IEEE International Conference on Robotics and Automation · 140 papers
Saturation-Aware Angular Velocity Estimation: Extending the Robustness of SLAM to Aggressive Motions*
Simon-Pierre Deschênes, François Pomerleau
CodeAutonomous DrivingRobotic IntelligenceSimultaneous Localization and Mapping
🎯 What it does: Proposed a method using accelerometers to estimate angular velocity, enhancing the robustness of SLAM in scenarios where intense motion causes gyroscope saturation
🎯 What it does: Proposed a self-correcting visual navigation method called SCALE, which can proactively avoid out-of-distribution (OOD) scenarios without human intervention.
Scene Informer: Anchor-based Occlusion Inference and Trajectory Prediction in Partially Observable Environments
Bernard Lange, Mykel J. Kochenderfer
CodeAutonomous DrivingTransformerMultimodality
🎯 What it does: Propose Scene Informer, a unified framework for predicting the trajectories of visible agents and inferring occluded agents in partially observable environments.
Semi-Supervised Learning for Visual Bird’s Eye View Semantic Segmentation
Junyu Zhu, Yong Liu
CodeSegmentationImage
🎯 What it does: A visual bird's-eye-view (BEV) semantic segmentation framework based on semi-supervised learning is proposed, which enhances model performance by leveraging unlabeled images through consistency loss and feature constraints, and introduces a conjoint rotation data augmentation method to maintain geometric relationships between front-view and BEV.
SeqTrack3D: Exploring Sequence Information for Robust 3D Point Cloud Tracking
Yu Lin, Zheng Fang
CodeObject TrackingAutonomous DrivingPoint Cloud
🎯 What it does: Propose SeqTrack3D, which employs a sequence-to-sequence tracking paradigm, combining historical point clouds and bounding box sequences to capture target motion across consecutive frames.
🎯 What it does: Proposes a simulation-to-real domain adaptation method for RGB-D grasp detection, achieving domain adaptation through self-supervised rotation pre-training and a global-local alignment pipeline.
SocialGAIL: Faithful Crowd Simulation for Social Robot Navigation
Bo Ling, Weiwei Wu
CodeAutonomous DrivingReinforcement Learning from Human FeedbackGraph Neural NetworkReinforcement LearningGenerative Adversarial Network
🎯 What it does: Propose a data-driven crowd simulation method named SocialGAIL based on generative adversarial imitation learning, aiming to more realistically reproduce pedestrian navigation behavior in crowded environments;
SONIC: Sonar Image Correspondence using Pose Supervised Learning for Imaging Sonars
Samiran Gode, Michael Kaess
CodePose EstimationImage
🎯 What it does: Proposed the SONIC (SONar Image Correspondence) network, which achieves feature correspondence for underwater sonar images through pose-supervised learning, maintaining robustness under viewpoint changes;
Specifying and Monitoring Safe Driving Properties with Scene Graphs
Felipe Toledo, Matthew B. Dwyer
CodeAutonomous DrivingSafty and PrivacyGraph Neural NetworkGraph
🎯 What it does: Built and implemented a framework that can extract scene graphs (SG) from sensor inputs and construct propositions using domain-specific languages and temporal logic to monitor the safety driving properties of autonomous vehicles.
Stein Variational Guided Model Predictive Path Integral Control: Proposal and Experiments with Fast Maneuvering Vehicles
Kohei Honda, Tatsuya Suzuki
CodeAutonomous DrivingOptimization
🎯 What it does: Proposes a novel stochastic optimal control method called SVG-MPPI based on MPPI to handle rapidly changing multi-modal optimal action distributions.
Stereo-LiDAR Depth Estimation with Deformable Propagation and Learned Disparity-Depth Conversion
Ang Li, Danping Zou
CodeDepth EstimationImagePoint Cloud
🎯 What it does: Proposed a stereo-LiDAR depth estimation network called SDG-Depth based on semi-dense hint guidance, which generates semi-dense hint maps and confidence maps using a deformable propagation module, and uses these maps to guide cost aggregation in stereo matching, while introducing a disparity-depth conversion module to reduce triangulation errors.
🎯 What it does: This paper proposes a multi-level LiDAR-camera fusion network called SuperFusion for generating high-precision semantic maps within short-range (up to 30 meters) and long-range (up to 90 meters) distances.
🎯 What it does: Developed Surgical Gym, a high-performance GPU-based platform for direct physics simulation and reinforcement learning on surgical robots;
The LuViRA Dataset: Synchronized Vision, Radio, and Audio Sensors for Indoor Localization
Ilayda Yaman, Liang Liu
CodePose EstimationSimultaneous Localization and MappingMultimodalityBenchmark
🎯 What it does: Proposed and released the LuViRA dataset, which synchronously records data from visual, RF, audio, and IMU sensors along with 6DOF pose ground truth.
Tight Motion Planning by Riemannian Optimization for Sliding and Rolling with Finite Number of Contact Points
Dror Livnat, Dan Halperin
CodeOptimizationRobotic Intelligence
🎯 What it does: Proposed a compact motion planning method using Riemannian optimization to achieve sliding and rolling, capable of navigating through narrow passages in configuration space and smoothly switching between free and semi-free regions.
Toward Optimal Tabletop Rearrangement with Multiple Manipulation Primitives
Baichuan Huang, Jingjin Yu
CodeOptimizationRobotic Intelligence
🎯 What it does: Proposed and studied an algorithm that uses multiple manipulation primitives (such as grasp-place and push) to plan high-quality action sequences for desktop rearrangement tasks
🎯 What it does: Propose a unified evaluation pipeline to compare traditional and end-to-end motion prediction methods under real perceptual inputs, and conduct an in-depth analysis of the performance gap between curated data and perceptual data.
TreeScope: An Agricultural Robotics Dataset for LiDAR-Based Mapping of Trees in Forests and Orchards
Derek Cheng, Vijay Kumar
CodeSegmentationDepth EstimationRobotic IntelligencePoint CloudBenchmarkAgriculture Related
🎯 What it does: Collected and released the TreeScope v1.0 dataset, which includes LiDAR data acquired from UAV and mobile robot platforms, 1,800+ trunk semantic labels, and measured tree diameters, along with benchmark scripts and baseline results of open-source algorithms.
TSCM: A Teacher-Student Model for Vision Place Recognition Using Cross-Metric Knowledge Distillation
Yehui Shen, Xieyuanli Chen
CodeRecognitionKnowledge DistillationImage
🎯 What it does: This paper proposes a teacher-student model TSCM, which achieves the visual place recognition (VPR) task by utilizing span-based knowledge distillation;
Uncertainty-aware 3D Object-Level Mapping with Deep Shape Priors
Ziwei Liao, Steven L. Waslander
CodePose EstimationOptimizationSimultaneous Localization and MappingImageMultimodality
🎯 What it does: Construct a high-quality object-level 3D map of unknown objects using multi-frame RGB-D images, outputting dense 3D shapes and a 9-degree-of-freedom (DOF) pose with three scale parameters, while modeling uncertainties in both shape and pose.
🎯 What it does: Propose the UncertaintyTrack extension, which leverages the localization uncertainty from probabilistic object detectors to enhance multi-object tracking
Unsupervised Spike Depth Estimation via Cross-modality Cross-domain Knowledge Transfer
Jiaming Liu, Shanghang Zhang
CodeDepth EstimationDomain AdaptationKnowledge DistillationImageMultimodalityBiomedical Data
🎯 What it does: Proposed an unsupervised spike depth estimation framework called BiCross, which achieves spike depth estimation using open-source RGB data through cross-modal cross-domain knowledge transfer.
Utilizing a Malfunctioning 3D Printer by Modeling Its Dynamics with Machine Learning
Renzo Caballero, Jürgen Schmidhuber
CodeRestorationRobotic IntelligenceWorld Model
🎯 What it does: Developed a method utilizing machine learning models to model the dynamics of failed 3D printers, enabling them to operate reliably even after damage and achieving self-repair.
Vehicle Behavior Prediction by Episodic-Memory Implanted NDT
Peining Shen, Jianru Xue
CodeAutonomous DrivingExplainability and Interpretability
🎯 What it does: By constructing a neural decision tree eMem-NDT based on text embedding hierarchical clustering, replacing the softmax layer of a pre-trained deep learning model, and utilizing memory prototype matching and leaf node link aggregation to achieve interpretable prediction of vehicle behavior.
VeloVox: A Low-Cost and Accurate 4D Object Detector with Single-Frame Point Cloud of Livox LiDAR
Tao Ma, Hongsheng Li
CodeObject DetectionPoint Cloud
🎯 What it does: Proposed a 4D object detector named VeloVox based on single-frame point clouds from Livox LiDAR, achieving accurate object detection and speed estimation.
Vision-based Wearable Steering Assistance for People with Impaired Vision in Jogging
Xiaotong Liu, Zhijun Li
CodeObject DetectionSafty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Developed a visual wearable turning assist system for visually impaired individuals, capable of detecting track lines and obstacles in athletic field environments, and achieving safe and rapid motion assistance through sampling and spline curve-based path planning.
Vision-Language Interpreter for Robot Task Planning
Keisuke Shirai, Shinsuke Mori
CodeRobotic IntelligenceTransformerLarge Language ModelMultimodality
🎯 What it does: Propose the Vision-Language Interpreter (ViLaIn) framework, which generates machine-readable problem description files for robot task planning by leveraging multimodal information and continuously optimizes them through error feedback.
VOLoc: Visual Place Recognition by Querying Compressed Lidar Map
Xudong Cai, Deying Li
CodeRetrievalCompressionSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Propose a vision-based place recognition method called VOLoc, which retrieves locations directly in compressed LiDAR maps using geometric similarity.
VOOM: Robust Visual Object Odometry and Mapping using Hierarchical Landmarks
Yutong Wang, Xieyuanli Chen
CodeSimultaneous Localization and MappingImage
🎯 What it does: Proposes a hierarchical landmark visual odometry and mapping framework called VOOM, which combines high-level objects and low-level feature points, and improves the observation model and data association method using dual quadric surfaces.
VPE-SLAM: Neural Implicit Voxel-permutohedral Encoding for SLAM
Zhiyao Zhang, Yulong Li
CodeSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Propose VPE-SLAM based on voxel-arranged octahedron encoding, achieving incremental dense SLAM reconstruction for unknown scenes and improving geometric accuracy
WeatherDepth: Curriculum Contrastive Learning for Self-Supervised Depth Estimation under Adverse Weather Conditions
Jiyuan Wang, Rui Ai
CodeDepth EstimationContrastive Learning
🎯 What it does: Propose WeatherDepth, which enhances the robustness of depth estimation in adverse weather through self-supervised curriculum contrastive learning.
WLST: Weak Labels Guided Self-training for Weakly-supervised Domain Adaptation on 3D Object Detection
T. Tsou, Winston H. Hsu
CodeObject DetectionDomain Adaptation
🎯 What it does: Propose the WLST framework, combining an autolabeler with self-training to achieve weakly supervised domain adaptation guided by weak labels;
🎯 What it does: Proposes a zero-training LiDAR-camera extrinsic calibration method that utilizes the Segment Anything Model (SAM) to automatically generate masks, and optimizes the extrinsic parameters by maximizing the consistency between point cloud attributes (intensity, normal vector, segmentation category) and the masks.
ZS6D: Zero-shot 6D Object Pose Estimation using Vision Transformers
P. Ausserlechner, Markus Vincze
CodePose EstimationTransformerImage
🎯 What it does: Proposes the ZS6D method, which utilizes a pre-trained Vision Transformer to extract visual descriptors, matches rendered templates with query images, establishes local correspondences, and estimates object 6D pose through RANSAC-PnP.