specific adaptation. Additionally, a semantic segmentation model trained on underwater images is integrated to remove unreliable features from dynamic objects and the background.

V2X-DGW: Domain Generalization for Multi-Agent Perception Under Adverse Weather Conditions

Baolu Li, Hongkai Yu

CodeObject DetectionDomain AdaptationAutonomous DrivingContrastive LearningPoint Cloud

🎯 What it does: Propose a V2X-DGW method based on domain generalization for LiDAR multi-agent 3D object detection under adverse weather conditions while maintaining high performance in clear weather

VertiCoder: Self-Supervised Kinodynamic Representation Learning on Vertically Challenging Terrain

Mohammad Nazeri, Xuesu Xiao

CodeRepresentation LearningRobotic IntelligenceTransformerAuto Encoder

🎯 What it does: Proposes VertiCoder, a self-supervised representation learning method for vertical challenging terrain robot mobility, capable of simultaneously accomplishing four downstream tasks with a single representation: forward dynamics learning, inverse dynamics learning, behavior cloning, and patch reconstruction.

VisFly: An Efficient and Versatile Simulator for Training Vision-Based Flight

Fanxing Li, Danping Zou

CodeAutonomous DrivingComputational EfficiencyReinforcement Learning

🎯 What it does: Developed VisFly, a highly efficient simulator for training vision-based quadrotor flight control strategies.

vMF-Contact: Uncertainty-Aware Evidential Learning for Probabilistic Contact-Grasp in Noisy Clutter

Yitian Shi, R. Rayyes

CodeRobotic Intelligence

🎯 What it does: Proposes a noise-free evidence learning method called vMF-Contact for 6-DoF grasping, capable of simultaneously capturing aleatoric uncertainty from data noise and epistemic uncertainty in out-of-distribution (OOD) recognition;

Wcdt: World-Centric Diffusion Transformer for Traffic Scene Generation

Chen Yang, Arsalan Heydarian

CodeGenerationData SynthesisAutonomous DrivingTransformerDiffusion modelMultimodality

🎯 What it does: Proposes a World-Centered Diffusion Transformer (WcDT) framework for autonomous driving trajectory generation, combining diffusion probability models and transformers to optimize the entire generation process.