arXivSub Start free trial

IROS 2023 Papers — Page 5

IEEE/RSJ International Conference on Intelligent Robots and Systems · 1195 papers

ETAUS: An Edge and Trustworthy AI UAV System with Self-Adaptivity for Air Quality Monitoring

Chun-Hsian Huang, Ren Wang

ClassificationSafty and PrivacyComputational Efficiency

🎯 What it does: Developed an FPGA-based drone edge AI system called ETAUS for real-time air quality monitoring, which includes a custom neural engine for AQI level classification, a pre-trained model for detecting objects containing private information, and integrated de-identification, encryption functions, and a protection matrix.

Evaluation Metrics of Object Detection for Quantitative System-Level Analysis of Safety-Critical Autonomous Systems

Apurva Badithela, Richard M. Murray

Object DetectionAutonomous Driving

🎯 What it does: Proposes two evaluation metrics based on proposition marking and distance parameterized confusion matrices, and maps detection performance to formal specifications of a closed-loop system through a probabilistic model checking approach, calculating the probability that the system satisfies safety requirements; demonstrates the application of this method in a vehicle-pedestrian example.

Evaluation of a 7-DoFs Robotic Manipulator as Haptic Interface During Planar Reaching Tasks

A. Noccaro, Domenico Formica

Robotic Intelligence

🎯 What it does: Evaluate the applicability of a 7-degree-of-freedom robotic arm as a planar tactile interface for planar task accomplishment, measure human motion and force, and assess disturbances generated by the robot.

Evaluation of Underwater AprilTag Localization for Highly Agile Micro Underwater Robots

Nathalie Bauschmann, R. Seifried

Pose EstimationRobotic Intelligence

🎯 What it does: A method to evaluate the accuracy of underwater AprilTag localization is proposed, focusing on the performance of highly agile micro underwater robots under dynamic motion and calibration media. The experiments verify the importance of calibration and the linear decline of detection accuracy with increasing camera distance, while also identifying the camera that maximizes detection rate during high-speed motion.

EvCenterNet: Uncertainty Estimation for Object Detection Using Evidential Learning

Monish R. Nallapareddy, Abhinav Valada

Object DetectionAutonomous DrivingImagePoint Cloud

🎯 What it does: Propose the EvCenterNet framework, which directly estimates uncertainty in classification and regression using evidential learning, and enhances detection performance through active learning based on heatmap uncertainty.

Event Camera-Based Visual Odometry for Dynamic Motion Tracking of a Legged Robot Using Adaptive Time Surface

Shifan Zhu, Donghyun Kim

Pose EstimationRobotic IntelligenceSimultaneous Localization and MappingImageMultimodalityPoint Cloud

🎯 What it does: Propose a direct sparse visual odometry method combining event cameras and RGB-D for dynamic motion tracking of quadruped robot pose estimation.

EventTransAct: A Video Transformer-Based Framework for Event-Camera Based Action Recognition

Tristan de Blegiers, M. Shah

RecognitionTransformerContrastive LearningVideo

🎯 What it does: Proposed EventTransAct, an event camera action recognition framework based on video Transformer, utilizing spatial embeddings of event frames combined with temporal self-attention, and designing Event-Contrastive Loss along with event-specific data augmentation;

Eversion-Capable Fabric Robot Gripper with Novel Retraction Mechanism

Ahmed Hassan, K. Althoefer

Robotic Intelligence

🎯 What it does: Designed and tested a soft robotic gripper with fabric fingertips capable of expanding, bending, and retracting.

eViper: A Scalable Platform for Untethered Modular Soft Robots

Hsin Cheng, Minjie Chen

Robotic Intelligence

🎯 What it does: Developed an expandable electromagnetic oscillation intelligent piezoelectric soft robot, eViper, and its corresponding open-source simulation framework, SFERS, to study the impact of weight distribution and driving modes on motion in untethered modular soft robots.

evoBOT – Design and Learning-Based Control of a Two-Wheeled Compound Inverted Pendulum Robot

Patrick Klokowski, Sören Kerner

Domain AdaptationRobotic IntelligenceReinforcement LearningBenchmark

🎯 What it does: This paper introduces the evoBOT robot platform, elaborates on its mechanical and electronic design, and proposes a control method based on reinforcement learning for training high-dynamic actions, including steady-state balance and dynamic walking; extensive sim-to-real benchmark testing was conducted between simulation and real environments, along with an initial sim-to-real transfer process; the full robot simulation model is also made publicly available throughout the paper.

EVOLIN Benchmark: Evaluation of Line Detection and Association

K. Ivanov, A. Kornilova

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingImageBenchmark

🎯 What it does: Provides a complete visual line SLAM frontend benchmark, including line detection, line association, and pose error evaluation for RGB and RGB-D images.

Evolutionary-Based Online Motion Planning Framework for Quadruped Robot Jumping

Linzhu Yue, Yunhui Liu

OptimizationRobotic Intelligence

🎯 What it does: Proposed an online jumping motion planning framework based on differential evolution (DE), Latin hypercube sampling, and configuration space (DLC).

Evolving Physical Instinct for Morphology and Control Co-Adaption

Xinglin Chen, Wenjing Yang

Computational EfficiencyRobotic Intelligence

🎯 What it does: Proposed an evolvable instinctive controller to enhance the co-evolution of robot morphology and control, implementing a phase-based finite state machine controller for multi-legged gait control, and conducting experiments on different morphological prototypes using GPU parallel simulation.

Exact Point Cloud Downsampling for Fast and Accurate Global Trajectory Optimization

Kenji Koide, A. Banno

Autonomous DrivingOptimizationPoint Cloud

🎯 What it does: Proposed a point cloud downsampling algorithm that selects a weighted residual subset, making the quadratic registration error function at evaluation points identical to the original point cloud, thereby achieving fast and accurate trajectory optimization.

Experimental Evaluation of a Transparent Operation Mode for a Lower-Limb Exoskeleton Designed for Children with Cerebral Palsy

R. Andrade, Paolo Bonato

Robotic Intelligence

🎯 What it does: A control strategy with a transparent operation mode was implemented and tested on the lower-limb exoskeleton ExoRoboWalker for children with cerebral palsy.

Exploiting Spatio-Temporal Human-Object Relations Using Graph Neural Networks for Human Action Recognition and 3D Motion Forecasting

Dimitrios Lagamtzis, Steffen Schober

RecognitionGraph Neural NetworkVideo

🎯 What it does: Proposed a graph neural network (GNN)-based architecture that integrates human action recognition and motion prediction in industrial human-robot collaboration (HRC) environments, using human-centered 3D information and object labels to construct graph structures.

Exploiting Task Tolerances in Mimicry-Based Telemanipulation

Yeping Wang, M. Gleicher

Robotic Intelligence

🎯 What it does: Implemented a remote operating system that allows robots to autonomously adjust within task tolerance ranges, and compared two teleoperation paradigms: functional imitation and precise imitation.

Exploiting the Kinematic Redundancy of a Backdrivable Parallel Manipulator for Sensing During Physical Human-Robot Interaction

Arda Yiğit, Clément Gosselin

Robotic Intelligence

🎯 What it does: Leveraging the motion redundancy of parallel robots under back-driving conditions, inferring the operator's intention through the use of only motor encoders and forward kinematics solving, and switching between a position controller and a gravity-compensated free-motion guide controller.

Exploring Kinodynamic Fabrics for Reactive Whole-Body Control of Underactuated Humanoid Robots

Alphonsus Adu-Bredu, J. Grizzle

OptimizationRobotic Intelligence

🎯 What it does: Propose and implement the Kinodynamic Fabrics method for real-time multi-task whole-body control in bipedal humanoid robots, evaluated in both simulation and real-world environments.

Exploring Learning-Based Control Policy for Fish-Like Robots in Altered Background Flows

Xiaozhu Lin, Yang Wang

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose a learning-based control framework that utilizes the FishGym simulation system to train fish-like robots to perform motion control tasks such as approaching targets and maintaining position in non-stationary, unknown background flows, and applies the learned control policies combined with an online estimator to path tracking tasks.

Exploring Social Motion Latent Space and Human Awareness for Effective Robot Navigation in Crowded Environments

J. Ansari, Brojeshwar Bhowmick

Autonomous DrivingRepresentation LearningRobotic Intelligence

🎯 What it does: Learning to utilize the social motion latent space to generate robot control for achieving more effective social navigation

Exploring Visual Pre-training for Robot Manipulation: Datasets, Models and Methods

Ya Jing, Tao Kong

Representation LearningRobotic IntelligenceContrastive Learning

🎯 What it does: Investigate the effectiveness of visual pre-training in robot manipulation tasks and propose the Vi-PRoM approach

Expressing and Inferring Action Carefulness in Human-to-Robot Handovers

Linda Lastrico, J. Santos-Victor

Robotic IntelligenceTime SeriesSequential

🎯 What it does: Implemented an online classifier to distinguish human carefulness during cup handling, and designed an 'expressive' robot controller enabling the robot's motion to convey information based on whether the cup is full or empty, while comparing with a neutral controller.

Extensions to Dynamically-Consistent Collision Reaction Control for Collaborative Robots

Marie Harder, Alexander Dietrich

Robotic Intelligence

🎯 What it does: Designed and verified a dynamically consistent collision reaction controller, achieving reaction motion in specific directions in Cartesian space, and compared with traditional methods.

External Sensor-Less in-Hand Object Position Manipulation for an Under-Actuated Hand Using Data-Driven-Based Methods to Compensate for the Nonlinearity of Self-Locking Mechanism

Hackley Doan, Kenji Tahara

Robotic Intelligence

🎯 What it does: Apply a hybrid analytical model and data-driven methods to analyze internal sensor data, compensate for nonlinear constraints of the self-locking mechanism, propose an object position manipulation framework within the hand, and conduct experimental validation.

Extracting Dynamic Navigation Goal from Natural Language Dialogue

Lanjun Liang, Huaping Liu

Robotic IntelligenceSimultaneous Localization and MappingText

🎯 What it does: Extract dynamic position change information from group chat dialogues to construct a Dynamic Spatiotemporal Map (DSTM), and use this map to assist mobile robots in performing instruction-based navigation tasks within university buildings to locate target individuals.

Extrinsic Calibration of Camera to LIDAR Using a Differentiable Checkerboard Model

L. Fu, Maurice F. Fallon

Pose EstimationAutonomous DrivingOptimizationImagePoint Cloud

🎯 What it does: Proposes a method to achieve external calibration between a camera and LiDAR using only a standard chessboard.

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

Ekta U. Samani, A. Banerjee

GenerationData 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.

FABRIKv: A Fast, Iterative Inverse Kinematics Solver for Surgical Continuum Robot with Variable Curvature Model

Fuhao Wang, Xiuhong Tang

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: Proposes a fast inverse kinematics solver FABRIKv for a surgical continuum robot with a variable curvature model, first analyzing the robot's deformation and presenting a representation method for the variable curvature model, then improving the FABRIK algorithm to maintain real-time performance under load and correct deformation.

Falcon: A Wide-and-Deep Onboard Active Vision System

Masahiro Hirano, Yuji Yamakawa

Object TrackingAutonomous DrivingImageVideo

🎯 What it does: Designed and implemented an active vision system called Falcon, integrating an electric zoom lens, high-speed camera, and servo mirrors, achieving high-resolution imaging from close to distant ranges. Proposed a mapping-based external camera calibration method and a lightweight visual feedback algorithm for object tracking. Subsequently validated the system performance in indoor experiments and achieved continuous high-resolution imaging of curved mirrors during vehicle movement.

Fast Asymptotically Optimal Path Planning in Dynamic, Uncertain Environments

Lu Huang, Xingjian Jing

Optimization

🎯 What it does: Propose Fast Adaptive Tree (FAT), a sampling-based path planner that is incrementally optimal in dynamic uncertain environments.

Fast Bi-Monocular Visual Odometry Using Factor Graph Sparsification

César Debeunne, Damien Vivet

Pose EstimationOptimizationComputational Efficiency

🎯 What it does: Proposes an indirect dual monocular visual odometry method based on sliding window optimization, aiming to maintain problem sparsity and reduce computational burden under low-light conditions, applicable to scenarios such as lava tunnel exploration.

Fast Decision Support for Air Traffic Management at Urban Air Mobility Vertiports Using Graph Learning

Prajit KrisshnaKumar, Souma Chowdhury

OptimizationGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: Proposed a graph reinforcement learning-based urban aerial mobility (UAM) vertical runway scheduling management (UAM-VSM) method, which uses graph convolutional networks (GCN) to extract features of runway positions and aircraft, and employs perceptron layers to determine actions such as staying, cruising, taking off, or landing.

Fast Point to Mesh Distance by Domain Voxelization

Geordan Gutow, H. Choset

Computational EfficiencyMesh

🎯 What it does: Proposed a voxel-based point-to-triangle mesh distance calculation method

FATROP: A Fast Constrained Optimal Control Problem Solver for Robot Trajectory Optimization and Control

Lander Vanroye, W. Decré

OptimizationRobotic Intelligence

🎯 What it does: Proposed a fast constrained optimal control problem solver named FATROP for robot trajectory optimization and control.

FeatDANet: Feature-level Domain Adaptation Network for Semantic Segmentation

Jiao Li, Jiamao Li

SegmentationDomain AdaptationConvolutional Neural NetworkImage

🎯 What it does: Proposes an unsupervised domain adaptation network called FeatDANet, which focuses on aligning feature-level domain distributions at each encoder layer to achieve semantic segmentation transfer from synthetic data to real data.

Feature Explanation for Robust Trajectory Prediction

Xukai Zhai, Zhishuai Yin

Autonomous DrivingExplainability and InterpretabilityTransformerTime SeriesBenchmark

🎯 What it does: Proposed and implemented a Parallel Explainable Transformer (PXT) framework that uses a dual-branch encoder to separate road information from historical trajectories, and employs explainability methods to select the most contributing features for trajectory prediction.

Feature-based Visual Odometry for Bronchoscopy: A Dataset and Benchmark

Jianning Deng, Mohsen Khadem

Pose EstimationConvolutional Neural NetworkSimultaneous Localization and MappingVideoBiomedical DataBenchmark

🎯 What it does: Developed a bronchoscopy visual odometry dataset containing 34 video clips and over 23,000 frames, along with benchmark experiments.

Feedback Motion Prediction for Safe Unicycle Robot Navigation

Aykut Isleyen, Ömür Arslan

Robotic Intelligence

🎯 What it does: Proposes a novel conical feedback motion prediction method for two-wheel differential robots (modeled as a kinematic unicycle model), and applies it to reference governors to achieve safe obstacle navigation;

Few-Shot Segmentation and Semantic Segmentation for Underwater Imagery

Imran Kabir, Md. Alimoor Reza

SegmentationMeta 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

Finding Biomechanically Safe Trajectories for Robot Manipulation of the Human Body in a Search and Rescue Scenario

Elizabeth Peiros, Michael C. Yip

OptimizationSafty and PrivacyRobotic IntelligenceBiomedical Data

🎯 What it does: Designed and verified a robot trajectory planning method considering biomechanical safety constraints in search and rescue scenarios, for repositioning limbs of unconscious human victims to achieve safe extraction.

Finding the Goal: Insect-Inspired Spiking Neural Network for Heading Error Estimation

Thorben Schoepe, Elisabetta Chicca

Robotic IntelligenceSpiking Neural Network

🎯 What it does: Proposed and implemented a spiking neural network model based on insect navigation for estimating heading error, and validated its functionality on a simulated robot platform.

FingerTac - An Interchangeable and Wearable Tactile Sensor for the Fingertips of Human and Robot Hands

Prathamesh Sathe, Shigeki Sugano

Robotic Intelligence

🎯 What it does: Developed a wearable fingertip tactile sensor that is interchangeable between human and robotic fingers and can capture triaxial force vectors on the fingertips

FISS+: Efficient and Focused Trajectory Generation and Refinement Using Fast Iterative Search and Sampling Strategy

Shuo Sun, Marcelo H. Ang

Autonomous DrivingOptimizationComputational EfficiencyBenchmark

🎯 What it does: Proposes a two-stage coarse-to-fine sampling-based trajectory planning framework

Flexible Gear Assembly with Visual Servoing and Force Feedback

J. Ming, M. Caccamo

Object DetectionData SynthesisRobotic IntelligenceConvolutional Neural NetworkReinforcement LearningImage

🎯 What it does: This paper proposes a visual-guided two-stage gear assembly method with force feedback, first using YOLO for rough localization of the workpiece, and then using deep reinforcement learning to complete the insertion.

Flexible Handover with Real-Time Robust Dynamic Grasp Trajectory Generation

Gu Zhang, Cewu Lu

Robotic IntelligenceBenchmark

🎯 What it does: Proposed a flexible hand interaction method with high success rates, enabling robots to grasp objects in complex continuous motion scenarios.

Flexible Multi-DoF Aerial 3D Printing Supported with Automated Optimal Chunking

Marios-Nektarios Stamatopoulos, G. Nikolakopoulos

OptimizationRobotic IntelligencePhysics Related

🎯 What it does: Proposes a block-based distributed 3D printing framework for unmanned aerial vehicles (UAVs), which divides models into manageable blocks through an optimization process, allocates them to UAVs for partial printing, and achieves fully autonomous operation.

FM-Loc: Using Foundation Models for Improved Vision-Based Localization

Reihaneh Mirjalili, Wolfram Burgard

Pose EstimationTransformerLarge Language ModelVision Language ModelImage

🎯 What it does: Proposes an FM-Loc visual localization method based on foundation models (GPT-3 and CLIP), constructing semantic image descriptors to enhance the robustness of indoor visual localization

FogROS2-SGC: A ROS2 Cloud Robotics Platform for Secure Global Connectivity

Kai-Peng Chen, Ken Goldberg

Safty and PrivacyRobotic Intelligence

🎯 What it does: Introduce and implement FogROS2-SGC, a ROS2 cloud robotics platform that enables secure global connectivity across different physical locations, networks, and DDS implementations.

Force Map: Learning to Predict Contact Force Distribution from Vision

Ryo Hanai, T. Ogata

Data SynthesisRobotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Utilize visual prediction of contact force distribution (force map) to plan the lifting direction, thereby reducing disturbances in stacked objects.

Force-Based Pose Regulation of a Cable-Suspended Load Using UAVs with Force Bias

C. Gabellieri, Antonio Franchi

OptimizationRobotic Intelligence

🎯 What it does: Studied the impact of force measurement/estimation errors on force-based cooperative control of a beam-shaped load suspended by two drones via cables; first calculated the system's equilibrium configuration; then proved that inducing internal forces within the load can enhance the robustness of the load's attitude error to force deviations; finally proposed a method to achieve zero load position error.

Formal Composition of Robotic Systems as Contract Programs

Mason Nakamura, Stuart Russell

OptimizationRobotic Intelligence

🎯 What it does: Proposes a meta-reasoning framework that formalizes robot systems as contract programs, incorporating programming constructs with functional, conditional, and cyclic semantics.

Forward/Inverse Kinematics Modeling for Tensegrity Manipulator Based on Goal-Conditioned Variational Autoencoder

Yuhei Yoshimitsu, Shuhei Ikemoto

Robotic IntelligenceAuto Encoder

🎯 What it does: A data-driven approach is used to model the forward and inverse kinematics of a high-redundancy tension structure manipulator, training a VAE-based kinematic model and extracting sub-networks for forward kinematics, inverse kinematics, and null space.

FPECMV: Learning-Based Fault-Tolerant Collaborative Localization Under Limited Connectivity

Rong Ou, Tin Lun Lam

Convolutional Neural NetworkSimultaneous Localization and Mapping

🎯 What it does: Propose an FPECMV algorithm for fault-tolerant collaborative localization under limited connectivity;

FPGADDS: An Intra-FPGA Data Distribution Service for ROS 2 Robotics Applications

Christian Lienen, M. Platzner

Autonomous DrivingRobotic Intelligence

🎯 What it does: Designed and implemented fpgaDDS, a lightweight data distribution service for hardware-mapped ROS 2 nodes, and evaluated its performance in examples and autonomous driving case studies.

FRoGGeR: Fast Robust Grasp Generation via the Min-Weight Metric

Albert Li, A. Ames

OptimizationRobotic IntelligenceMesh

🎯 What it does: Fast generation of robust and precise grasping plans

From “Thumbs Up” to “10 out of 10”: Reconsidering Scalar Feedback in Interactive Reinforcement Learning

Hang Yu, E. Short

Reinforcement Learning from Human FeedbackReinforcement Learning

🎯 What it does: Compared the effects of scalar feedback and binary feedback in interactive reinforcement learning, and proposed the STEADY method to enhance the learning performance of scalar feedback.

From Crowd Motion Prediction to Robot Navigation in Crowds

S. Poddar, S. Srinivasa

OptimizationRobotic IntelligenceGenerative Adversarial Network

🎯 What it does: Integrate the S-GAN motion prediction model into a model predictive controller (MPC) in the laboratory, deploy it on a self-balancing robot, and test its navigation performance under different crowd behaviors.

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

Kévin Riou, P. Callet

Robotic 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.

Fully Proprioceptive Slip-Velocity-Aware State Estimation for Mobile Robots via Invariant Kalman Filtering and Disturbance Observer

Xihang Yu, Maani Ghaffari

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Developed a full slip velocity perception state estimator based on complete self-perception, utilizing invariant observer design theory and disturbance observer (DOB) to achieve real-time estimation of slip velocity in mobile robots.

FUSE-D: Framework for UAV System-Parameter Estimation with Disturbance Detection

Christoph Böhm, Stephan Weiss

Anomaly DetectionAutonomous Driving

🎯 What it does: Propose the FUSE-D framework, integrating online system identification, sensor self-calibration, and external disturbance analysis. Based on a rotor speed-level system model and a single global attitude sensor, it estimates rotor geometry, mass, moment of inertia, rotor aerodynamic characteristics, as well as external forces and their points of application.

Fusing Visual Appearance and Geometry for Multi-Modality 6DoF Object Tracking

Manuel Stoiber, Rudolph Triebel

Object TrackingPose EstimationVideoMultimodality

🎯 What it does: Developed a multi-modal 6DoF object tracker that integrates visual appearance and geometric information for real-time target pose estimation.

FVLoc-NeRF : Fast Vision-Only Localization within Neural Radiation Field

Wenzhi Guo, Lijun Chen

Pose EstimationRetrievalRobotic IntelligenceNeural Radiance FieldImage

🎯 What it does: Propose a fast visual localization framework called FVLoc-NeRF that uses only RGB monocular images, leveraging NeRF to encode 3D geometry and environmental appearance for robot localization.

Game-Theoretical Approach to Multi-Robot Task Allocation Using a Bio-Inspired Optimization Strategy

Shengkang Chen, Fumin Zhang

OptimizationRobotic Intelligence

🎯 What it does: Propose a game theory-based multi-robot task allocation method, considering self-interested robots that do not share utility functions, and design a utility function incorporating conflict penalties and path costs. Task allocation is achieved through consensus communication and SUSD search.

GAPSLAM: Blending Gaussian Approximation and Particle Filters for Real-Time Non-Gaussian SLAM

Qiangqiang Huang, John J. Leonard

OptimizationSimultaneous Localization and MappingTabular

🎯 What it does: Combine Gaussian approximation with particle filters to real-time infer the edge posterior distribution in SLAM, enabling the expression and computational scalability of non-Gaussian posteriors.

gatekeeper: Online Safety Verification and Control for Nonlinear Systems in Dynamic Environments

Devansh R. Agrawal, Dimitra Panagou

🎯 What it does: Developed a real-time, lightweight gatekeeper algorithm integrated into path planning and feedback controllers to verify the safety executability of the trajectory

Gaussian Max-Value Entropy Search for Multi-Agent Bayesian Optimization

Haitong Ma, Na Li

Optimization

🎯 What it does: Proposed and implemented Gaussian Max-value Entropy Search for multi-agent Bayesian optimization.

GelSight Svelte: A Human Finger-Shaped Single-Camera Tactile Robot Finger with Large Sensing Coverage and Proprioceptive Sensing

Jialiang Zhao, E. Adelson

Robotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Propose GelSight Svelte, a curved-surface, human-like fingertip, single-camera tactile sensor, and estimate fingertip bending and torsion moments using convolutional neural networks, validating its tactile and proprioceptive capabilities through colloidal deformation experiments and grasping tasks.

Generalized Few-shot Semantic Segmentation for LiDAR Point Clouds

Pengze Wu, Yu Hu

SegmentationRepresentation LearningMeta LearningPoint Cloud

🎯 What it does: Proposes a generic few-shot semantic segmentation method based on LiDAR point cloud data that can simultaneously predict base classes and novel classes.

Generalized Robot Dynamics Learning and Gen2Real Transfer

Dengpeng Xing, Bo Xu

Knowledge DistillationRobotic IntelligenceTransformerSequential

🎯 What it does: Learned a general model covering a large variety of robot dynamics, and proposed the Gen2Real method to transfer the general model generated in simulation to specific physical robots.

Generalizing Surgical Instruments Segmentation to Unseen Domains with One-to-Many Synthesis

An-Chi Wang, Hongliang Ren

SegmentationData SynthesisDomain AdaptationImageBiomedical Data

🎯 What it does: Synthetic surgical instrument segmentation datasets are synthesized using minimal source images through transformations, pooling, and various hybrid techniques, and models are trained on this dataset to improve generalization performance on unseen domains.

Generating Executable Action Plans with Environmentally-Aware Language Models

Maitrey Gramopadhye, D. Szafir

GenerationRobotic 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.

Generating Scenarios from High-Level Specifications for Object Rearrangement Tasks

S. V. Waveren, Danica Kragic

GenerationData SynthesisRobotic Intelligence

🎯 What it does: Generate training scenarios for object rearrangement tasks based on high-level specifications, sort scenarios according to difficulty, and use a generative model based on spatial logic specifications to produce spatially structured scenarios that meet specifications and desired difficulty levels.

Generation of Time-Varying Impedance Attacks Against Haptic Shared Control Steering Systems

Alireza Mohammadi, Hafiz Abid Mahmood Malik

Autonomous DrivingAdversarial Attack

🎯 What it does: Studies the attack methods where attackers exploit time-varying impedance to disrupt human-vehicle interaction dynamics, leading to instability in the vehicle steering system.

Geometric Fault-Tolerant Control of Quadrotors in Case of Rotor Failures: An Attitude Based Comparative Study

Jennifer Yeom, Giuseppe Loianno

Robotic Intelligence

🎯 What it does: Propose a fault-tolerant control strategy for quadrotors applicable to both single and dual full rotor failures, enhanced based on the classical geometric tracking controller on SO(3)×ℝ^3.

Geometric Gait Optimization for Inertia-Dominated Systems with Nonzero Net Momentum

Yanhao Yang, Ross L. Hatton

OptimizationPhysics Related

🎯 What it does: Proposed a kinematic and momentum dual gait optimization algorithm for inertial dominance systems with non-zero net momentum, demonstrating its effectiveness in forward and turning motions in systems with/without fluid added mass.

Geometrically Consistent Monocular Metric-Semantic 3D Mapping for Indoor Environments with Transparent and Reflecting Objects

M. Mohrat, S. Kolyubin

SegmentationPose EstimationDepth EstimationConvolutional Neural NetworkSimultaneous Localization and MappingPoint CloudMesh

🎯 What it does: Developed a geometric consistency metric semantic 3D mapping pipeline based on a monocular camera, capable of generating high-quality point clouds in indoor environments containing transparent and reflective objects.

Global Localization in Unstructured Environments Using Semantic Object Maps Built from Various Viewpoints

Jacqueline Ankenbauer, J. How

Autonomous DrivingOptimizationGraph Neural NetworkSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Proposes a framework for achieving vehicle global localization and guided relocalization in unstructured environments by associating and registering the vehicle's local semantic object maps with compact semantic reference maps.

Global Localization: Utilizing Relative Spatio-Temporal Geometric Constraints from Adjacent and Distant Cameras

Mohammad Altillawi, Ziyuan Liu

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Leverages relative spatiotemporal geometric constraints from adjacent and distant cameras to estimate 6-DoF camera pose from a single image using a deep network.

Global Map Assisted Multi-Agent Collision Avoidance via Deep Reinforcement Learning around Complex Obstacles

Yuanyuan Du, Shuguang Cui

Autonomous DrivingRobotic IntelligenceReinforcement LearningWorld Model

🎯 What it does: Proposed a global map-assisted multi-agent collision avoidance algorithm that utilizes a distance map to simultaneously consider obstacles and other agents.

GloPro: Globally-Consistent Uncertainty-Aware 3D Human Pose Estimation & Tracking in the Wild

Simon Schaefer, Stefan Leutenegger

Object TrackingPose EstimationMesh

🎯 What it does: Proposes the GloPro framework, which can predict the uncertainty distribution of 3D body meshes that include body shape, pose, and root position, achieving this through an effective fusion of visual cues and learned motion models.

GOATS: Goal Sampling Adaptation for Scooping with Curriculum Reinforcement Learning

Yaru Niu, Liangjun Zhang

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed and implemented a robot water scooping task based on goal-conditioned reinforcement learning, and designed the GOATS curriculum reinforcement learning method.

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

Ihab S. Mohamed, Lantao Liu

Autonomous 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.

GraNet: A Multi-Level Graph Network for 6-DoF Grasp Pose Generation in Cluttered Scenes

Haowen Wang, Chungang Zhuang

Pose EstimationRobotic IntelligenceGraph Neural NetworkPoint Cloud

🎯 What it does: Propose GraNet—a 6-DoF object-agnostic grasping pose generation framework based on multi-layer graph networks, which converts point cloud scenes into multi-layer graphs and propagates features through graph neural networks.

Graph Matching Optimization Network for Point Cloud Registration

Qianliang Wu, Jian Yang

Pose EstimationOptimizationPoint Cloud

🎯 What it does: Proposed a graph matching optimization-based network (GMONet), which explicitly imposes isometric constraints in the dense layer via a graph matching optimizer to enhance feature representation for point cloud registration;

Graph-Based Global Robot Localization Informing Situational Graphs with Architectural Graphs

Muhammad Shaheer, Holger Voos

Robotic IntelligenceGraph Neural NetworkSimultaneous Localization and MappingPoint CloudGraph

🎯 What it does: Proposes a scheme for multi-legged robot localization using architectural floor plans. First, the architectural floor plan is converted into an architectural graph (A-Graph). Subsequently, during robot movement, the robot estimates an online situational graph (S-Graph) based on sensor data. By performing graph-to-graph matching, the S-Graph is aligned and merged with the A-Graph to obtain an information-rich situational graph (is-Graph), thereby achieving global localization.

Graph-Based View Motion Planning for Fruit Detection

Tobias Zaenker, Maren Bennewitz

Graph Neural NetworkAgriculture Related

🎯 What it does: Proposes a graph-based perspective motion planner for effectively monitoring and discovering fruits in bell pepper plants.

Graph-Based Visual-Kinematic Fusion and Monte Carlo Initialization for Fast-Deployable Cable-Driven Robots

R. Khorrambakht, Stephan Weiss

Robotic IntelligenceSimultaneous Localization and MappingImageTime Series

🎯 What it does: A unified localization and calibration system based on statistical fusion was constructed using vehicle-mounted cameras and kinematic sensors, employing Monte Carlo initialization and factor graph representation to simultaneously identify kinematic parameters, visual odometry scale, and their uncertainty.

Grasp Region Exploration for 7-DoF Robotic Grasping in Cluttered Scenes

Zibo Chen, Wei-Shi Zheng

Pose EstimationRobotic IntelligencePoint Cloud

🎯 What it does: Proposes a grasp region exploration module and a grasp region attention module to enhance point cloud density around grasp points in cluttered scenes and dynamically aggregate features, thereby improving the grasping performance of 7-DoF robotic hands.

Grasp Stability Assessment Through Attention-Guided Cross-Modality Fusion and Transfer Learning

Zhuangzhuang Zhang, Q. Cao

Domain AdaptationRobotic IntelligenceConvolutional Neural NetworkTransformerMultimodality

🎯 What it does: Proposed an attention-guided cross-modal fusion architecture to comprehensively integrate visual and tactile features for evaluating grasp stability

Grasp State Classification in Agricultural Manipulation

Benjamin Walt, Girish Krishnan

ClassificationTime SeriesAgriculture Related

🎯 What it does: Studied different grasping states (successful grasping, sliding, grasping failure) during fruit harvesting in agricultural environments, and constructed a learning-based classifier using low-cost sensors (IMU and IR reflection).

GVCCI: Lifelong Learning of Visual Grounding for Language-Guided Robotic Manipulation

Junghyun Kim, Byoung-Tak Zhang

Object DetectionRobotic IntelligenceMeta LearningVision-Language-Action ModelImageTextMultimodality

🎯 What it does: Proposed the GVCCI lifelong learning framework, which utilizes object detection to generate synthetic instructions and continuously trains a visual localization model without human supervision, aiming to enhance language-guided robot manipulation performance.

HALO: A Safe, Coaxial, and Dual-Ducted UAV Without Servo

Haotian Li, Fu Zhang

🎯 What it does: Designed and verified a novel drone HALO, utilizing a swashplateless mechanism to achieve pitch and roll control, and employing a coaxial dual-channel ducted fan design to enhance safety and aerodynamic efficiency.

Hand Design Approach for Planar Fully Actuated Manipulators

Keegan Nave, Cindy Grimm

Robotic IntelligenceBenchmark

🎯 What it does: Proposed a fully active planar manipulator design method tailored for specific grasping actions

HANDAL: A Dataset of Real-World Manipulable Object Categories with Pose Annotations, Affordances, and Reconstructions

Andrew Guo, Stan Birchfield

Pose EstimationRobotic IntelligenceImageVideoMeshBenchmark

🎯 What it does: Created the HANDAL dataset for category-level object pose estimation and functional prediction, focusing on tool-like objects graspable by robots. It employs a single camera and semi-automated processing to achieve high-quality 3D annotations, providing 308k frames, 212 objects, 17 categories, and full 3D reconstructed meshes.

Haptic Dataset Augmentation with Subjective QoE Labels using Conditional Generative Adversarial Network

Zican Wang, Eckehard G. Steinbach

Data SynthesisGenerative Adversarial Network

🎯 What it does: Proposed a GAN-based generative method to automatically expand the subjective tactile QoE dataset, avoiding time-consuming experiments.

Hardware-in-the-Loop Simulation of Vehicle-Manipulator Systems for Physical Interaction Tasks

Hemjyoti Das, Christian Ott

Robotic IntelligencePhysics Related

🎯 What it does: A method for impedance matching is proposed to match the dynamics of the end-effector of a fixed-base robotic arm with the dynamics of the target vehicle-robotic arm system (VMS) in hardware-in-the-loop simulation, considering redundant null-space dynamics to ensure that the torque applied to the environment is consistent with the simulation system, and the method is validated using a suspended drone robotic arm.

Harnessing the Power of Human Biomechanics in Force-Position Domain: A 3D Passivity Index Map for Upper Limb Physical Human-(Tele) Robot Interaction

Xingyuan Zhou, S. F. Atashzar

Robotic IntelligenceBiomedical Data

🎯 What it does: This paper constructs a three-dimensional inability index map, quantitatively evaluates the inability margin of the upper limb in the force-position domain, and explores the effects of synergistic contraction levels, interaction frequency, and geometric direction on the inability margin.

Helical Propulsion in Low-Re Numbers with Near-Zero Angle of Attack

Leendert-Jan W. Ligtenberg, Islam S. M. Khalil

Physics Related

🎯 What it does: By establishing a simplified one-dimensional model of UHMD under low Reynolds numbers, the gap with rotating permanent magnets is predicted and adjusted to achieve constrained behavior at near-zero angles of attack, and the method's feasibility for zero-sinking swimming is validated through numerical simulations and experiments.

HELSA: Hierarchical Reinforcement Learning with Spatiotemporal Abstraction for Large-Scale Multi-Agent Path Finding

Zhaoyi Song, Xiang Cheng

Reinforcement Learning

🎯 What it does: A hierarchical reinforcement learning framework is employed to address large-scale multi-agent path planning, leveraging spatial and temporal abstraction to enhance sampling efficiency.