arXivSub Start free trial

IROS 2023 Papers with Code

IEEE/RSJ International Conference on Intelligent Robots and Systems Β· 94 papers with a public code repository

360FusionNeRF: Panoramic Neural Radiance Fields with Joint Guidance

Shreyas Kulkarni, S. Scherer

CodeGenerationVision Language ModelNeural Radiance FieldImage

🎯 What it does: A semi-supervised learning framework based on NeRF, which generates new view renderings from a single 360° panorama using geometric supervision and semantic consistency guidance;

A Game-Theoretic Framework for Joint Forecasting and Planning

K. Kedia, Sanjiban Choudhury

CodeRobotic IntelligenceTime SeriesSequential

🎯 What it does: Propose a game theory framework for joint prediction and planning of robots in human environments, and provide a practical end-to-end trained algorithm.

A Grasp Pose is All You Need: Learning Multi-Fingered Grasping with Deep Reinforcement Learning from Vision and Touch

Federico Ceola, Lorenzo Natale

CodeRobotic IntelligenceReinforcement LearningVideo

🎯 What it does: Proposed and implemented a deep reinforcement learning method called G-PAYN for multi-fingered grasping tasks by the iCub humanoid hand in a simulated environment; the method initializes the policy by automatically collecting task demonstrations and combines grasp poses generated by external algorithms as the starting point of the grasping process; subsequently, the trained control policy completes the grasping; experiments were conducted on the MuJoCo simulator, using objects from the YCB-Video dataset for evaluation.

A Multiplicative Value Function for Safe and Efficient Reinforcement Learning

Nick BΓΌhrer, L. Gool

CodeReinforcement LearningImagePoint Cloud

🎯 What it does: Propose a safe model-free reinforcement learning algorithm that employs a multiplicative value function to separate the safety estimator from the reward estimator;

Accurate and Interactive Visual-Inertial Sensor Calibration with Next-Best-View and Next-Best-Trajectory Suggestion

Christopher Choi, Stefan Leutenegger

CodeSimultaneous Localization and MappingMultimodality

🎯 What it does: Proposes a calibration process for visual-inertial (VI) sensors, utilizing a graphical user interface (GUI) and information theory to guide non-experts in collecting information-rich calibration data through recommendations for the next best view and next best trajectory, enabling calibration of intrinsic parameters, extrinsic parameters, and time errors.

Adaptive PD Control Using Deep Reinforcement Learning for Local-Remote Teleoperation with Stochastic Time Delays

Lucy McCutcheon, Saber Fallah

CodeOptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose an adaptive PD control method using deep reinforcement learning to address random time delay issues in local-remote collaborative systems

An Attentional Recurrent Neural Network for Occlusion-Aware Proactive Anomaly Detection in Field Robot Navigation

Andre Schreiber, K. Driggs-Campbell

CodeAnomaly DetectionRecurrent Neural NetworkAgriculture Related

🎯 What it does: Proposes an attention-based recurrent neural network for proactive anomaly detection in mobile robots within agricultural environments, integrating current perceptual inputs, planned control actions, and potential representations of previous states.

An MCTS-DRL Based Obstacle and Occlusion Avoidance Methodology in Robotic Follow-Ahead Applications

Sahar Leisiazar, Mo Chen

CodeRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed an obstacle and occlusion avoidance method for robot following applications, and developed a high-level decision algorithm to generate short-term navigation goals.

An Open-Source Robotic Chinese Chess Player

Shan An, Hong Zhang

CodeObject DetectionRobotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Designed and implemented a low-cost, open-source robot for playing Chinese chess.

ANEC: Adaptive Neural Ensemble Controller for Mitigating Latency Problems in Vision-Based Autonomous Driving

Aws Khalil, Jaerock Kwon

CodeAutonomous DrivingComputational EfficiencyMixture of ExpertsImage

🎯 What it does: Studied the impact of algorithmic latency in vision-driven neural networks for lane-keeping tasks, and proposed the Adaptive Neural Ensemble Controller (ANEC).

Autonomous Exploration and Mapping for Mobile Robots via Cumulative Curriculum Reinforcement Learning

Z. Li, Ning Li

CodeRobotic IntelligenceReinforcement LearningSimultaneous Localization and Mapping

🎯 What it does: Proposes a Cumulative Curriculum Reinforcement Learning (CCRL) framework that combines deep reinforcement learning with curriculum learning for autonomous exploration and mapping in mobile robots.

Autonomous Marker-Less Rapid Aerial Grasping

Erik Bauer, Robert K. Katzschmann

CodeDepth EstimationRobotic IntelligenceConvolutional Neural NetworkPoint Cloud

🎯 What it does: Proposed a vision-based drone rapid grasping system that uses Mask R-CNN segmentation, depth cameras to generate dense point clouds, and geometric grasping planning without relying on markers or known appearances

Boosting Feedback Efficiency of Interactive Reinforcement Learning by Adaptive Learning from Scores

Shukai Liu, Liang Zhang

CodeRobotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningScore-based Model

🎯 What it does: Improve the feedback efficiency of interactive reinforcement learning by using human-provided scores instead of pairwise preferences, and propose an adaptive learning scheme to handle unstable scores.

BSH-Det3D: Improving 3D Object Detection with BEV Shape Heatmap

You Shen, D. Kerr

CodeObject DetectionAutonomous DrivingPoint CloudBenchmark

🎯 What it does: Proposed and implemented the BSH-Det3D model, leveraging BEV shape heatmap to enhance spatial features, and designed the Pillar-based Shape Completion (PSC) module and Attention-based Densification Fusion (ADF) module.

CEFHRI: A Communication Efficient Federated Learning Framework for Recognizing Industrial Human-Robot Interaction

Umar Khalid, Chen Chen

CodeRecognitionFederated LearningVideoBenchmark

🎯 What it does: Proposes a communication-efficient federated learning framework called CEFHRI for industrial human-robot interaction, addressing issues of data heterogeneity and communication costs.

CoFlyers: A Universal Platform for Collective Flying of Swarm Drones

Jialei Huang, Tianjiang Hu

CodeRobotic Intelligence

🎯 What it does: Proposed and developed an open-source platform named CoFlyers for end-to-end development from cluster models to real-world drone swarm flights.

CompUDA: Compositional Unsupervised Domain Adaptation for Semantic Segmentation Under Adverse Conditions

Ziqiang Zhengl, Sai-Kit Yeung

CodeSegmentationDomain AdaptationImage

🎯 What it does: This paper proposes an unsupervised domain adaptation method called CompUDA based on multi-factor decomposition for semantic segmentation under adverse weather conditions. The method improves semantic segmentation performance by decomposing domain differences into three factors: style, visibility, and image quality, and adapting them individually using intermediate domains.

ConSOR: A Context-Aware Semantic Object Rearrangement Framework for Partially Arranged Scenes

Kartik Ramachandruni, Sonia Chernova

Code

🎯 What it does: Proposes ConSOR, which rearranges objects by leveraging contextual cues from partially arranged scenes without requiring users to explicitly specify the target scene;

Depth-Based 6DoF Object Pose Estimation Using Swin Transformer

Zhujun Li, I. Stamos

CodePose EstimationConvolutional Neural NetworkTransformerImagePoint Cloud

🎯 What it does: Proposes the SwinDePose framework, which utilizes the geometric information of depth images to achieve accurate 6D pose estimation.

DR-Pose: A Two-Stage Deformation-and-Registration Pipeline for Category-Level 6D Object Pose Estimation

Lei Zhou, Marcelo H ANG Jr

CodePose EstimationPoint CloudBenchmark

🎯 What it does: Proposes a two-stage deformation and registration pipeline, DR-Pose, for category-level 6D object pose estimation.

Driver Distraction Detection for Daytime and Nighttime with Unpaired Visible and Infrared Image Translation

Hong-Ze Shen, Huei-Yung Lin

CodeImage TranslationPose EstimationAutonomous DrivingImage

🎯 What it does: Propose a network model named V2IA-Net that utilizes daytime visible light and nighttime infrared images for driver distraction detection, combining driver action recognition and head pose detection to achieve real-time analysis.

Efficient and Feasible Robotic Assembly Sequence Planning via Graph Representation Learning

Matan Atad, Rudolph Triebel

CodeRepresentation LearningRobotic IntelligenceGraph Neural NetworkGraph

🎯 What it does: Proposed a holistic graph method that includes Assembly Graph representation and Graph Assembly Processing Network (GRACE) for robot assembly sequence planning.

Efficient Heuristics for Multi-Robot Path Planning in Crowded Environments

Teng Guo, Jingjin Yu

CodeOptimizationRobotic Intelligence

🎯 What it does: Proposed and implemented two hybrid algorithms, DCBS and SCBS, for multi-robot path planning in dense scenarios.

Efficient Visuo-Haptic Object Shape Completion for Robot Manipulation

Lukas Rustler, M. Hoffmann

CodeRobotic IntelligenceVision-Language-Action ModelImageMultimodality

🎯 What it does: Achieving robotic object shape completion through closed-loop visual and tactile reconstruction.

ElC-OIS: Ellipsoidal Clustering for Open-World Instance Segmentation on LiDAR Data

Wenbang Deng, Xieyuanli Chen

CodeSegmentationDiffusion modelPoint Cloud

🎯 What it does: Propose an open-world instance segmentation framework suitable for LiDAR point clouds that can accurately segment both known and unknown instances.

Enhancing State Estimation in Robots: A Data-Driven Approach with Differentiable Ensemble Kalman Filters

Xinyu Liu, H. B. Amor

CodeRobotic Intelligence

🎯 What it does: Proposes a novel robotic state estimation framework based on the differentiable ensemble Kalman filter (DEnKF), which implicitly models process noise using stochastic neural networks.

F2BEV: Bird's Eye View Generation from Surround-View Fisheye Camera Images for Automated Driving

Ekta U. Samani, A. Banerjee

CodeGenerationData SynthesisAutonomous DrivingTransformerImage

🎯 What it does: Proposed a baseline method called F2BEV for generating discretized bird's-eye-view (BEV) height maps and semantic segmentation maps from fisheye camera images.

Few-Shot Segmentation and Semantic Segmentation for Underwater Imagery

Imran Kabir, Md. Alimoor Reza

CodeSegmentationMeta LearningConvolutional Neural NetworkImageBenchmark

🎯 What it does: Proposes a new dense pixel-level annotated dataset based on underwater animals and addresses the few-shot segmentation and semantic segmentation tasks on this dataset

From Temporal-Evolving to Spatial-Fixing: A Keypoints-Based Learning Paradigm for Visual Robotic Manipulation

KΓ©vin Riou, P. Callet

CodeRobotic Intelligence

🎯 What it does: Propose a hierarchical behavior cloning method that decomposes traditional BC into high-level planning to convert initial observations into spatial waypoints, and low-level execution of predefined primitives to reach waypoints.

Generating Executable Action Plans with Environmentally-Aware Language Models

Maitrey Gramopadhye, D. Szafir

CodeGenerationRobotic IntelligenceTransformerLarge Language ModelScore-based ModelText

🎯 What it does: Propose a scheme to integrate environmental objects and their relationships as additional inputs into large language models (LLMs), generating executable and environment-matching action plans.

GP-Guided MPPI for Efficient Navigation in Complex Unknown Cluttered Environments

Ihab S. Mohamed, Lantao Liu

CodeAutonomous DrivingOptimizationReinforcement Learning

🎯 What it does: Proposed an online learning control strategy GP-MPPI based on sparse Gaussian processes (SGP), combining MPPI with SGP for efficient navigation in unknown, crowded environments.

Hybrid Map-Based Path Planning for Robot Navigation in Unstructured Environments

Jiayang Liu, Huimin Lu

CodeRobotic Intelligence

🎯 What it does: Proposes a new hybrid map representation and develops a path planning method that considers robot posture for traversability assessment, aiming to achieve safe and efficient robot navigation.

Hybrid Object Tracking with Events and Frames

Zhichao Li, C. Bartolozzi

CodeObject TrackingPose EstimationConvolutional Neural NetworkImageMultimodalityPoint Cloud

🎯 What it does: Developed a dual Kalman filter combining event cameras and RGB-D for high-speed and accurate object pose tracking.

IDA: Informed Domain Adaptive Semantic Segmentation

Zheng Chen, Lantao Liu

CodeSegmentationDomain AdaptationImage

🎯 What it does: Proposed the IDA self-training framework guided by category-level segmentation performance for unsupervised domain adaptation in semantic segmentation

Image Restoration via UAVFormer for Under-Display Camera of UAV

Zhuoran Zheng, X. Jia

CodeRestorationTransformerImage

🎯 What it does: This paper proposes a deep network called UAVFormer specifically designed to address image degradation caused by transparent film covering UAV-mounted cameras, achieving image restoration.

InsMOS: Instance-Aware Moving Object Segmentation in LiDAR Data

Neng Wang, Xieyuanli Chen

CodeSegmentationAutonomous DrivingConvolutional Neural NetworkPoint CloudBenchmark

🎯 What it does: Proposed a new network that can predict point-level moving labels and detect instance information of major traffic participants, achieving instance-aware LiDAR moving object segmentation.

InteractionNet: Joint Planning and Prediction for Autonomous Driving with Transformers

Jiawei Fu, Nanning Zheng

CodeAutonomous DrivingTransformer

🎯 What it does: Proposed InteractionNet, which uses Transformer to achieve joint reasoning of planning and prediction, and captures interactions among traffic participants through global context sharing; additionally, a specialized Transformer module focusing on critical or unseen vehicles was added.

Interactive Spatiotemporal Token Attention Network for Skeleton-Based General Interactive Action Recognition

Yuhang Wen, Mengyuan Liu

CodeRecognitionConvolutional Neural NetworkTransformerGraph

🎯 What it does: Proposes an ISTA-Net network for skeletal-based interactive action recognition, capable of simultaneously modeling spatial, temporal, and interaction relationships.

Joint Out-of-Distribution Detection and Uncertainty Estimation for Trajectory Prediction

Julian Wiederer, Vasileios Belagiannis

CodeAnomaly DetectionAutonomous DrivingSequential

🎯 What it does: Proposes a trajectory prediction method that combines Gaussian Mixture Models for outlier detection and error regression networks for uncertainty estimation.

Keypoints-Based Adaptive Visual Servoing for Control of Robotic Manipulators in Configuration Space

Sreejani Chatterjee, B. Γ‡alli

CodeRobotic IntelligenceImage

🎯 What it does: Proposes an adaptive visual servoing method based on keypoints, achieving robotic arm control in configuration space using natural features;

LAMP: Leveraging Language Prompts for Multi-Person Pose Estimation

Shengnan Hu, G. Sukthankar

CodePose EstimationPrompt EngineeringVision Language ModelImageText

🎯 What it does: Proposed a language prompt-based multi-person human pose estimation method called LAMP, which utilizes text representations generated by CLIP to assist in pose reasoning.

Learning Bifunctional Push-Grasping Synergistic Strategy for Goal-Agnostic and Goal-Oriented Tasks

Dafa Ren, Xiaoqiang Ren

CodeRobotic IntelligenceReinforcement LearningImage

🎯 What it does: A dual-functional push-grasp collaborative strategy is proposed, combining push and grasp actions to achieve efficient grasping of all objects in the workspace (target-agnostic) and predefined target objects (target-specific). The strategy uses visual observations as input, employs a dual-functional network to generate pixel-level Q-value dense maps for push and grasp primitives, and builds a hierarchical reinforcement learning framework. It treats target-agnostic tasks as combinations of multiple target-specific tasks, and adopts a two-stage training method to separately train the two types of tasks. Finally, the method directly transfers from simulation environments to the real world without requiring additional fine-tuning.

Learning Soft Robot Dynamics Using Differentiable Kalman Filters and Spatio-Temporal Embeddings

Xinyu Liu, H. B. Amor

CodeRobotic IntelligenceTime SeriesSequential

🎯 What it does: Using differentiable Kalman filters and spatiotemporal embedding methods, end-to-end training of soft robot dynamics models to learn system dynamics, noise characteristics, and temporal behavior.

Learning to Solve Tasks with Exploring Prior Behaviours

Ruiqi Zhu, O. Γ‡eliktutan

CodeRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose an intrinsic reward-driven example control method (IRDEC), enabling agents to complete tasks in sparse reward environments by exploring prior behaviors and connecting them with task-specific behaviors from examples.

LIO-PPF: Fast LiDAR-Inertial Odometry via Incremental Plane Pre-Fitting and Skeleton Tracking

Xingyu Chen, Thomas H. Li

CodePose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Propose a fast LiDAR-Inertial odometry method based on incremental plane pre-fitting (PPF) and skeleton tracking

LocalViT: Analyzing Locality in Vision Transformers

Yawei Li, L. Gool

CodeClassificationTransformerImage

🎯 What it does: Study the locality mechanism in Vision Transformers and systematically validate it by incorporating a locality module into the feed-forward network.

MDT3D: Multi-Dataset Training for LiDAR 3D Object Detection Generalization

Louis Soum-Fontez, Franccois Goulette

CodeObject DetectionPoint Cloud

🎯 What it does: Proposed and implemented a multi-dataset training (MDT3D) method to enhance the robustness of 3D object detection models in new environments with different sensor configurations.

MIMIR-UW: A Multipurpose Synthetic Dataset for Underwater Navigation and Inspection

Olaya Álvarez-Tuñón, Erdal Kayacan

CodeSegmentationData SynthesisDepth EstimationRobotic IntelligenceSimultaneous Localization and MappingImageMultimodalityBenchmark

🎯 What it does: Created a multi-purpose synthetic underwater dataset MIMIR-UW for SLAM, depth estimation, and object segmentation, evaluated in pipeline inspection scenarios.

Multi-Scale Point Octree Encoding Network for Point Cloud Based Place Recognition

Zhilong Tang, Hong Zhang

CodeRetrievalTransformerPoint Cloud

🎯 What it does: Propose a multi-scale point octree encoding network (MPOE-Net), which generates high-discrimination global descriptors through a point octree encoding module, a multi-Transformer with grouped offset-attention mechanism, and multi-layer NetVLAD, achieving efficient retrieval for point cloud location recognition.

Multi-View Robust Collaborative Localization in High Outlier Ratio Scenes Based on Semantic Features

Yujie Tang, Yufeng Yue

CodeRobotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Propose a robust collaborative localization algorithm for outdoor environments with high outlier rates, HORCL, which utilizes a Mixture Probability Model to compute inlier probabilities and combines a hierarchical EM algorithm to perform two-layer outlier filtering on loop closure constraints and point pairs, thereby enhancing the accuracy and robustness of multi-robot localization.

Multiplanar Self-Calibration for Mobile Cobot 3D Object Manipulation Using 2D Detectors and Depth Estimation

T. Dang, M. Huber

CodeDepth EstimationRobotic IntelligenceImagePoint Cloud

🎯 What it does: Propose a multi-plane self-calibration method to register the camera system with the robot end-effector for 3D object manipulation.

navlie: A Python Package for State Estimation on Lie Groups

C. Cossette, J. R. Forbes

CodeOptimization

🎯 What it does: Developed a Python package called navlie for fast prototyping of state estimation on Lie groups.

Object Manipulation Through Contact Configuration Regulation: Multiple and Intermittent Contacts

Orion Taylor, Alberto Rodriguez

CodeRobotic IntelligenceImage

🎯 What it does: Through a factor graph estimation framework that integrates limited visual feedback, force/torque sensing, and robot proprioception, the estimation and control of all contact points, geometry, and patterns between the robot, object, and environment are achieved, enabling manipulation tasks for unknown objects.

Online Adaptive Disparity Estimation for Dynamic Scenes in Structured Light Systems

Rukun Qiao, Hong-yan Zha

CodeDepth Estimation

🎯 What it does: Proposes a framework based on self-supervised online adaptation, leveraging unsupervised loss functions from long-sequence inputs and sparse trajectory and confidence masks from multi-frame pattern streams, significantly improving online adaptation speed and performance on unseen data

Online Monocular Lane Mapping Using Catmull-Rom Spline

Zhijian Qiao, S. Shen

CodeAutonomous DrivingOptimizationSimultaneous Localization and MappingImage

🎯 What it does: Proposed an online lane mapping method based on a monocular camera and odometry, generating a lane map based on Catmull-Rom splines.

Online Self-Supervised Thermal Water Segmentation for Aerial Vehicles

Connor T. Lee, Soon-Jo Chung

CodeSegmentationDomain AdaptationContrastive LearningImage

🎯 What it does: Through online self-supervision, utilizing texture and motion cues to transfer an RGB-trained water segmentation network to the aerial thermal imaging domain.

Optical Flow Boosts Unsupervised Localization and Segmentation

Xinyu Zhang, Abdeslam Boularias

CodeObject DetectionSegmentationTransformerContrastive LearningOptical Flow

🎯 What it does: Fine-tune an unsupervised vision transformer (ViT) using optical flow information to enhance performance in unsupervised object localization and segmentation.

PanelPose: A 6D Pose Estimation of Highly-Variable Panel Object for Robotic Robust Cockpit Panel Inspection

Han Sun, Qixin Cao

CodeData SynthesisPose EstimationRobotic IntelligenceSimultaneous Localization and MappingImage

🎯 What it does: Proposed a 6D pose estimation method called PanelPose for handling highly variable panel objects in robot inspection scenarios within aviation cockpits.

PANet: LiDAR Panoptic Segmentation with Sparse Instance Proposal and Aggregation

Jianbiao Mei, Yong Liu

CodeSegmentationAutonomous DrivingPoint Cloud

🎯 What it does: Propose the PANet framework, utilizing sparse instance proposal (SIP) and an instance aggregation module for LiDAR panoramic segmentation, removing the offset branch and enhancing performance on large objects.

Principled ICP Covariance Modelling in Perceptually Degraded Environments for the EELS Mission Concept

W. Talbot, V. Ila

CodeSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a principled modeling approach for LiDAR point-to-plane ICP covariance, and conducted comparative evaluation of new and old models within the complete SLAM pipeline.

PuSHR: A Multirobot System for Nonprehensile Rearrangement

Sidharth Talia, S. Srinivasa

CodeOptimizationRobotic Intelligence

🎯 What it does: This study addresses the problem of non-grasping object rearrangement using vehicle-type pushing robots, constructing the PuSHR multi-robot system. It optimizes task allocation and trajectory planning in the offline phase, and achieves decentralized trajectory tracking in the online phase.

Quantized Distillation: Optimizing Driver Activity Recognition Models for Resource-Constrained Environments

Calvin Tanama, Alina Roitberg

CodeRecognitionAutonomous DrivingOptimizationComputational EfficiencyKnowledge DistillationConvolutional Neural Network

🎯 What it does: Propose a lightweight framework that improves 3D MobileNet by combining knowledge distillation and model quantization to achieve driver behavior recognition in resource-constrained environments.

RACECAR - The Dataset for High-Speed Autonomous Racing

A. Kulkarni, Madhur Behl

CodeAutonomous DrivingMultimodalityPoint Cloud

🎯 What it does: Created and released the RACECAR high-speed full-scale autonomous racing car dataset, collected multi-modal sensor data, covering 11 track scenarios, 27 events, and 6.5 hours of racing;

RADA: Robust Adversarial Data Augmentation for Camera Localization in Challenging Conditions

Jialu Wang, Andrew Markham

CodeAutonomous DrivingAdversarial AttackImage

🎯 What it does: Propose a robust adversarial data augmentation method called RADA to enhance the robustness of camera localization under complex conditions.

RaPlace: Place Recognition for Imaging Radar using Radon Transform and Mutable Threshold

Hyesu Jang, Ayoung Kim

CodeRecognitionRetrievalImage

🎯 What it does: Propose a radar-based place recognition method that calculates similarity using sinogram images from the Radon transform and frequency domain cross-correlation.

RaSpectLoc: RAman SPECTroscopy-dependent robot LOCalisation

C. Thirgood, Simon J. Hadfield

CodeRobotic IntelligenceSimultaneous Localization and MappingBenchmark

🎯 What it does: Proposed and implemented the first robot localization algorithm based on material composition, which utilizes a Raman spectrometer to acquire material information and integrates it with visual, structural, and semantic features to improve localization accuracy.

Recurrent Macro Actions Generator for POMDP Planning

Yuanchu Liang, Hanna Kurniawati

CodeOptimizationRecurrent Neural NetworkReinforcement Learning

🎯 What it does: This paper proposes a simple recurrent neural network for generating appropriate macro action sets in POMDP planning, thereby reducing the effective planning cycle and lowering computational complexity.

RFDNet: Real-Time 3D Object Detection Via Range Feature Decoration

Hongda Chang, Jun Cheng

CodeObject DetectionAutonomous DrivingComputational EfficiencyPoint Cloud

🎯 What it does: Proposed a real-time 3D object detection framework named RFDNet

RGBD Fusion Grasp Network with Large-Scale Tableware Grasp Dataset

J. Yoon, Sungchul Kang

CodePose EstimationRobotic IntelligenceMultimodality

🎯 What it does: Propose an RGBD fusion-based grasping network, construct a large-scale utensil grasping dataset, and design a stable grasping pose sampling method.

RMBench: Benchmarking Deep Reinforcement Learning for Robotic Manipulator Control

Yanfei Xiang, Siwei Lyu

CodeRobotic IntelligenceReinforcement LearningImageBenchmark

🎯 What it does: Proposed RMBench, the first benchmark for robotic manipulation in high-dimensional continuous action and state spaces; implemented and evaluated deep reinforcement learning algorithms using pixel inputs, reporting their average performance and learning curves.

Rollvox: Real-Time and High-Quality LiDAR Colorization with Rolling Shutter Camera

Sheng Hong, Shaojie Shen

CodeAutonomous DrivingSimultaneous Localization and MappingOptical FlowImagePoint Cloud

🎯 What it does: This study proposes a system that uses a low-cost rolling shutter camera to perform real-time high-quality coloring of LiDAR point clouds.

RVWO: A Robust Visual-Wheel SLAM System for Mobile Robots in Dynamic Environments

J. Mahmoud, S. Kolyubin

CodeRobotic IntelligenceSimultaneous Localization and MappingImage

🎯 What it does: Proposed a robust visual-wheel SLAM system (RVWO) for localization and mapping of mobile robots in dynamic environments

SDF-Pack: Towards Compact Bin Packing with Signed-Distance-Field Minimization

Jia-Yu Pan, Chi-Wing Fu

CodeOptimizationRobotic Intelligence

🎯 What it does: Proposes the SDF-Pack method based on signed distance fields (SDF), representing object geometry with truncated SDF, and computes object placement positions and arrangement sequences within containers through an SDF-minimization heuristic algorithm to achieve more compact robotic bin packing.

See What the Robot Can't See: Learning Cooperative Perception for Visual Navigation

Jan Blumenkamp, Amanda Prorok

CodeRobotic IntelligenceGraph Neural NetworkImage

🎯 What it does: Train sensor encoders and communicate with robots to achieve visual navigation without localization;

Shape Completion with Prediction of Uncertain Regions

Matthias Humt, U. Hillenbrand

CodeRestorationGenerationImage

🎯 What it does: Propose two shape completion methods for predicting the shape of uncertain regions, and generate and use a real rendered depth image dataset derived from ShapeNet for training and evaluation.

SSC-RS: Elevate LiDAR Semantic Scene Completion with Representation Separation and BEV Fusion

Jianbiao Mei, Yong Liu

CodeSegmentationAutonomous DrivingRepresentation LearningConvolutional Neural NetworkPoint Cloud

🎯 What it does: Propose a network called SSC-RS for outdoor LiDAR semantic scene completion, combining representation separation and BEV fusion techniques.

SSGM: Spatial Semantic Graph Matching for Loop Closure Detection in Indoor Environments

Yujie Tang, Yufeng Yue

CodeGraph Neural NetworkSimultaneous Localization and MappingGraph

🎯 What it does: Proposed a spatial semantic graph matching method (SSGM) for loop closure detection in indoor environments, achieving loop closure detection by aligning semantic graphs and evaluating spatial distribution similarity.

Stackelberg Meta-Learning for Strategic Guidance in Multi-Robot Trajectory Planning

Yuhan Zhao, Quanyan Zhu

CodeRobotic IntelligenceMeta Learning

🎯 What it does: This paper proposes a meta-learning framework based on the Stackelberg game for collaborative guidance of leader-follower robots in multi-trajectory planning.

Streaming Motion Forecasting for Autonomous Driving

Ziqi Pang, Yu-Xiong Wang

CodeAutonomous DrivingTime SeriesSequentialBenchmark

🎯 What it does: Propose a continuous streaming trajectory prediction benchmark and introduce a meta-algorithm called 'Predictive Streamer' compatible with existing prediction models to handle occluded agents and maintain temporal consistency.

T-UDA: Temporal Unsupervised Domain Adaptation in Sequential Point Clouds

Awet Haileslassie Gebrehiwot, TomΓ‘Ε‘ Svoboda

CodeSegmentationDomain AdaptationPoint Cloud

🎯 What it does: Proposed an unsupervised domain adaptation framework combining temporal point cloud geometric consistency with the mean teacher method to enhance the robustness of 3D semantic segmentation.

T2FPV: Dataset and Method for Correcting First-Person View Errors in Pedestrian Trajectory Prediction

Ben Stoler, Jean Oh

CodeData SynthesisAutonomous DrivingVideoBenchmark

🎯 What it does: Proposed the T2FPV method to generate high-fidelity first-person view (FPV) datasets from real-world top-down trajectory data, designed the CoFE module for end-to-end correction of FPV errors in trajectory prediction algorithms, and released the T2FPV-ETH dataset along with software tools.

Thoracic Cartilage Ultrasound-CT Registration Using Dense Skeleton Graph

Zhongliang Jiang, N. Navab

CodeGraph Neural NetworkPoint CloudBiomedical DataComputed TomographyUltrasound

🎯 What it does: Propose a non-rigid registration method based on a dense skeleton graph to align thoracic cartilage ultrasound with CT images, thereby transferring planning paths from a generic atlas to individual patients.

Towards a Robust Adversarial Patch Attack Against Unmanned Aerial Vehicles Object Detection

Samridha Shrestha, Eduardo Viegas

CodeObject DetectionAdversarial AttackImage

🎯 What it does: Propose a robust adversarial patch attack method for UAV target detection, considering factors such as UAV camera perspective, distance, and brightness variations;

Trajectory Tracking via Multiscale Continuous Attractor Networks

Therese Joseph, Michael Milford

CodeAutonomous DrivingHyperparameter SearchRobotic IntelligenceImagePoint Cloud

🎯 What it does: Proposed and implemented a multi-scale continuous attractor network (MCAN) to achieve trajectory tracking over a wide speed range, realized automatic parameter tuning through genetic algorithms, and open-sourced a city-scale navigation simulator applicable to any street network.

Transparent Object Tracking with Enhanced Fusion Module

Kalyan Garigapati, Haibin Ling

CodeObject TrackingTransformerBenchmark

🎯 What it does: Proposes a novel feature fusion technique that embeds transparency information into a fixed feature space to enhance tracking accuracy for transparent objects; and designs a new tracker architecture based on this technique.

Uncertainty-Aware Lidar Place Recognition in Novel Environments

Keita Mason, Dimity Miller

CodeAutonomous DrivingMixture of ExpertsSimultaneous Localization and MappingPoint CloudBenchmark

🎯 What it does: Proposed and implemented an uncertainty-aware LiDAR localization method, established a new evaluation protocol, and constructed the first comprehensive benchmark, testing the performance of five uncertainty estimation techniques on three large-scale datasets.

Ungar - A C++ Framework for Real-Time Optimal Control Using Template Metaprogramming

Flavio De Vincenti, Stelian Coros

CodeOptimizationRobotic Intelligence

🎯 What it does: Developed an open-source C++ framework called Ungar for implementing high-dimensional optimal control problems, demonstrated in quadrupedal locomotion and multi-robot cooperative manipulation using model predictive control.

UnLoc: A Universal Localization Method for Autonomous Vehicles using LiDAR, Radar and/or Camera Input

Muhammad Ibrahim, Ajmal Saeed Mian

CodeAutonomous DrivingConvolutional Neural NetworkTransformerImageMultimodalityPoint Cloud

🎯 What it does: Proposes UnLoc, a unified multi-sensor localization method that supports LiDAR, Camera, and RADAR as needed under all weather conditions;

Unsupervised OmniMVS: Efficient Omnidirectional Depth Inference via Establishing Pseudo-Stereo Supervision

Zisong Chen, Yao Zhao

CodeDepth EstimationConvolutional Neural NetworkImage

🎯 What it does: Proposed the first unsupervised omnidirectional multi-view stereo (Omnidirectional MVS) framework based on multiple fisheye images, and designed an efficient Un-OmniMVS network.

VDBblox: Accurate and Efficient Distance Fields for Path Planning and Mesh Reconstruction

Yi-Feng Bai, Yaonan Wang

CodeComputational EfficiencyRobotic IntelligenceSimultaneous Localization and MappingPoint CloudMesh

🎯 What it does: Proposed the VDBblox mapping framework, which can incrementally construct Euclidean Signed Distance Fields (ESDF) from TSDF mapping and improve mesh reconstruction quality.

VIW-Fusion: Extrinsic Calibration and Pose Estimation for Visual-IMU-Wheel Encoder System

Chunxia Qiao, Dan Zhang

CodePose EstimationAutonomous DrivingOptimizationSimultaneous Localization and MappingImageMultimodalityTime Series

🎯 What it does: Propose a joint extrinsic calibration algorithm for the camera-IMU-wheel encoder system and a multi-sensor fusion pose estimation algorithm, while improving the VIO initialization method.

VL-Grasp: a 6-Dof Interactive Grasp Policy for Language-Oriented Objects in Cluttered Indoor Scenes

Yuhao Lu, Shengjin Wang

CodePose EstimationRobotic IntelligenceVision-Language-Action ModelMultimodality

🎯 What it does: Propose a vision-and-language interactive grasping strategy, VL-Grasp, to achieve target object grasping in indoor scenes.

Walking in Narrow Spaces: Safety-Critical Locomotion Control for Quadrupedal Robots with Duality-Based Optimization

Qiayuan Liao, K. Sreenath

CodeOptimizationRobotic Intelligence

🎯 What it does: Propose a safety-critical quadruped robot locomotion control framework enabling the robot to navigate safely in crowded environments

WIT-UAS: A Wildland-Fire Infrared Thermal Dataset to Detect Crew Assets from Aerial Views

Andrew Jong, S. Scherer

CodeObject DetectionData-Centric LearningImageBenchmark

🎯 What it does: Proposed the WIT-UAS dataset, which collects drone flight data and manually annotated images captured using long-wave infrared thermal imaging technology in controlled wildfire environments, for detecting personnel and vehicle assets in fire scenes.