arXivSub Start free trial

ICRA 2024 Papers — Page 18

IEEE International Conference on Robotics and Automation · 1760 papers

VFAS-Grasp: Closed Loop Grasping with Visual Feedback and Adaptive Sampling

Pedro Piacenza, Volkan Isler

Robotic IntelligenceOptical FlowImage

🎯 What it does: Proposed a closed-loop robotic grasping planner VFAS-Grasp, which uses visual feedback and uncertainty-aware adaptive sampling to iteratively improve grasp candidate solutions.

VG4D: Vision-Language Model Goes 4D Video Recognition

Zhichao Deng, Mengyuan Liu

RecognitionTransformerVision Language ModelVideoPoint Cloud

🎯 What it does: Proposed the VG4D framework, which transfers knowledge from pre-trained vision-language models to 4D point cloud networks, and improved the dynamic point cloud backbone network, introducing im-PSTNet;

VICAN: Very Efficient Calibration Algorithm for Large Camera Networks

Gabriel Moreira, Alexander Hauptmann

Pose EstimationOptimizationSimultaneous Localization and MappingImage

🎯 What it does: Proposed an efficient calibration algorithm for large camera networks, utilizing movable rigid objects and camera-object relative transformations to construct a bipartite graph, and extending the Pose Graph Optimization (PGO) framework based on this; employing iterative primal-dual algorithms to handle large-scale graph structures;

VIDAR: Data Quality Improvement for Monocular 3D Reconstruction through In-situ Visual Interaction

Han Gao, Sheng Zhong

Depth EstimationData-Centric LearningMesh

🎯 What it does: Propose the VIDAR system, which uses real-time visual guidance through incremental reconstruction mesh visualization during the data acquisition phase of monocular 3D reconstruction to improve data quality.

VINSat: Solving the Lost-in-Space Problem with Visual-Inertial Navigation

Kyle McCleary, Brandon Lucia

OptimizationSimultaneous Localization and MappingImagePhysics Related

🎯 What it does: Proposed a visual-inertial navigation-based satellite orbit determination method called VINSat, addressing the satellite missing positioning problem.

VioLA: Aligning Videos to 2D LiDAR Scans

Jun-Jee Chao, Volkan Isler

Pose EstimationDepth EstimationVision Language ModelSimultaneous Localization and MappingVideoTextPoint Cloud

🎯 What it does: This paper proposes a method called VioLA that aligns video with panoramic 2D LiDAR scans. It first constructs a local semantic map from image sequences and extracts fixed-height points for registration with the LiDAR map. When reconstruction errors or insufficient camera coverage lead to insufficient registration information, a pre-trained text-image inpainting model and depth completion model are used for geometrically consistent scene completion to support pose registration.

ViPlanner: Visual Semantic Imperative Learning for Local Navigation

Pascal Roth, Marco Hutter

Autonomous DrivingOptimizationImage

🎯 What it does: Proposes ViPlanner, a local path planning method based on visual semantics and geometric information.

Virtual Borders in 3D: Defining a Drone’s Movement Space Using Augmented Reality

Malte Riechmann, Jan Rexilius

Robotic Intelligence

🎯 What it does: Using a tablet AR device to define 3D virtual boundaries to restrict the drone's movement space and change its navigation behavior;

Vision-based Tip Force Estimation on a Soft Continuum Robot

Xingyu Chen, T. G. Thuruthel

Robotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Proposed a soft continuous robot end-effector force estimation framework based on visual deformation, using camera images from arbitrary positions and drive inputs to predict the force applied at the end-effector

Vision-Based Water Clearance Determination in Maritime Environment

Carl H. Schiller, Stefano Maranò

Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkImage

🎯 What it does: Propose a visual method based on a single camera and semantic segmentation network, using back projection technology to calculate the water gap distance between the ship's hull and water surface obstacles, and validate its effectiveness on actual ships.

Vision-based Wearable Steering Assistance for People with Impaired Vision in Jogging

Xiaotong Liu, Zhijun Li

Object 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

Robotic 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.

Visual CPG-RL: Learning Central Pattern Generators for Visually-Guided Quadruped Locomotion

Guillaume Bellegarda, A. Ijspeert

OptimizationRobotic IntelligenceRecurrent Neural NetworkReinforcement Learning

🎯 What it does: Vision-guided quadruped gait coordination and collision avoidance were achieved by integrating external perception and a central pattern generator (CPG) into a deep reinforcement learning framework.

Visual Feedback Control of an Underactuated Hand for Grasping Brittle and Soft Foods

Ryogo Kai, Kazunori Umeda

Robotic IntelligenceImage

🎯 What it does: Proposes an underactuated hand control method using only a monocular camera without internal sensors

Visual Inertial Odometry using Focal Plane Binary Features (BIT-VIO)

Matthew Lisondra, Sajad Saeedi

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingImageMultimodality

🎯 What it does: Implement high-frame-rate visual-inertial odometry (BIT-VIO) using SCAMP-5, achieving 300 FPS visual odometry by executing visual algorithms on the image plane and combining 400 Hz IMU predictions to provide accurate and smooth trajectories

Visual Localization in Repetitive and Symmetric Indoor Parking Lots using 3D Key Text Graph

Joohyung Kim, N. Doh

Pose EstimationGraph Neural NetworkImageTextGraph

🎯 What it does: Proposed a visual localization method for repetitive and symmetric indoor parking lots, enhancing localization accuracy through four new modules.

Visual Noun Modifiers: The Problem of Binding Visual and Linguistic Cues*

Mohamadreza Faridghasemnia, Alessandro Saffiotti

RecognitionVision Language ModelMultimodality

🎯 What it does: Proposed an image description method based on nouns and modifiers, and constructed a new embedding binding space; studied the performance of existing models in identifying nouns and modifiers in images; introduced a new method by incorporating a dataset and CLIP-like recognition techniques (transfer learning and metric learning).

ViTacTip: Design and Verification of a Novel Biomimetic Physical Vision-Tactile Fusion Sensor

Wen Fan, Dandan Zhang

RecognitionPose EstimationRobotic IntelligenceGenerative Adversarial NetworkMultimodality

🎯 What it does: Designed and verified a novel bio-inspired perception sensor called ViTacTip that can simultaneously capture tactile and visual information in a single sensing unit;

VLEIBot: A New 45-mg Swimming Microrobot Driven by a Bioinspired Anguilliform Propulsor

Elijah K. Blankenship, N. O. P'erez-Arancibia

Robotic Intelligence

🎯 What it does: Proposed a 45 mg eel-like microrobot named VLEIBot, and achieved dual rudder control through its improved version VLEIBot+.

VLFM: Vision-Language Frontier Maps for Zero-Shot Semantic Navigation

Naoki Yokoyama, Bernadette Bucher

Robotic IntelligenceTransformerVision Language ModelSimultaneous Localization and MappingImageMultimodality

🎯 What it does: Proposed and implemented a zero-shot vision-language frontier map (VLFM) navigation method that can guide robots to search for target objects in unknown environments, with successful demonstration on the real robot Spot.

VO-Safe Reinforcement Learning for Drone Navigation

Feiqiang Lin, Ze Ji

Pose EstimationAutonomous DrivingReinforcement Learning

🎯 What it does: Propose a drone navigation method based on visual odometry, employing a hierarchical control architecture. A high-level policy trained with reinforcement learning (RL) generates waypoints, while low-level controllers execute path tracking.

VOLoc: Visual Place Recognition by Querying Compressed Lidar Map

Xudong Cai, Deying Li

RetrievalCompressionSimultaneous 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

Simultaneous 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

Simultaneous 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

VPRTempo: A Fast Temporally Encoded Spiking Neural Network for Visual Place Recognition

Adam D. Hines, Tobias Fischer

RecognitionSpiking Neural NetworkImage

🎯 What it does: Proposed a fast temporal coding spiking neural network called VPRTempo for visual place recognition, which can be trained within minutes and achieves query speeds in milliseconds.

VWDER:A Variable Wheel-Diameter Ellipsoidal Robot

Ziao Qin, Changrui Liu

Robotic Intelligence

🎯 What it does: Designed and experimentally verified a variable wheel diameter elliptical robot (VWDER), including its structural, kinematic, and dynamic analysis.

Wait, That Feels Familiar: Learning to Extrapolate Human Preferences for Preference-Aligned Path Planning

Haresh Karnan, Peter Stone

Representation LearningRobotic IntelligenceReinforcement Learning from Human FeedbackMultimodality

🎯 What it does: Propose and implement the PATERN framework, which achieves preference alignment in visual navigation by mapping robot inertial-proprioceptive-tactile measurements into a representation space and performing nearest neighbor search in this space to extrapolate the operator's preferences for visually novel terrains.

Walking-by-Logic: Signal Temporal Logic-Guided Model Predictive Control for Bipedal Locomotion Resilient to External Perturbations

Zhaoyuan Gu, Ye Zhao

OptimizationRobotic Intelligence

🎯 What it does: Propose a planning framework based on model predictive control (MPC) combined with signal temporal logic (STL) to achieve safe and task-compliant trajectory planning for bipedal robots when subjected to external impacts.

WARABI Hand: Five-fingered Robotic Hand with Flexible Skin and Force Sensors for Social Interaction

Aoi Nakane, Masayuki Inaba

Robotic Intelligence

🎯 What it does: Proposed a human-sized five-finger robot hand called WARABI Hand, equipped with multi-layer flexible rubber skin and phalangeal force sensors; conducted experiments where it grasped a forearm, shook hands, interlocked fingers, and evaluated grasping performance.

Watching the Air Rise: Learning-Based Single-Frame Schlieren Detection

Florian Achermann, Roland Siegwart

Data SynthesisConvolutional Neural NetworkSupervised Fine-TuningOptical FlowImagePhysics Related

🎯 What it does: This paper demonstrates the feasibility of predicting and visualizing schlieren-induced optical flow generated by thermal convection from a single grayscale image (captured by a moving camera in an unknown background) through training a convolutional neural network.

WAVE: An open-source underWater Arm-Vehicle Emulator

Marcus Rosette, Joseph R. Davidson

Robotic Intelligence

🎯 What it does: A 10-degree-of-freedom underwater manipulator-vehicle simulation platform named WAVE was developed to bridge the gap between simulation and real-world experiments, and was validated in a wave tank laboratory.

Waverider: Leveraging Hierarchical, Multi-Resolution Maps for Efficient and Reactive Obstacle Avoidance

Victor Reijgwart, Lionel Ott

Autonomous DrivingComputational EfficiencyRobotic Intelligence

🎯 What it does: Proposed an efficient, reactive obstacle avoidance system that combines hierarchical multi-resolution maps (wavemap) and Riemannian Motion Policies (RMP) algorithms, achieving low-latency, high-quality avoidance at hundreds of Hz speeds.

WayEx: Waypoint Exploration using a Single Demonstration

Mara Levy, Abhinav Shrivastava

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose a single demonstration goal control robot task learning method called WayEx, which significantly reduces training time by utilizing a waypoint exploration strategy;

WayFASTER: a Self-Supervised Traversability Prediction for Increased Navigation Awareness

M. V. Gasparino, Girish Chowdhary

Pose EstimationDepth EstimationAutonomous DrivingContrastive LearningVideoMultimodality

🎯 What it does: Propose a self-supervised neural network method for cross-temporal and spatial traversability prediction using RGB and depth image sequences along with pose estimation.

WayIL: Image-based Indoor Localization with Wayfinding Maps

Obin Kwon, Donghwan Lee

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Estimate the robot's pose in an abstract wayfinding map using RGB images captured by a perspective camera.

Weakly-Supervised Depth Completion during Robotic Micromanipulation from a Monocular Microscopic Image

Han Yang, Zhuoran Zhang

Depth EstimationRobotic IntelligenceImageBiomedical Data

🎯 What it does: Propose a weakly supervised deep completion network that generates dense depth maps in cell manipulation by utilizing monocular microscopy images and sparse depth information obtained through contact detection.

Wearable Haptics for a Marionette-inspired Teleoperation of Highly Redundant Robotic Systems

Davide Torielli, D. Prattichizzo

Robotic Intelligence

🎯 What it does: Developed a wearable haptic interface to enhance a teleoperation system based on the 'puppet' concept, and evaluated it with novice users

WeatherDepth: Curriculum Contrastive Learning for Self-Supervised Depth Estimation under Adverse Weather Conditions

Jiyuan Wang, Rui Ai

Depth EstimationContrastive Learning

🎯 What it does: Propose WeatherDepth, which enhances the robustness of depth estimation in adverse weather through self-supervised curriculum contrastive learning.

Weathering Ongoing Uncertainty: Learning and Planning in a Time-Varying Partially Observable Environment

Gokul Puthumanaillam, M. Ornik

Reinforcement Learning

🎯 What it does: Proposed the time-varying partially observable Markov decision process (TV-POMDP) and designed two core technologies: Memory Prioritized State Estimation (MPSE) and long-term reward optimization planning based on MPSE;

Weighting Online Decision Transformer with Episodic Memory for Offline-to-Online Reinforcement Learning

Xiao Ma, Wu-Jun Li

TransformerReinforcement Learning

🎯 What it does: Propose a method called WODTEM, which improves the sample efficiency of offline-to-online reinforcement learning by incorporating a weighted mechanism and episodic memory into the online decision Transformer with a replay buffer.

Well-Connected Set and Its Application to Multi-Robot Path Planning

Teng Guo, Jingjin Yu

OptimizationRobotic Intelligence

🎯 What it does: This paper proposes and studies the maximum well-connected set (LWCS) problem, explores its application in multi-robot path planning layout design, and designs optimal and approximate LWCS algorithms, while providing priority planning methods based on LWCS.

What Do We Learn from a Large-Scale Study of Pre-Trained Visual Representations in Sim and Real Environments?

S. Silwal, Oleksandr Maksymets

Domain AdaptationRepresentation LearningRobotic IntelligenceSupervised Fine-TuningImage

🎯 What it does: This paper conducts a large-scale empirical investigation of pre-trained visual representations (PVR) in downstream policy training for real-world tasks, encompassing five different PVRs, five manipulation or indoor navigation tasks corresponding to each PVR, and evaluation using three different robots and two policy learning paradigms.

What Matters for Active Texture Recognition With Vision-Based Tactile Sensors

Alina Böhm, Jan Peters

ClassificationRecognitionRobotic IntelligenceImage

🎯 What it does: This paper studies the active perception and classification of fabric textures using vision-based tactile sensors, formalizes the active sampling problem, and implements an information-theoretic exploration strategy based on predictive entropy and variance minimization; investigates the key factors for rapid and reliable texture recognition through ablation studies and human experiments; evaluates different neural network architectures, uncertainty representations, data augmentation, and dataset variability; and verifies the method on the published Active Clothing Perception Dataset and real robotic systems.

When Prolog Meets Generative Models: a New Approach for Managing Knowledge and Planning in Robotic Applications

Enrico Saccon, Luigi Palopoli

Robotic IntelligenceLarge Language ModelText

🎯 What it does: Proposes a Prolog-based knowledge representation system for robot applications, utilizing a specialized knowledge base organization method to achieve efficient knowledge base population, time-parallel plan generation, and automatic plan executability.

When to Replan? An Adaptive Replanning Strategy for Autonomous Navigation using Deep Reinforcement Learning

Kohei Honda, Tadashi Kozuno

Autonomous DrivingReinforcement Learning

🎯 What it does: This paper compares various common replanning strategies through large-scale simulation experiments, and based on the experimental results, proposes an adaptive replanning strategy based on deep reinforcement learning, which can learn appropriate replanning timing according to the environment and planning settings.

Whisker-Based Tactile Navigation Algorithm For Underground Robots

Tanel Kossas, M. Kruusmaa

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Explored the effectiveness of enhancing tactile perception and navigation in underground robots using artificial whisker sensors, developing and testing wall-following, Theta* navigation, and hybrid algorithms.

WiBot 1.0: A Modular Reconfigurable Glass Cleaning Robot for High-rise Buildings

S. A. Kariyawasam, U-Xuan Tan

Robotic IntelligenceImage

🎯 What it does: Developed and prototyped a modular reconfigurable glass cleaning robot named WiBot 1.0, which employs a kinematic chain with pneumatic joints and two rotational joints, each equipped with suction cups for mobility and adhesion. Experimental evaluations were conducted on its movement, window frame detection, and crossing-frame movement mechanisms.

Wind Field Modeling for Formation Planning in Multi-Drone Systems

Minhyuk Park, T. Au

OptimizationRobotic IntelligenceSupervised Fine-Tuning

🎯 What it does: This study extends the grid-based reservation system in a multi-quadcopter system by introducing non-exclusive reservations to address wind field effects, and proposes a novel formation planning algorithm that avoids collisions by adjusting the start time of motion plans.

Wireless Communication Infrastructure Building for Mobile Robot Search and Inspection Missions

Martin Zoula, J. Faigl

OptimizationRobotic Intelligence

🎯 What it does: By deploying relay nodes, a wireless communication infrastructure is constructed to support mobile robot search and inspection tasks, ensuring network connectivity.

WiTHy A*: Winding-Constrained Motion Planning for Tethered Robot using Hybrid A*

Vishnu S. Chipade, Sze Zheng Yong

OptimizationRobotic Intelligence

🎯 What it does: An improved Hybrid A* algorithm is developed to find the shortest path for a curvature-constrained robot with cable bundle constraints, satisfying user-defined wrapping angle limits under a fixed start point.

WLST: Weak Labels Guided Self-training for Weakly-supervised Domain Adaptation on 3D Object Detection

T. Tsou, Winston H. Hsu

Object 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;

WOMD-LiDAR: Raw Sensor Dataset Benchmark for Motion Forecasting

K. Chen, Drago Anguelov

Autonomous DrivingPoint CloudBenchmark

🎯 What it does: Added large-scale, high-quality, and diverse LiDAR point cloud data to the Waymo Open Motion Dataset (WOMD), constructing the WOMD-LiDAR dataset, and provided baseline models to study the role of LiDAR in motion prediction.

X-Tacformer : Spatio-tempral Attention Model for Tactile Recognition

Jiarui Hu, Bin He

RecognitionTransformerTime Series

🎯 What it does: Proposed the X-Tacformer model for tactile recognition of spatiotemporal sequences generated by sensor arrays

You Only Scan Once: A Dynamic Scene Reconstruction Pipeline for 6-DoF Robotic Grasping of Novel Objects

Lei Zhou, M. H. Ang

Pose EstimationRobotic IntelligenceSimultaneous Localization and MappingPoint CloudMesh

🎯 What it does: Through a two-stage dynamic scene reconstruction pipeline, real-time acquisition of the target object's 3D information and tracking of its pose, providing precise point cloud input for 6-DoF robotic grasping.

ZAPP! Zonotope Agreement of Prediction and Planning for Continuous-Time Collision Avoidance with Discrete-Time Dynamics

L. Paparusso, Marco Pavone

Autonomous DrivingOptimizationOrdinary Differential Equation

🎯 What it does: Proposed the ZAPP method, unifying prediction and planning under a single uncertainty representation framework

Zero-Shot Constrained Motion Planning Transformers Using Learned Sampling Dictionaries

Jacob J. Johnson, Michael C. Yip

Robotic IntelligenceTransformer

🎯 What it does: This paper proposes using a pre-trained VQ-MPT transformer model for constrained motion planning without retraining or fine-tuning, and improves the planning performance by updating the network output to bring the sampling region closer to the constraint manifold. Experiments conducted in simulation environments and on a real Franka Panda robot demonstrate improvements in planning time and path accuracy.

Zero-Shot Open-Vocabulary Tracking with Large Pre-Trained Models

Wen-Hsuan Chu, Katerina Fragkiadaki

Object DetectionObject TrackingSegmentationTransformerOptical FlowVideo

🎯 What it does: This paper combines a large-scale pre-trained open-vocabulary detector, segmenter, and dense optical flow estimator to build a model capable of tracking and segmenting any object in 2D videos.

Zero-Shot Wireless Indoor Navigation through Physics-Informed Reinforcement Learning

Mingsheng Yin, Quanyan Zhu

Reinforcement LearningPhysics Related

🎯 What it does: Proposed a physics-informed reinforcement learning (PIRL) framework for achieving zero-shot transfer in wireless indoor navigation.

Zero-training LiDAR-Camera Extrinsic Calibration Method Using Segment Anything Model

Zhaotong Luo, Botian Shi

Autonomous DrivingOptimizationTransformerImagePoint Cloud

🎯 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

Pose 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.