arXivSub Start free trial

IROS 2024 Papers with Code β€” Page 2

IEEE/RSJ International Conference on Intelligent Robots and Systems Β· 156 papers

Progressive Representation Learning for Real-Time UAV Tracking

Changhong Fu, Jia Pan

CodeObject TrackingRepresentation LearningVideo

🎯 What it does: Proposed an advanced representation learning framework called PRL-Track for real-time visual object tracking in unmanned aerial vehicles (UAVs).

Prompt-Driven Temporal Domain Adaptation for Nighttime UAV Tracking

Changhong Fu, Jia Pan

CodeObject TrackingDomain AdaptationPrompt EngineeringGenerative Adversarial NetworkVideoBenchmark

🎯 What it does: Proposed a prompt-based temporal domain adaptation training framework for nighttime UAV tracking and implemented the TDA-Track tracker.

Proprioception Is All You Need: Terrain Classification for Boreal Forests

D. LaRocque, Franccois Pomerleau

CodeClassificationConvolutional Neural NetworkTime Series

🎯 What it does: Propose the BorealTC dataset and evaluate the performance of CNN and Mamba on terrain classification tasks

RaceMOP: Mapless Online Path Planning for Multi-Agent Autonomous Racing using Residual Policy Learning

Raphael Trumpp, Marco Caccamo

CodeAutonomous DrivingReinforcement Learning

🎯 What it does: Proposed a mapless online path planning method called RaceMOP, which utilizes residual policy learning to achieve high-speed overtaking decisions for multi-vehicle F1TENTH racing.

Raising Body Ownership in End-to-End Visuomotor Policy Learning via Robot-Centric Pooling

Zheyu Zhuang, D. Kragic

CodeRobotic IntelligenceContrastive LearningImage

🎯 What it does: Proposed a Robot-Centric Pooling (RcP) method to enhance the performance of end-to-end visual motion policies.

RATE: Real-time Asynchronous Feature Tracking with Event Cameras

Mikihiro Ikura, W. StΓΌrzl

CodeObject TrackingOptical Flow

🎯 What it does: Proposed a real-time asynchronous event camera feature tracking pipeline called RATE

Real-Time Semantic Segmentation in Natural Environments with SAM-assisted Sim-to-Real Domain Transfer

Han Wang, L. Teixeira

CodeSegmentationDomain AdaptationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkTransformer

🎯 What it does: This paper proposes a two-step training pipeline for semantic segmentation in natural environments. First, a domain adaptation model is used for training, followed by pseudo label refinement using masks generated by the Segment Anything Model (SAM), ultimately distilling a MobileNetV3 model deployable in real-time.

Realistic Rainy Weather Simulation for LiDARs in CARLA Simulator

Donglin Yang, Xinyu Cai

CodeData SynthesisAutonomous DrivingPoint Cloud

🎯 What it does: Implement LiDAR point cloud simulation under rainy conditions, including raindrop spray and light intensity effects, using the CARLA simulator to generate synthetic rainy weather data for data augmentation.

Robot Shape and Location Retention in Video Generation Using Diffusion Models

Peng Wang, Minh Huy Pham

CodeGenerationRobotic IntelligenceConvolutional Neural NetworkDiffusion modelVideo

🎯 What it does: Proposed a Diffusion model specifically tailored for shape and position preservation in mobile robots, capable of generating videos that precisely retain the robot's morphology and positional information.

Robot Traversability Prediction: Towards Third-Person-View Extension of Walk2Map with Photometric and Physical Constraints

J. Liang, Kanji Tanaka

CodeRobotic IntelligenceSimultaneous Localization and MappingImage

🎯 What it does: Propose a third-person perspective-based robot traversability prediction method called Walk2Map++, which addresses visual uncertainty issues by integrating photometric constraints (occlusion ordering) and physical constraints (collision avoidance);

Robust Two-View Geometry Estimation with Implicit Differentiation

Vladislav A. Pyatov, Stamatios Lefkimmiatis

CodePose Estimation

🎯 What it does: Proposes a two-view geometry estimation framework based on differentiable robust loss function fitting, treating the robust fundamental matrix estimation as an implicit layer, and constructing an end-to-end trainable unified pipeline

Safe multi-agent reinforcement learning for bimanual dexterous manipulation

Weishu Zhan, Peter Chin

CodeSafty and PrivacyRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a multi-agent reinforcement learning algorithm called MAC-PAO for safe bimanual coordination, and conducted experiments on various tasks with safety constraints

Safe Offline-to-Online Multi-Agent Decision Transformer: A Safety Conscious Sequence Modeling Approach

A. Shah, Xin Fu

CodeTransformerReinforcement LearningSequentialBenchmark

🎯 What it does: Proposed a safety-aware offline-to-online multi-agent decision Transformer framework SO2-MADT, achieving safety-priority multi-agent reinforcement learning.

Safety-First Tracker: A Trajectory Planning Framework for Omnidirectional Robot Tracking

Yue Lin, Huchuan Lu

CodeOptimizationRobotic Intelligence

🎯 What it does: Proposes a safety-priority trajectory planning framework (SF-Tracker) for omnidirectional autonomous tracking robots, which separates robot position and orientation for step-by-step planning. First, it constructs a reference path that is independent of conflicts and occlusions, then performs safety trajectory optimization, and designs an orientation planner to ensure target visibility.

SCANet: Correcting LEGO Assembly Errors with Self-Correct Assembly Network

Yuxuan Wan, Hao Dong

CodeAnomaly DetectionContrastive LearningImage

🎯 What it does: Proposes a self-correcting assembly network, SCANet, for the task of identifying and correcting LEGO assembly errors

Scheduling of Robotic Cellular Manufacturing Systems with Timed Petri Nets and Reinforcement Learning

ZhuTao Yao, Shaohua Yu

CodeOptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a Q-learning scheduling method based on Petri nets for efficient scheduling in robot cell manufacturing systems.

SDGE: Stereo Guided Depth Estimation for 360Β°Camera Sets

Jialei Xu, Xianming Liu

CodeDepth EstimationSupervised Fine-TuningImageVideo

🎯 What it does: Propose a stereo-guided depth estimation method that leverages multi-view stereo results and self-calibrated camera poses to enhance the quality of full-image depth estimation for 360° camera sets.

Self Supervised Detection of Incorrect Human Demonstrations: A Path Toward Safe Imitation Learning by Robots in the Wild

Noushad Sojib, M. Begum

CodeSafty and PrivacyRobotic IntelligenceSequentialBenchmark

🎯 What it does: Created the Layman V1.0 dataset and proposed the Behavior Cloning for Error Detection (BED) framework to automatically detect and discard erroneous demonstrations, thereby enhancing the safety of imitation learning.

Self-Selecting Semi-Supervised Transformer-Attention Convolutional Network for Four Class EEG-Based Motor Imagery Decoding

Han Wei Ng, Cuntai Guan

CodeClassificationConvolutional Neural NetworkTransformerAuto EncoderBiomedical Data

🎯 What it does: Propose a multi-class EEG motor imagery classification method based on Variational Autoencoder and Transformer-based Attention Convolutional Network (SSTACNet).

Semantics from Space: Satellite-Guided Thermal Semantic Segmentation Annotation for Aerial Field Robots

Connor T. Lee, Soon-Jo Chung

CodeSegmentationTransformerSimultaneous Localization and MappingImageAgriculture Related

🎯 What it does: Automatically generate semantic segmentation annotations for aerial thermal imaging using satellite data and onboard positioning information.

SiCP: Simultaneous Individual and Cooperative Perception for 3D Object Detection in Connected and Automated Vehicles

Deyuan Qu, Qing Yang

CodeObject DetectionAutonomous DrivingPoint Cloud

🎯 What it does: Proposes the Simultaneous Individual and Cooperative Perception (SiCP) framework, and employs a lightweight Dual-Perception Network (DP-Net) to simultaneously achieve individual and collaborative 3D object detection.

Similarity Distance-Based Label Assignment for Tiny Object Detection

Shuohao Shi, Xin Xu

CodeObject DetectionHyperparameter SearchImage

🎯 What it does: Proposed a simple and effective strategy called Similarity Distance (SimD) for evaluating the similarity of bounding boxes and assigning labels in small object detection.

Skeleton-Based Human Action Recognition with Noisy Labels

Yi Xu, Rainer Stiefelhagen

CodeRecognitionMixture of ExpertsGraph

🎯 What it does: This paper proposes and verifies a framework targeting label noise in skeleton action recognition, and on this basis, designs a new method called NoiseEraSAR

SoftNeRF: A Self-Modeling Soft Robot Plugin for Various Tasks

Jiwei Shan, Hesheng Wang

CodeRobotic IntelligenceNeural Radiance FieldImage

🎯 What it does: Developed SoftNeRF, a self-modeling method for soft robots based on self-supervised vision, which can learn the geometry and nonlinear motion of soft robots from RGB images through differentiable rendering, thereby enabling simulation and prediction of its future states.

SPVSoAP3D: A Second-order Average Pooling Approach to enhance 3D Place Recognition in Horticultural Environments

T. Barros, U. Nunes

CodeRecognitionPoint CloudAgriculture Related

🎯 What it does: Proposed the SPVSoAP3D model and added two new horticultural environment sequences to the HORTO-3DLM dataset

SSCBench: A Large-Scale 3D Semantic Scene Completion Benchmark for Autonomous Driving

Yiming Li, Chen Feng

CodeAutonomous DrivingImagePoint CloudBenchmark

🎯 What it does: Built a large-scale 3D semantic scene completion benchmark named SSCBench, integrating automotive datasets such as KITTI-360, nuScenes, and Waymo, and unified semantic labels to facilitate exploration of SSC methods in street view scenarios.

SSL-RGB2IR: Semi-supervised RGB-to-IR Image-to-Image Translation for Enhancing Visual Task Training in Semantic Segmentation and Object Detection

Aniruddh Sikdar, Suresh Sundaram

CodeImage TranslationObject DetectionSegmentationData SynthesisGenerative Adversarial NetworkImageMultimodality

🎯 What it does: Proposed and implemented the SSL-RGB2IR semi-supervised RGB-to-IR image-to-image translation model for generating synthetic IR images from RGB images.

Stick Roller: Precise In-hand Stick Rolling with a Sample-Efficient Tactile Model

Yipai Du, Yu She

CodeData-Centric LearningRobotic Intelligence

🎯 What it does: Learning High-Resolution Tactile Dynamics for Precise In-Hand Rod Rolling

SurrealDriver: Designing LLM-powered Generative Driver Agent Framework based on Human Drivers’ Driving-thinking Data

Ye Jin, Jiangtao Gong

CodeAutonomous DrivingTransformerLarge Language ModelTextChain-of-Thought

🎯 What it does: Proposes an LLM-driven generative driving agent framework based on human driver driving thought data, collects high-quality natural language data through urban driving experiments, and validates its effectiveness in simulation and human evaluation.

SWCF-Net: Similarity-weighted Convolution and Local-global Fusion for Efficient Large-scale Point Cloud Semantic Segmentation

Zhenchao Lin, Hong Zhang

CodeSegmentationConvolutional Neural NetworkTransformerPoint Cloud

🎯 What it does: Propose a network called SWCF-Net, which combines Similarity-Weighted Convolution and local-global Fusion for efficient semantic segmentation of large-scale point clouds.

SwinMTL: A Shared Architecture for Simultaneous Depth Estimation and Semantic Segmentation from Monocular Camera Images

P. Taghavi, Gaurav Pandey

CodeSegmentationDepth EstimationGenerative Adversarial NetworkImage

🎯 What it does: Proposes a multi-task learning framework that enables monocular cameras to perform simultaneous depth estimation and semantic segmentation.

Switching Sampling Space of Model Predictive Path-Integral Controller to Balance Efficiency and Safety in 4WIDS Vehicle Navigation

Mizuho Aoki, Tatsuya Suzuki

CodeAutonomous DrivingOptimization

🎯 What it does: Implementing navigation for a four-wheel independently driven vehicle to avoid collisions and reach the target point among obstacles of arbitrary shapes using the MPPI control algorithm

Task and Domain Adaptive Reinforcement Learning for Robot Control

Y. Liu, Aamir Ahmad

CodeDomain AdaptationRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose an adaptive agent based on transfer learning that can dynamically adjust strategies to adapt to different tasks and environmental conditions, and verify its multi-task and environmental adaptability in the balloon control challenge.

TD-NeRF: Novel Truncated Depth Prior for Joint Camera Pose and Neural Radiance Field Optimization

Zhen Tan, Dewen Hu

CodePose EstimationDepth EstimationOptimizationNeural Radiance Field

🎯 What it does: Propose a TD-NeRF method based on truncated depth prior, achieving training from unknown camera poses through joint optimization of camera poses and NeRF.

TeFF: Tracking-enhanced Forgetting-free Few-shot 3D LiDAR Semantic Segmentation

Junbao Zhou, Yu Hu

CodeObject TrackingSegmentationPoint Cloud

🎯 What it does: This paper utilizes a tracking model to generate pseudo labels for data augmentation and combines LoRA to achieve few-shot 3D LiDAR semantic segmentation without forgetting.

Temporal- and Viewpoint-Invariant Registration for Under-Canopy Footage using Deep-Learning-based Bird’s-Eye View Prediction

Jiawei Zhou, L. Teixeira

CodeImage TranslationConvolutional Neural NetworkImagePoint CloudAgriculture Related

🎯 What it does: Proposed a cross-seasonal and cross-day under-canopy image sequence registration method that combines GPS and deep learning to generate bird's-eye views for estimating tree positions and achieving registration.

Thermal-NeRF: Neural Radiance Fields from an Infrared Camera

Tian Ye, Ling Pei

CodeGenerationNeural Radiance FieldImage

🎯 What it does: Developed the Thermal-NeRF method, which generates a NeRF volume field representation using only infrared images, and employs thermal mapping and structural thermal constraints to enhance recovery quality in low-light/visually impaired scenarios.

Tightly-Coupled Factor Graph Formulation For Radar-Inertial Odometry

J. Michalczyk, Stephan Weiss

CodeOptimizationSimultaneous Localization and Mapping

🎯 What it does: Propose a radar-inertial odometry (RIO) method based on factor graph nonlinear optimization, implemented within a sliding window framework.

Towards a Surgeon-in-the-Loop Ophthalmic Robotic Apprentice using Reinforcement and Imitation Learning

Amr Gomaa, Antonio KrΓΌger

CodeRobotic IntelligenceReinforcement LearningImageBiomedical Data

🎯 What it does: Developed an image-based surgeon-centered autonomous robotic assistant, utilizing both reinforcement learning and imitation learning for simultaneous training, specifically for autonomous operation during the incision phase of cataract surgery in ophthalmology.

Towards Accurate And Robust Dynamics and Reward Modeling for Model-Based Offline Inverse Reinforcement Learning

Gengyu Zhang, Yan Yan

CodeRobotic IntelligenceReinforcement LearningDiffusion modelScore-based Model

🎯 What it does: Enhance model-based offline inverse reinforcement learning by improving the conservative MDP framework and utilizing score-based diffusion generative models to improve dynamics and reward modeling.

Towards Cross-View-Consistent Self-Supervised Surround Depth Estimation

Laiyan Ding, Rui Huang

CodeDepth EstimationAutonomous DrivingConvolutional Neural NetworkImage

🎯 What it does: Proposed a cross-view consistent self-supervised surround view depth estimation method, designed an efficient pose estimation scheme using only the front view, and introduced dense depth consistency loss, multi-view reconstruction consistency loss, and flip augmentation techniques.

Towards Long Term SLAM on Thermal Imagery

Colin Keil, H. Singh

CodeSimultaneous Localization and MappingImage

🎯 What it does: Developed a learning-based feature thermal imaging visual SLAM baseline system, demonstrating good tracking and day-night relocalization performance on thermal imaging.

Transformer-Based Relationship Inference Model for Household Object Organization by Integrating Graph Topology and Ontology

Xiaodong Li, Yu Gu

CodeRobotic IntelligenceGraph Neural NetworkTransformerLarge Language ModelGraph

🎯 What it does: By constructing a dataset containing ontological attributes and relationships of household items, and using a feature extraction method that combines Graph Attention Network (GAT) with BERT, training under the Transformer framework to achieve relationship inference between items, thereby improving the systematic organization of household items by service robots.

TΕ™iVis: Versatile, Reliable, and High-Performance Tool for Computing Visibility in Polygonal Environments

Jan Mikula, L. PΕ™eučil

CodeMeshBenchmark

🎯 What it does: Developed the TΕ™iVis library for computing visibility-related queries in complex polygonal environments.

UMAD: University of Macau Anomaly Detection Benchmark Dataset

Dong Li (University of Macau), Hui Kong (University of Macau)

CodeAnomaly DetectionImageBenchmark

🎯 What it does: Proposed a reference-based anomaly detection benchmark dataset named UMAD specifically for robot patrolling scenarios, and evaluated the baseline ADr model on this dataset.

Uncertainty-Aware Deployment of Pre-trained Language-Conditioned Imitation Learning Policies

Bo Wu, Nikolai Matni

CodeTransformerReinforcement LearningText

🎯 What it does: Proposes an uncertainty-based deployment method for pre-trained language-conditioned imitation learning agents, calibrating the model using temperature scaling and achieving uncertainty-aware decision-making through the aggregation of local information from candidate actions.

UWB-Based Localization System Considering Antenna Anisotropy and NLOS/Multipath Conditions

Taekyun Kim, Dongjun Lee

CodeSimultaneous Localization and MappingPhysics Related

🎯 What it does: Researched and implemented a localization system based on UWB ranging error models, solving problems of antenna anisotropy, NLOS, and multipath interference, and including anchor self-calibration and state estimation using iterative Kalman filtering;

V2I-Calib: A Novel Calibration Approach for Collaborative Vehicle and Infrastructure LiDAR Systems

Qianxin Qu, Shichun Guo

CodeAutonomous DrivingPoint Cloud

🎯 What it does: Proposed a real-time robust calibration method for vehicle-infrastructure collaborative LiDAR systems based on perceived target spatial association information

V3D-SLAM: Robust RGB-D SLAM in Dynamic Environments with 3D Semantic Geometry Voting

T. Dang, M. Huber

CodeSimultaneous Localization and MappingPoint CloudBenchmark

🎯 What it does: Proposes a robust RGB-D SLAM method called V3D-SLAM, which removes moving objects in dynamic environments through a two-stage lightweight re-evaluation process.

Vertebrae-based Global X-ray to CT Registration for Thoracic Surgeries

Lilu Liu, Yue Wang

CodePose EstimationOptimizationConvolutional Neural NetworkMultimodalityBiomedical DataComputed Tomography

🎯 What it does: Proposes a global X-ray to CT registration method based on vertebrae, automatically locating vertebrae centers via CNN, introducing a 4-DoF solver and AE2 estimator, ultimately achieving an end-to-end trained registration framework under clinical settings.

VIRUS-NeRF - Vision, InfraRed and UltraSonic based Neural Radiance Fields

Nicolaj Schmid (ETH Zurich), Florian Tschopp (ETH Zurich)

CodeAutonomous DrivingComputational EfficiencyNeural Radiance FieldSimultaneous Localization and MappingMultimodalityUltrasound

🎯 What it does: Proposed a VIRUS-NeRF model that utilizes visual, infrared, and ultrasonic sensors for low-cost local map construction.

Volumetric Mapping with Panoptic Refinement using Kernel Density Estimation for Mobile Robots

K. Nguyen, M. Huber

CodeDepth EstimationRobotic IntelligenceConvolutional Neural Network

🎯 What it does: Refine panoramic segmentation of 3D scenes using kernel density estimation, remove anomalies in depth perception, and reconstruct using projected signed distance functions (SDF).

Volumetric Semantically Consistent 3D Panoptic Mapping

Yang Miao, DΓ‘niel BarΓ‘th

CodeAutonomous DrivingOptimizationSimultaneous Localization and Mapping

🎯 What it does: Propose an online 2D-to-3D semantic instance mapping algorithm for generating comprehensive, accurate, and efficient semantic 3D maps, applicable to unmanned systems operating in unstructured environments.

WasteGAN: Data Augmentation for Robotic Waste Sorting through Generative Adversarial Networks

Alberto Bacchin, Takuya Kiyokawa

CodeSegmentationData SynthesisRobotic IntelligenceGenerative Adversarial NetworkImage

🎯 What it does: Proposed the WasteGAN data augmentation method to enhance the semantic segmentation performance of robot garbage classification under scenarios with extremely limited labeled samples (e.g., 100 samples), and utilized the enhanced segmentation results to achieve semantic-aware grasping poses, improving garbage identification and separation efficiency.

Weakly Scene Segmentation Using Efficient Transformer

Hao Huang, Yi Fang

CodeSegmentationTransformerPoint Cloud

🎯 What it does: Proposes a weakly supervised large-scale indoor point cloud scene semantic segmentation method, requiring only 1‰ of points to be labeled, and develops an efficient point neighborhood Transformer and low-rank sparse self-attention approximation.

WidthFormer: Toward Efficient Transformer-based BEV View Transformation

Chenhongyi Yang, Elliot Crowley

CodeAutonomous DrivingTransformerImageBenchmark

🎯 What it does: Proposed WidthFormer, a Transformer-based module for generating real-time and efficient bird's-eye view (BEV) representations from multi-view cameras, applicable to autonomous driving scenarios.