arXivSub Start free trial

ICRA 2023 Papers — Page 6

IEEE International Conference on Robotics and Automation · 1341 papers

Failure Detection and Fault Tolerant Control of a Jet-Powered Flying Humanoid Robot

Gabriele Nava, D. Pucci

Robotic Intelligence

🎯 What it does: Proposes a fault detection and fault-tolerant control framework for the flying humanoid robot iRonCub in the case of losing one turbine.

Failure Detection for Motion Prediction of Autonomous Driving: An Uncertainty Perspective

Wenbo Shao, Hong Wang

Anomaly DetectionAutonomous Driving

🎯 What it does: Proposes a motion prediction failure detection framework from an uncertainty perspective, considering motion uncertainty and model uncertainty, and designing multiple uncertainty scores based on different prediction stages.

Failure-aware Policy Learning for Self-assessable Robotics Tasks

Kechun Xu, R. Xiong

Robotic IntelligenceRecurrent Neural NetworkReinforcement Learning

🎯 What it does: Propose a failure-aware strategy learning method for self-evaluation robot tasks, utilizing RNN to record previous self-evaluation results of failures and perform action reselection.

Fast and Scalable Signal Inference for Active Robotic Source Seeking

Chris Denniston, Ali-akbar Agha-mohammadi

Robotic Intelligence

🎯 What it does: Propose a model and scalable planner based on global and local factor graphs for large-scale active source tracking, verified in experiments to outperform baseline methods.

Fast Event-based Double Integral for Real-time Robotics

Shijie Lin, Jia Pan

Computational EfficiencyRobotic Intelligence

🎯 What it does: Proposed a fast event double integration (fast EDI) method for real-time online computation on a single-core CPU

Fast Extrinsic Calibration for Multiple Inertial Measurement Units in Visual-Inertial System

Youwei Yu, Jiamao Li

Optimization

🎯 What it does: Propose a fast extrinsic calibration method that fuses multiple IMUs to enhance the positioning accuracy of visual-inertial odometry (VIO).

Fast Region of Interest Proposals on Maritime UAVs

Benjamin Kiefer, A. Zell

Object DetectionComputational EfficiencyVideo

🎯 What it does: Proposes an end-to-end future frame prediction model for real-time generation of region of interest (ROI) proposals in maritime video streams on embedded GPUs.

Fast Staircase Detection and Estimation using 3D Point Clouds with Multi-detection Merging for Heterogeneous Robots

Prasanna Sriganesh, M. Travers

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes a method for rapid staircase detection and estimation using 3D point clouds, and achieves autonomous localization and stair climbing for heterogeneous robots in multi-level environments through multi-detection merging.

Fast Untethered Soft Robotic Crawler with Elastic Instability

Zechen Xiong, Hod Lipson

Robotic IntelligencePhysics Related

🎯 What it does: Designed and manufactured a single-driven, tangle-free soft robot capable of running at a speed of 313 mm/s (1.56 BL/s).

Fast-Grasp'D: Dexterous Multi-finger Grasp Generation Through Differentiable Simulation

Dylan Turpin, Animesh Garg

Data SynthesisOptimizationRobotic IntelligenceMultimodality

🎯 What it does: Proposed Fast-Grasp'D, a differentiable grasp simulator, and the Grasp'D-1M dataset for multi-finger grasp generation.

Fast, Reliable Constrained Manipulation Using a VSA Driven Planar Robot

Andrew L. Bernhard, J. Schimmels

Robotic Intelligence

🎯 What it does: Designed and evaluated a planar 3R robot equipped with a novel variable stiffness actuator (VSA), achieving restricted flexible manipulation in rigid environments, with performance validated through a steel crank lifting task.

FDLNet: Boosting Real-time Semantic Segmentation by Image-size Convolution via Frequency Domain Learning

Qingqing Yan, Qi Chen

SegmentationConvolutional Neural NetworkImage

🎯 What it does: Proposes a real-time semantic segmentation network based on frequency domain learning called FDLNet, and designs Image Size Convolution (IS-Conv), Global Structural Representation Path (GSRP), and Decomposed Stereoscopic Attention (FSA) modules.

Feature Extraction for Effective and Efficient Deep Reinforcement Learning on Real Robotic Platforms

P. Böhm, Archie C. Chapman

Robotic IntelligenceRecurrent Neural NetworkReinforcement Learning

🎯 What it does: Propose using gated feature extraction (untrained GRU) to improve the training and execution performance of deep reinforcement learning on real robot platforms.

Feature-Realistic Neural Fusion for Real-Time, Open Set Scene Understanding

Kirill Mazur, A. Davison

SegmentationConvolutional Neural NetworkNeural Radiance FieldSimultaneous Localization and Mapping

🎯 What it does: Efficient 3D geometric neural field representations that integrate the general features of pre-trained networks into real-time SLAM for open-set scene understanding.

Few-shot 3D LiDAR Semantic Segmentation for Autonomous Driving

Jilin Mei, Yu Hu

SegmentationAutonomous DrivingKnowledge DistillationPoint Cloud

🎯 What it does: Propose a few-shot 3D LiDAR semantic segmentation method that simultaneously predicts new and base classes.

Few-Shot Point Cloud Semantic Segmentation via Contrastive Self-Supervision and Multi-Resolution Attention

Jiahui Wang, T. Lee

SegmentationContrastive LearningPoint Cloud

🎯 What it does: Propose a contrastive self-supervised pre-training framework for few-shot point cloud semantic segmentation, combined with a multi-resolution attention module.

FewSOL: A Dataset for Few-Shot Object Learning in Robotic Environments

P. JishnuJaykumar, Yu Xiang

ClassificationSegmentationRobotic IntelligenceMeta LearningImageBenchmark

🎯 What it does: Constructed a Few-Shot Object Learning (FEWSOL) dataset and investigated few-shot object classification as well as joint object segmentation and classification on this dataset.

FG-Depth: Flow-Guided Unsupervised Monocular Depth Estimation

Junyu Zhu, Hongbo Zhang

Depth EstimationConvolutional Neural NetworkOptical FlowImage

🎯 What it does: Propose an unsupervised monocular depth estimation framework FG-Depth based on Flow-Net, aiming to utilize flow information to guide optimization and improve performance.

Finding Optimal Modular Robots for Aerial Tasks

Jiawei Xu, D. Saldaña

OptimizationRobotic Intelligence

🎯 What it does: A modular multirotor drone capable of altering its operational capabilities through module reconfiguration is proposed, along with the establishment of its kinematic model and task feasibility determination method, leading to the search for the optimal configuration scheme for a given task.

Finding the Optimal Incision Point in Robotic Assisted Surgery

Kyriakos Almpanidis, Z. Doulgeri

Robotic Intelligence

🎯 What it does: Proposed a simulation tool for finding the optimal incision points in robotic-assisted surgery, integrating preoperative path planning and sensitive area protection.

Finding Things in the Unknown: Semantic Object-Centric Exploration with an MAV

Sotiris Papatheodorou, Stefan Leutenegger

Object DetectionRobotic IntelligenceSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Expanding UAV (MAV) exploration tasks in unknown spaces through semantic and object-level mapping techniques, which not only search for specified target objects but also reconstruct them while meeting precision requirements; simultaneously adding maximum observation distance constraints for background and target objects to the traditional objective function of 'covering as much free space as possible as quickly as possible'; and developing a MAV semantic exploration simulator based on Habitat, validated using a drone equipped with an Intel RealSense D455 RGB-D camera in real-world scenarios.

FingerSLAM: Closed-loop Unknown Object Localization and Reconstruction from Visuo-tactile Feedback

Jialiang Zhao, E. Adelson

Pose EstimationSimultaneous Localization and MappingMultimodalityPoint Cloud

🎯 What it does: Proposes FingerSLAM, a closed-loop factor graph optimization-based pose estimator that integrates fingertip tactile sensing with global vision from a wrist-mounted camera for 6-DoF localization and 3D reconstruction of unknown handheld objects.

Fisher Information Based Active Planning for Aerial Photogrammetry

Jaeyoung Lim, R. Siegwart

OptimizationRobotic IntelligenceImage

🎯 What it does: Proposed a viewpoint utility function based on Fisher information, achieving active path planning for UAVs in complex terrains for photogrammetric reconstruction, and collecting geometrically rich image data through online decision-making.

Flipbot: Learning Continuous Paper Flipping via Coarse-to-Fine Exteroceptive-Proprioceptive Exploration

Chao Zhao, Qifeng Chen

Robotic IntelligenceMultimodality

🎯 What it does: Flipbot learns and performs paper flipping tasks by combining exteroceptive and proprioceptive sensing, and using a coarse-to-fine exploration process.

FloorplanNet: Learning Topometric Floorplan Matching for Robot Localization

Delin Feng, Liangjun Zhang

Data SynthesisRobotic IntelligenceGraph Neural NetworkSimultaneous Localization and MappingPoint CloudGraph

🎯 What it does: Proposes FloorplanNet, which matches robot-measured metric maps with building floorplans using semantic information, and applies this matching to robot localization; utilizes graph neural networks to learn node descriptors from vertex-metric graphs, enabling the matching of 3D point cloud submaps with 2D floorplans;

Flow-Based Rendezvous and Docking for Marine Modular Robots in Gyre-Like Environments

G. Knizhnik, M. A. Hsieh

Robotic IntelligenceFlow-based Model

🎯 What it does: Propose a flow-field-based controller for water-land modular robots to achieve rendezvous and docking by utilizing environmental currents in vortex-like marine environments.

FlowDrone: Wind Estimation and Gust Rejection on UAVs Using Fast-Response Hot-Wire Flow Sensors

Nathaniel Simon, Anirudha Majumdar

Autonomous DrivingRobotic IntelligenceReinforcement LearningTime Series

🎯 What it does: Integration of MEMS hot-wire flow sensors for wind speed estimation in drones, and training of a wind-aware residual controller using reinforcement learning to enhance hover performance

FlowMap: Path Generation for Automated Vehicles in Open Space Using Traffic Flow

Wenchao Ding, Zhongxue Gan

Autonomous DrivingFlow-based ModelGraph

🎯 What it does: Propose the FlowMap framework, which integrates a traffic flow layer into the lightweight semantic map RoadMap, generating human-like paths for autonomous vehicles in complex intersections without HD maps by leveraging traffic flow fields.

FLYOVER: A Model-Driven Method to Generate Diverse Highway Interchanges for Autonomous Vehicle Testing

Yuan Zhou, Yang Liu

Data SynthesisAutonomous DrivingOptimizationGraph

🎯 What it does: Propose a model-driven method called Flyover for generating a dataset of highway interchanges with measurable diversity.

Focused Adaptation of Dynamics Models for Deformable Object Manipulation

P. Mitrano, D. Berenson

Domain AdaptationRobotic Intelligence

🎯 What it does: Propose a new method that adapts the model only in regions where the dynamics of the source and target environments are similar, combined with unreliable dynamics planning to achieve data-efficient online adaptation (FOCUS)

FOGL: Federated Object Grasping Learning

Seok-Kyu Kang, Changhyun Choi

Federated LearningRobotic IntelligenceImage

🎯 What it does: Propose a federated learning method called FOGL to address the impact of non-IID data distributions collected by multiple robots in different environments on the training of robotic arm grasping models.

FogROS2: An Adaptive Platform for Cloud and Fog Robotics Using ROS 2

Jeffrey Ichnowski (BerkeleyAutomation), K. Goldberg (BerkeleyAutomation)

Robotic IntelligenceSimultaneous Localization and MappingVideo

🎯 What it does: Provides the FogROS2 platform, supporting cloud and fog robots, and integrated into ROS 2;

Follow The Rules: Online Signal Temporal Logic Tree Search for Guided Imitation Learning in Stochastic Domains

J. J. Aloor, S. Scherer

OptimizationReinforcement Learning

🎯 What it does: A heuristic method combining Signal Temporal Logic (STL) rules with Monte Carlo Tree Search (MCTS) guides learners to achieve better constraint satisfaction in stochastic domains, thereby improving the performance of example-based learning (LfD) strategies.

Foot gestures to control the grasping of a surgical robot

Yijun Cheng, E. Burdet

Robotic IntelligenceBiomedical Data

🎯 What it does: Investigated the use of foot pressure gestures for controlling surgical robot grasping, and developed three modular foot-machine interfaces. Combined with other motion control interfaces, these achieved grasping control of laparoscopic tools. Evaluations showed that non-expert participants could smoothly perform pick-and-place operations.

Foot Stepping Algorithm of Humanoids with Double Support Time Adjustment based on Capture Point Control

Myeong-Ju Kim, Jaeheung Park

Robotic Intelligence

🎯 What it does: Proposed a footstep algorithm under a model predictive control (MPC) framework based on capture point control, which can dynamically adjust the double support phase (DSP) time

Force control for Robust Quadruped Locomotion: A Linear Policy Approach

Aditya Shirwatkar, Shishir N Y Kolathaya

OptimizationRobotic Intelligence

🎯 What it does: Proposed a simple linear strategy for direct force control in quadruped robots, generating foot trajectory parameters and center of mass (CoM) torques, followed by allocating ground reaction forces using a 12-variable QP based on foot contact information.

Force/Torque Sensing for Soft Grippers using an External Camera

Jeremy Collins, Charles C. Kemp

Robotic IntelligenceConvolutional Neural NetworkImageMultimodality

🎯 What it does: Developed a method called Visual Force/Torque Estimation for Soft grippers (VFTS) using an external camera, capable of outputting 6-axis force/torque (F/T) measurements.

Forming and Controlling Hitches in Midair Using Aerial Robots

Diego S. D’antonio, D. Saldaña

Robotic Intelligence

🎯 What it does: Propose using a team of aerial drones to form and deform knots in the air to secure objects, and provide systematic algorithms and action sets.

FourStr: When Multi-sensor Fusion Meets Semi-supervised Learning

Bangquan Xie, Bing Li

Autonomous DrivingMultimodalityPoint Cloud

🎯 What it does: Proposed a semi-supervised learning framework called FourStr, which enhances the fusion and annotation efficiency of 3D multi-sensor detectors through a four-stream model (composed of two two-stream models);

FRAME: Fast and Robust Autonomous 3D Point Cloud Map-Merging for Egocentric Multi-Robot Exploration

Nikolaos Stathoulopoulos, G. Nikolakopoulos

Pose EstimationComputational EfficiencyRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Designed and implemented a 3D point cloud map fusion framework based on overlap detection and alignment for autonomous heterogeneous multi-robot panoramic exploration without requiring manual prior poses.

FreDSNet: Joint Monocular Depth and Semantic Segmentation with Fast Fourier Convolutions from Single Panoramas

Bruno Berenguel-Baeta, J. J. Guerrero

SegmentationDepth EstimationConvolutional Neural NetworkImage

🎯 What it does: Propose FreDSNet, a deep learning solution for achieving semantic 3D understanding of indoor environments from a single panoramic image, jointly realizing monocular depth estimation and semantic segmentation.

FRIDA: A Collaborative Robot Painter with a Differentiable, Real2Sim2Real Planning Environment

Peter Schaldenbrand, Jean Oh

Robotic IntelligenceVision-Language-Action ModelWorld ModelImageText

🎯 What it does: Developed a framework and robotic planning system that enables humans to collaborate with robots to complete canvas painting through language descriptions or images.

From Concept to Field Tests: Accelerated Development of Multi-AUV Missions Using a High-Fidelity Faster-than-Real-Time Simulator

T. Player, B. Hobson

OptimizationComputational EfficiencyRobotic IntelligenceSimultaneous Localization and MappingWorld ModelMultimodality

🎯 What it does: Designed and verified a novel simulator for efficient development of multi-robot marine tasks

From Semi-supervised to Omni-supervised Room Layout Estimation Using Point Clouds

Huanhuan Gao, H. Zha

SegmentationPoint CloudBenchmark

🎯 What it does: This paper proposes a new framework for the point cloud room layout estimation task, transitioning from semi-supervised to fully supervised. The core components include a quad set matching strategy, a dedicated consistency loss based on layout quadrilaterals, and an online pseudo-label collection algorithm that does not require thresholds;

Frontier Semantic Exploration for Visual Target Navigation

Bangguo Yu, M. Cao

Autonomous DrivingReinforcement LearningSimultaneous Localization and MappingImage

🎯 What it does: Proposed a visual goal navigation framework based on frontier semantic strategies, which constructs maps from current observations using semantic maps and frontier maps, and learns to select frontier cells as long-term goals through deep reinforcement learning to achieve efficient exploration.

Fruit Tracking Over Time Using High-Precision Point Clouds

Alessandro Riccardi, C. Stachniss

Object TrackingPoint CloudAgriculture Related

🎯 What it does: Monitor fruit growth using high-precision point cloud data in commercial greenhouses, and study the matching problem between fruits at different growth stages.

FSG-Net: a Deep Learning model for Semantic Robot Grasping through Few-Shot Learning

L. Barcellona, S. Ghidoni

SegmentationRobotic IntelligenceMeta LearningConvolutional Neural NetworkImagePoint Cloud

🎯 What it does: Proposed a deep learning architecture for few-shot semantic grasping tasks, which infers correct grasp poses on unseen target objects using only five annotated images through a few-shot semantic segmentation module.

Fully Robotized 3D Ultrasound Image Acquisition for Artery

Mingcong Chen, Hongbin Liu

Robotic IntelligenceConvolutional Neural NetworkBiomedical DataUltrasound

🎯 What it does: A method for fully automated 3D arterial imaging using a linear ultrasound probe, a 6-DoF robotic arm, and a 3D camera was developed.

Fusing Event-based Camera and Radar for SLAM Using Spiking Neural Networks with Continual STDP Learning

A. Safa, Georges G. E. Gielen

Autonomous DrivingSpiking Neural NetworkSimultaneous Localization and MappingMultimodality

🎯 What it does: Designed and implemented a SLAM architecture that integrates event cameras and FMCW radar, using spiking neural networks (SNNs) with continuous STDP learning for feature extraction and map construction.

Fusion of Events and Frames using 8-DOF Warping Model for Robust Feature Tracking

Min-Seok Lee, Chan Gook Park

Object TrackingOptical FlowMultimodality

🎯 What it does: Propose a robust feature tracking method using an 8-DOF deformation model, which fuses complementary information from event cameras and traditional frame cameras by minimizing the brightness increment patch differences between events and frames.

Gait Event Detection with Proprioceptive Force Sensing in a Powered Knee-Ankle Prosthesis: Validation over Walking Speeds and Slopes

Emily G. Keller, R. Gregg

Robotic IntelligenceTime Series

🎯 What it does: Propose a walking event detection method based on proprioceptive sensing, utilizing the algebraic relationship between joint torque and ground contact force to derive a floating body dynamics model and estimate ground reaction force, thereby achieving gait event detection without load sensors, and verifying its feasibility through treadmill experiments at different speeds and slopes.

Gaka-Chu: A Self-Employed Autonomous Robot Artist

Eduardo Castell'o Ferrer, P. Tarasov

Robotic IntelligenceAgentic AIFinance Related

🎯 What it does: Developed a robot named Gaka-chu capable of autonomously creating and selling Japanese character paintings, using sales revenue to automatically replenish materials and repay investors, achieving economic autonomy.

GAN-Based Interactive Reinforcement Learning from Demonstration and Human Evaluative Feedback

Jie Huang, Guangliang Li

Reinforcement Learning from Human FeedbackReinforcement LearningGenerative Adversarial Network

🎯 What it does: Proposes GAIRL, a GAN-based interactive reinforcement learning method that learns from demonstrations and human evaluation feedback;

GANet: Goal Area Network for Motion Forecasting

Mingkun Wang, Wenjing Yang

Autonomous DrivingVideoPoint CloudBenchmark

🎯 What it does: Proposed a trajectory prediction framework called GANet based on the target area.

GaPT: Gaussian Process Toolkit for Online Regression with Application to Learning Quadrotor Dynamics

Francesco Crocetti, Giuseppe Loianno

Computational EfficiencyRobotic Intelligence

🎯 What it does: Proposed the GaPT toolkit, which converts Gaussian processes into state space form, performs regression in linear time, and validates quadrotor dynamics in single-input and multi-input settings.

GDIP: Gated Differentiable Image Processing for Object Detection in Adverse Conditions

Sanket Kalwar, K. Krishna

Object DetectionImage

🎯 What it does: The study proposes a gated differentiable image processing (GDIP) module, which can be plugged into existing object detection networks, and end-to-end learns image enhancement through detection loss under adverse conditions such as foggy weather and low illumination.

GenDexGrasp: Generalizable Dexterous Grasping

Puhao Li, Siyuan Huang

Data SynthesisOptimizationRobotic IntelligenceGraph Neural NetworkGraph

🎯 What it does: Proposed a hand-agnostic and generalizable grasping algorithm called GenDexGrasp, which can quickly generate diverse and high-success-rate grasping poses and transfer across different multi-fingered robotic hands.

General, Single-shot, Target-less, and Automatic LiDAR-Camera Extrinsic Calibration Toolbox

Kenji Koide, A. Banno

Autonomous DrivingTransformerImagePoint Cloud

🎯 What it does: Designed and implemented a generic, single-shot, target-free, and fully automatic LiDAR-camera extrinsic calibration toolbox.

Generalizable Movement Intention Recognition with Multiple Heterogeneous EEG Datasets

Xiao Gu, Benny P. L. Lo

RecognitionDomain AdaptationKnowledge DistillationBiomedical Data

🎯 What it does: Developed two networks to learn information from shared channels and complete channels, and constructed an online knowledge co-distillation framework to enhance cross-subject generalization performance.

Generalizable Pose Estimation Using Implicit Scene Representations

Vaibhav Saxena, Yotto Koga

Pose EstimationNeural Radiance Field

🎯 What it does: Proposes a 6-DoF pose estimation method based on implicit scene representation, achieving pose inference from input images through inverse neural rendering.

Generalization of Impact Response Factors for Proprioceptive Collaborative Robots

Carlos Relaño, C. Monje

Robotic Intelligence

🎯 What it does: This paper proposes a generic impact absorption factor (GIAF) applicable to both floating and fixed-base robots, along with its mathematical definition and potential application examples.

Generating a Terrain-Robustness Benchmark for Legged Locomotion: A Prototype via Terrain Authoring and Active Learning

Chong Zhang

Data SynthesisRobotic IntelligenceBenchmark

🎯 What it does: A terrain dataset is prototyped using terrain creation and active learning, with the learned sampler capable of stably generating diverse and high-quality terrains.

Generating Formal Safety Assurances for High-Dimensional Reachability

Albert Lin, Somil Bansal

OptimizationSafty and PrivacyRobotic IntelligencePhysics Related

🎯 What it does: Propose a method to compute an upper bound on the error of DeepReach solutions, and use this error upper bound to correct the reachable tube to obtain a safe approximation of the true reachable tube; also propose a scenario-based optimization method to calculate the probabilistic upper bound of error correction; verify the effectiveness of the method in high-dimensional rocket landing and multi-vehicle collision avoidance problems.

Generating Stable and Collision-Free Policies through Lyapunov Function Learning

Alexandre Coulombe, Hsiu-Chin Lin

OptimizationReinforcement Learning

🎯 What it does: Proposes a method that uses a single neural network to simultaneously learn Lyapunov functions and collision safety strategies, verifying its feasibility in several simulation environments and real-world scenarios.

Getting Air: Modelling and Control of a Hybrid Pneumatic-Electric Legged Robot

Christopher Mailer, Amir Patel

OptimizationRobotic Intelligence

🎯 What it does: Developed an operational model and representation procedure for standard double-acting cylinders controlled by on/off solenoid valves, and applied this model to trajectory optimization, further enabling the quadruped robot Kemba to achieve jumping and landing motions by integrating electric and pneumatic actuators.

GIDP: Learning a Good Initialization and Inducing Descriptor Post-enhancing for Large-scale Place Recognition

Zhaoxin Fan, Jun He

RetrievalContrastive LearningPoint Cloud

🎯 What it does: Proposed the GIDP method, combining unsupervised momentum contrast point cloud pre-training with a descriptor post-enhancement module based on re-ranking to improve large-scale place recognition performance.

Global and Reactive Motion Generation with Geometric Fabric Command Sequences

Weiming Zhi, Fabio Ramos

OptimizationRobotic Intelligence

🎯 What it does: A novel robot motion generation method combining global optimization with an adaptive Geometric Fabric strategy is studied, achieving globally optimal, smooth, and intuitive motion through attractor state sequences;

Global Localization in Repetitive and Ambiguous Environments

Zhenyu Wu, Danwei W. Wang

Robotic IntelligenceSimultaneous Localization and MappingPoint CloudSequential

🎯 What it does: Proposes a probabilistic global localization system for AMRs in repetitive ambiguous environments using 2D LiDAR and rotation-invariant magnetic fields, achieving localization through two-step initialization and pose tracking based on magnetic field sequences.

Globally Defined Dynamic Modelling and Geometric Tracking Controller Design for Aerial Manipulator

Byeongjun Kim, H. J. Kim

Robotic IntelligencePhysics Related

🎯 What it does: This paper proposes a globally defined dynamic model for conventional multi-rotor vehicles equipped with a single n-DOF robotic arm, and designs a geometric tracking controller based on this model.

Globally Guided Trajectory Planning in Dynamic Environments

O. D. Groot, Javier Alonso-Mora

OptimizationRobotic Intelligence

🎯 What it does: Studied the problem of mobile robot navigation in shared human dynamic environments, proposing a method to identify multiple local optimal passing behaviors through topological information, maintaining consistency during continuous iterations, and ultimately guiding local optimization planners with the most suitable high-level trajectory to achieve fast and safe motion planning.

GMCR: Graph-based Maximum Consensus Estimation for Point Cloud Registration

Michael Gentner, Mohsen Kaboli

Pose EstimationGraph Neural NetworkPoint CloudBenchmark

🎯 What it does: Proposed a graph-based maximum consistency estimation (GMCR) method for point cloud registration.

GMM Registration: a Probabilistic scan matching approach for sonar-based AUV navigation

Pau Vial, M. Carreras

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Developed a sonar scan matching method based on the Gaussian Mixture Model (GMM) that achieves sparse, noisy point cloud registration in AUV navigation and returns uncertainty estimates of the matching results.

GNM: A General Navigation Model to Drive Any Robot

Dhruv Shah, S. Levine

Robotic IntelligenceTime SeriesSequential

🎯 What it does: Trained a general-purpose visual navigation model that can share data across multiple robots and generalize across robots, trained on multi-robot data

GNN-Based Point Cloud Maps Feature Extraction and Residual Feature Fusion for 3D Object Detection

Wei-Hsiang Liao, Wen-Chieh Lin

Object DetectionAutonomous DrivingGraph Neural NetworkPoint Cloud

🎯 What it does: Construct a point cloud map feature extraction module based on graph neural networks (GNN), and perform residual feature fusion with PVRCNN's point-level and voxel-level features for 3D object detection.

Goal-Conditioned Action Space Reduction for Deformable Object Manipulation

Shengyin Wang, M. Dogar

Computational EfficiencyRobotic IntelligencePoint CloudMesh

🎯 What it does: Propose to reduce the number of pick points in deformable object manipulation by identifying key particles, thereby lowering planning computational costs.

Goal-Image Conditioned Dynamic Cable Manipulation through Bayesian Inference and Multi-Objective Black-Box Optimization

K. Takahashi, T. Taniguchi

OptimizationRobotic IntelligenceImage

🎯 What it does: Achieving dynamic cable manipulation under target image conditions through Bayesian inference and multi-objective black-box optimization

GoRela: Go Relative for Viewpoint-Invariant Motion Forecasting

Alexander Cui, R. Urtasun

Autonomous DrivingGraph Neural NetworkImageBenchmark

🎯 What it does: Proposed a viewpoint-invariant motion prediction method called GoRela, which uses diagonal relative position encoding to perform shared encoding for all pedestrians and maps, reducing online computation while maintaining accuracy and generalization capabilities.

GP-Frontier for Local Mapless Navigation

Mahmoud Ali, Lantao Liu

Autonomous DrivingRobotic Intelligence

🎯 What it does: Propose a frontier concept based on Gaussian processes called GP-Frontier, which can locally guide robots toward a target without requiring map construction.

GPF-BG: A Hierarchical Vision-Based Planning Framework for Safe Quadrupedal Navigation

Shiyu Feng, P. Vela

Safty and PrivacyRobotic IntelligenceImage

🎯 What it does: Proposed a hierarchical vision-based planning framework called GPF-BG, integrating the previous global path follower GPF with a Bézier curve-based gap planner (BG) for safe navigation of quadruped robots in unknown environments.

Gradient-Based Trajectory Optimization With Learned Dynamics

Bhavya Sukhija, Stelian Coros

OptimizationRobotic IntelligenceWorld Model

🎯 What it does: Using machine learning to learn differentiable dynamics models and utilizing their gradients for gradient-based trajectory optimization to achieve high dynamic complex tasks without an accurate analytical dynamics model.

Graph Neural Networks for Multi-Robot Active Information Acquisition

Mariliza Tzes, George Pappas

Robotic IntelligenceGraph Neural NetworkGraph

🎯 What it does: Proposed an information-aware graph block network (I-GBNet) for multi-robot active information acquisition, capable of aggregating information in distributed communication graphs and performing sequential decision-making.

Graph-based Pose Estimation of Texture-less Surgical Tools for Autonomous Robot Control

Haozheng Xu, S. Giannarou

Pose EstimationRobotic IntelligenceGraph Neural NetworkImage

🎯 What it does: Proposed a keypoint map-based network for estimating the pose of cylindrical surgical tools with missing textures and small diameters, achieving stable and accurate poses through a PnP solver.

Grasp Planning with CNN for Log-loading Forestry Machine

Elie Ayoub, I. Sharf

Depth EstimationRobotic IntelligenceConvolutional Neural NetworkImageAgriculture Related

🎯 What it does: Propose a grasp planning pipeline based on CNN and virtual depth camera for log loading in forest machines.

GraspAda: Deep Grasp Adaptation through Domain Transfer

Yiting Chen, Miao Li

Domain AdaptationRobotic IntelligenceGenerative Adversarial NetworkContrastive LearningImage

🎯 What it does: Propose a new grasp adaptation strategy based on visual data, utilizing a novel grasp feature representation to transfer the learned grasp capability to new domains; employ a conditional generative model for visual data transformation, and bridge the gap between the training domain and the new domain by maintaining consistency during adaptation through deep feature representation, feature-level contrastive learning, and adversarial learning in the output space; the pre-trained model can generalize to the new domain without fine-tuning based on the transformed input.

GraspNeRF: Multiview-based 6-DoF Grasp Detection for Transparent and Specular Objects Using Generalizable NeRF

Qiyu Dai, He Wang

Object DetectionData SynthesisPose EstimationNeural Radiance FieldImage

🎯 What it does: Proposed GraspNeRF, a 6-DoF grasping detection network based on multi-view RGB, which achieves material-agnostic grasping of transparent and specular objects using a generalizable NeRF.

Grey-Box Learning of Adaptive Manipulation Primitives for Robotic Assembly

Marco Braun, S. Wrede

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose Adaptive Manipulation Primitives (AMP), a gray-box learning method that combines compliance operation task specifications with policy gradient reinforcement learning for robot assembly task learning

GRM: Gradient Rectification Module for Visual Place Retrieval

Boshu Lei, Xi Qiu

RetrievalConvolutional Neural NetworkImage

🎯 What it does: Proposed a gradient correction module (GRM) for learning global descriptors in visual place retrieval tasks, aiming to address the problem of low-dimensional principal space caused by gradient distribution degradation.

Ground then Navigate: Language-guided Navigation in Dynamic Scenes

Kanishk Jain, Vineet Gandhi

SegmentationAutonomous DrivingExplainability and InterpretabilityVision-Language-Action ModelMultimodalityBenchmark

🎯 What it does: In outdoor autonomous driving environments, explicitly benchmarking navigable areas using language instructions, with the model predicting corresponding segmentation masks at each moment to complete the visual language navigation task.

Grounding Language with Visual Affordances over Unstructured Data

Oier Mees (University of Freiburg), Wolfram Burgard (University of Freiburg)

Robotic IntelligenceLarge Language ModelPrompt EngineeringVision-Language-Action ModelMultimodalityBenchmark

🎯 What it does: Propose a method that leverages a self-supervised vision-language empowerment model to efficiently learn general language-conditioned robot skills from unstructured, offline, and non-reset real-world data.

GSMR-CNN: An End-to-End Trainable Architecture for Grasping Target Objects from Multi-Object Scenes

Valerija Holomjova, P. Meissner

Object DetectionSegmentationRobotic IntelligenceSpiking Neural NetworkImage

🎯 What it does: Propose an end-to-end trainable multi-task model called GSMR-CNN for locating and grasping target objects in multi-object scenes.

GSNet: Model Reconstruction Network for Category-level 6D Object Pose and Size Estimation

Penglei Liu, Jun Cheng

Pose EstimationGraph Neural NetworkAuto EncoderImage

🎯 What it does: Propose a category-level 6D pose and size estimation method based on RGB-D images, which uses an autoencoder-decoder to learn shape variation features and constructs a GSNet network to reconstruct instance 3D models, thereby achieving rotation, translation, and size estimation for unseen objects.

Guided Conditional Diffusion for Controllable Traffic Simulation

Ziyuan Zhong, M. Pavone

Data SynthesisAutonomous DrivingDiffusion modelVideoPoint Cloud

🎯 What it does: Developed a conditional diffusion model for controllable traffic generation;

Guided Learning from Demonstration for Robust Transferability

F. Sukkar, J. Deuse

Domain AdaptationRobotic IntelligenceReinforcement Learning from Human Feedback

🎯 What it does: Developed and verified an interactive GUI-guided learning demonstration method to ensure reproducibility of demonstration actions in the target system.

Guiding Reinforcement Learning with Shared Control Templates

A. Padalkar, F. Stulp

Reinforcement Learning

🎯 What it does: Use Shared Control Templates (SCT) to guide reinforcement learning in completing the pouring task, achieving safe and rapid learning.

GuILD: Guided Incremental Local Densification for Accelerated Sampling-based Motion Planning

Aditya Mandalika, S. Srinivasa

Optimization

🎯 What it does: Propose Guided Incremental Local Densification (GuILD), which improves sampling strategies in sample-based motion planning by leveraging local information.

GUTS: Generalized Uncertainty-Aware Thompson Sampling for Multi-Agent Active Search

Nikhil Angad Bakshi, J. Schneider

OptimizationReinforcement Learning

🎯 What it does: Proposed and verified the GUTS (Generalized Uncertainty-aware Thompson Sampling) algorithm for asynchronous multi-agent active search tasks, achieving rapid target localization in simulations and field tests;

H-SAUR: Hypothesize, Simulate, Act, Update, and Repeat for Understanding Object Articulations from Interactions

Keita Ota, J. Tenenbaum

Robotic IntelligenceBenchmark

🎯 What it does: Propose a probability generation framework based on the 'Hypothesis-Simulation-Action-Update-Repeat' (H-SAUR) loop, which generates joint hypothesis distributions from observations, tracks confidence over time, and infers feasible exploration and goal-based manipulation actions.

HALO: Hazard-Aware Landing Optimization for Autonomous Systems

Christopher R. Hayner, Behçet Açikmese

OptimizationRobotic IntelligencePoint Cloud

🎯 What it does: Developed a system integrating perception and planning, utilizing HALSS for safe landing zone identification under point cloud information, and achieving multi-objective, adaptive landing trajectory optimization and emergency planning through Adaptive-DDTO.

Handling Sparse Rewards in Reinforcement Learning Using Model Predictive Control

Murad Dawood, Maren Bennewitz

Robotic IntelligenceReinforcement Learning

🎯 What it does: Using MPC as an experience source to train RL agents, solving reward design challenges in sparse reward environments, applied to mobile robot navigation (simulation and real-world Kuboki Turtlebot 2)

HaPPArray: Haptic Pneumatic Pouch Array for Feedback in handheld Robots

Xiaolei Luo, Tania. K. Morimoto

Robotic Intelligence

🎯 What it does: Designed and evaluated a 3x3 pneumatic bladder array called HaPPArray for handheld robot handles to provide lightweight, easily integrable, and distinguishable haptic feedback.