arXivSub Start free trial

ICRA 2024 Papers — Page 6

IEEE International Conference on Robotics and Automation · 1760 papers

DVI-SLAM: A Dual Visual Inertial SLAM Network

Xiongfeng Peng, Qiang Wang

Autonomous DrivingRobotic IntelligenceSimultaneous Localization and MappingMultimodality

🎯 What it does: Propose a dual visual factor depth SLAM network that integrates photometric factors and reprojection factors into an end-to-end differentiable network.

DyHGDAT: Dynamic Hypergraph Dual Attention Network for multi-agent trajectory prediction

Weilong Lin, Jing Liu

Autonomous DrivingGraph Neural NetworkAuto EncoderTime SeriesSequential

🎯 What it does: Propose a dynamic hypergraph dual attention network (DyHGDAT) to capture high-order interactions in multi-agent trajectory prediction

DynaInsRemover: A Real-time Dynamic Instance-Aware Static 3D LiDAR Mapping Framework for Dynamic Environment

Huanfeng Zhao, Bo Zheng

Autonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a real-time dynamic instance-aware static 3D LiDAR mapping framework called DynaInsRemover, which efficiently removes dynamic objects while preserving more static map details by leveraging geometric differences between instances.

Dynamic Adaptive Imaging System on Optoelectronic Tweezers Platform

Ao Wang, Lin Feng

OptimizationRobotic IntelligenceImagePhysics Related

🎯 What it does: Developed a dynamic adaptive imaging system to enhance autofocus and illumination uniformity on an optoelectronic micromanipulation platform.

Dynamic Coalition Formation and Routing for Multirobot Task Allocation via Reinforcement Learning

Weiheng Dai, Guillaume Sartoretti

OptimizationRobotic IntelligenceTransformerReinforcement Learning

🎯 What it does: Propose a method for dynamic coalition formation and path planning in multi-robot systems based on reinforcement learning, using attention networks to enable robots to learn collaborative scheduling during task allocation, and incorporating a leader-follower mechanism to enhance cooperative learning efficiency.

Dynamic Crane Scheduling with Reinforcement Learning for a Steel Coil Warehouse

Sang-Hyun Cho, Hyun-Jung Kim

OptimizationGraph Neural NetworkReinforcement Learning

🎯 What it does: Proposed a real-time reinforcement learning algorithm for dynamic crane scheduling in a steel coil warehouse, aiming to minimize the average task waiting time through task assignment and execution sequence optimization.

Dynamic evaluation of a suction based gripper for fruit picking using a physical twin

Alejandro Velasquez, Joseph R. Davidson

Robotic IntelligenceWorld ModelAgriculture Related

🎯 What it does: Designed and evaluated a three-suction-cup vacuum gripper for fruit picking.

Dynamic Interaction Control in Legged Mobile Manipulators: A Decoupled Approach

Qikai Li, Kun Xu

Robotic Intelligence

🎯 What it does: Proposes a separation control method for legged mobile robots and robotic arms, explicitly estimating the torque exerted by the robotic arm on the base and incorporating it into the legged robot dynamics, using nonlinear model predictive control (NMPC) to control the legged robot; the robotic arm side employs impedance control to achieve force control at the end effector.

Dynamic modeling of wing-assisted inclined running with a morphing multi-modal robot

Eric Sihite, Morteza Gharib

Robotic Intelligence

🎯 What it does: Designed a deformable multifunctional robot, derived its dynamic model, proposed a nonlinear model predictive controller (MPC) to achieve wing-assisted incline running (WAIR), and verified its feasibility and performance through numerical simulations and experiments.

Dynamic Multi-Agent Deep Deterministic Policy Gradient for Autonomous Navigation of Reconfigurable Unmanned Aerial Vehicle

Xin Lu, Fusheng Li

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a dynamic multi-agent deep deterministic policy gradient (DMADDPG) algorithm for autonomous navigation of reconfigurable UAVs in complex environments;

Dynamic Occupancy Grids for Object Detection: A Radar-Centric Approach

M. Ronecker, Daniel Watzenig

Object DetectionPoint CloudBenchmark

🎯 What it does: Proposed a radar-based dynamic occupancy grid mapping algorithm to generate a local environment map containing both static and dynamic information.

Dynamic Targeting of Satellite Observations Incorporating Slewing Costs and Complex Observation Utility *

Akseli Kangaslahti, Steve A. Chien

OptimizationPhysics Related

🎯 What it does: Implemented a more general dynamic target localization framework, including a physics-based pivot model, dynamic observation utility model, and provided greedy and depth-first search algorithms for efficient observation.

Eclares: Energy-Aware Clarity-Driven Ergodic Search

Kaleb Ben Naveed, Dimitra Panagou

OptimizationRobotic Intelligence

🎯 What it does: Proposes the Eclares framework, which includes constructing the target information space distribution based on clarity and an energy-aware filter to optimize energy-aware ergodic search trajectories.

EdgePoint: Efficient Point Detection and Compact Description via Distillation

Haodi Yao, Fenghua He

CompressionComputational EfficiencyKnowledge DistillationRepresentation LearningImage

🎯 What it does: Propose the EdgePoint lightweight network, integrating UnfoldSoftmax detection loss, Ortho-Alignment loss, and LocalPCA compression to generate a 32-dimensional compact descriptor, and quantize the descriptor as integers to achieve fast interest point detection and compact description.

EdgeSoil 2.0 – Soil Analyzer Using Convolutional Neural Network and Camera Imaging for Agricultural Robotics

R. Kasemi, Markus Vincze

RecognitionComputational EfficiencyConvolutional Neural NetworkImageVideoAgriculture Related

🎯 What it does: Developed a real-time soil pH prediction system called EdgeSoil 2.0, based on convolutional neural networks (CNN) and camera imaging, capable of non-intrusive soil analysis on mobile robots.

EDMP: Ensemble-of-costs-guided Diffusion for Motion Planning

Kallol Saha, Madhava Krishna

Autonomous DrivingDiffusion model

🎯 What it does: Propose a motion planning method called EDMP based on diffusion models, which trains the network using a set of diverse kinematically valid trajectories and guides the diffusion process during inference to generate trajectories that satisfy constraints according to scene-specific costs (e.g., collision costs).

EDOPT: Event-camera 6-DoF Dynamic Object Pose Tracking

Arren J. Glover, C. Bartolozzi

Pose Estimation

🎯 What it does: Proposed a real-time 6-DoF dynamic object pose tracking algorithm called EDOPT based on event cameras, achieving high-frequency tracking by leveraging the low latency and continuous capture capabilities of event cameras;

Effective Representation Learning is More Effective in Reinforcement Learning than You Think

Jiawei Zheng, Yonghong Song

Representation LearningReinforcement LearningContrastive Learning

🎯 What it does: Propose a visual reinforcement learning method utilizing teacher-student collaboration, jointly estimating the Q-function through contrastive learning representations obtained from teacher and student encoders, and using TD error to guide the update of the teacher encoder.

Effectively Detecting Loop Closures using Point Cloud Density Maps

Saurabh Gupta, C. Stachniss

Autonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Propose a method for detecting 3D LiDAR loop closure using a density map view, constructing local maps based on local consistent odometry to enhance robustness and real-time performance.

Efficient 3D Instance Mapping and Localization with Neural Fields

G. Tang, Antonio Torralba

SegmentationPose EstimationSupervised Fine-TuningNeural Radiance Field

🎯 What it does: Propose the 3DIML framework, which learns a label field renderable from new perspectives through a two-stage process (InstanceMap and InstanceLift), achieving efficient 3D instance segmentation and real-time localization.

Efficient and Accurate Mapping of Subsurface Anatomy via Online Trajectory Optimization for Robot Assisted Surgery

Brian Y. Cho, Alan Kuntz

Robotic IntelligenceSimultaneous Localization and MappingBiomedical Data

🎯 What it does: Proposes an automated perception method that utilizes the probabilistic 3D occupancy map from Bayesian Hilbert mapping to map 3D subsurface anatomical structures and plans surface perception paths via A* search.

Efficient and Accurate Transformer-Based 3D Shape Completion and Reconstruction of Fruits for Agricultural Robots

Federico Magistri, C. Stachniss

RestorationTransformerAgriculture Related

🎯 What it does: A Transformer-based neural network combined with template deformation technology was used to achieve 3D shape completion and reconstruction of fruits under occluded conditions.

Efficient Clothoid Tree-Based Local Path Planning for Self-Driving Robots

Minhyeong Lee, Dongjun Lee

Autonomous DrivingOptimizationOrdinary Differential Equation

🎯 What it does: Propose a real-time path planning method for autonomous driving robots based on Clothoid trees

Efficient Composite Learning Robot Control Under Partial Interval Excitation

Tian Shi, Yongping Pan

Robotic Intelligence

🎯 What it does: A time segmentation multi-channel (TDMC) composite learning robot control strategy is proposed, which utilizes filtered regressors over multiple time intervals to generate generalized prediction errors, achieving rapid and accurate parameter estimation, and realizing global exponential stability and parameter convergence in closed-loop systems.

Efficient End-to-End Detection of 6-DoF Grasps for Robotic Bin Picking

Yushi Liu, Andreas Geiger

Pose EstimationRobotic IntelligenceImage

🎯 What it does: Propose a parametric grasp distribution model based on the Power-Spherical distribution for learning dense and diverse 6-DoF grasp poses, generating multiple collision-free grasp directions from a single top-down depth image.

Efficient Gesture Recognition on Spiking Convolutional Networks Through Sensor Fusion of Event-Based and Depth Data

Lea Steffen, R. Dillmann

RecognitionConvolutional Neural NetworkSpiking Neural NetworkMultimodality

🎯 What it does: Proposed a spiking convolutional neural network that integrates event and depth data for gesture recognition.

Efficient Hybrid Neuromorphic-Bayesian Model for Olfaction Sensing: Detection and Classification

Rizwana Kausar, Jorge Dias

ClassificationRobotic IntelligenceSpiking Neural Network

🎯 What it does: Designed and implemented a hybrid model combining convolutional spiking neural networks and Bayesian spiking neural networks for detection and classification in autonomous robot olfactory sensing.

Efficient ISO/TS 15066 Compliance through Model Predictive Control

Andrea Pupa, Cristian Secchi

Optimization

🎯 What it does: Proposes a nonlinear optimal control method for human-robot collaborative industrial scenarios, focusing on ensuring precise robot path execution, utilizing redundancy to shorten task completion time, while meeting ISO/TS 15066 safety standards.

Efficient Model Learning and Adaptive Tracking Control of Magnetic Micro-Robots for Non-Contact Manipulation

Yongyi Jia, Xiang Li

OptimizationRobotic IntelligenceConvolutional Neural NetworkPhysics Related

🎯 What it does: A data-driven non-contact manipulation method for magnetic microrobots is proposed, which includes constructing a neural network to estimate the motion model and designing an approximate model optimal control scheme based on this model to track time-varying trajectories, while introducing a planner to evaluate adaptability in crowded unstructured environments.

Efficient Motion Planning for Manipulators with Control Barrier Function-Induced Neural Controller

Mingxin Yu, Chuchu Fan

OptimizationRobotic IntelligencePoint Cloud

🎯 What it does: Propose a neural controller based on Control Barrier Functions (CBF), combined with RRT sampling-based planning, to reduce the required number of samples and achieve real-time collision avoidance.

Efficient Object Rearrangement via Multi-view Fusion

Dehao Huang, Hong Zhang

Pose EstimationRobotic IntelligenceImage

🎯 What it does: Estimate the target pose before moving objects by observing the current scene through multi-view fusion, thus achieving object rearrangement in the image target scene.

Efficient Polynomial Sum-Of-Squares Programming for Planar Robotic Arms

Daniel Keren, Roi Poranne

OptimizationRobotic Intelligence

🎯 What it does: Proposes a more concise Sum-Of-Squares (SOS) planning formulation for 2D robotic arms to identify large obstacle-free convex regions, ensuring collision safety for linear trajectories and supporting further optimization.

Efficient Pose Prediction with Rational Regression applied to vSLAM

G. Terzakis, Manolis I. A. Lourakis

Pose EstimationSimultaneous Localization and Mapping

🎯 What it does: Proposed a least squares method using rational regression for 6D pose prediction, which can handle fixed data points and avoid division by zero in the rational function's denominator.

Efficient RRT*-based Safety-Constrained Motion Planning for Continuum Robots in Dynamic Environments

Peiyu Luo, M. Meng

OptimizationRobotic Intelligence

🎯 What it does: Studied an RRT*-based motion planning method for continuum robots, integrating posture safety constraints to achieve autonomous navigation and obstacle avoidance in dynamic environments.

Efficient Semantic Segmentation for Compressed Video

Jiaxin Cai, Wenxi Liu

SegmentationVideo

🎯 What it does: Proposed an efficient semantic segmentation paradigm for compressed video, designed the WTDecomNet network structure based on wavelet transform principles, and introduced the efficient axial subband approximation module and lightweight temporal alignment module.

Efficient Terrain Map Using Planar Regions for Footstep Planning on Humanoid Robots

Bhavyansh Mishra, Robert J. Griffin

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Develop a method that utilizes planar region measurements and kinematic inertial state estimation to construct a dense yet efficient finite planar surface map for bipedal robot footprint planning.

Efficient, Dynamic Locomotion through Step Placement with Straight Legs and Rolling Contacts

Stefan Fasano, Robert J. Griffin

OptimizationRobotic Intelligence

🎯 What it does: Proposed a motion controller for efficient walking on nearly flat ground, validated in simulation and on the actual robot Nadia.

EfficientDPS: Efficient and End-to-End Depth-aware Panoptic Segmentation

Shengkai Wu, Wenyu Liu

SegmentationDepth EstimationConvolutional Neural NetworkImagePoint Cloud

🎯 What it does: Proposed an efficient, end-to-end unified depth-aware panoptic segmentation model called EfficientDPS, which uses query features extracted by convolutional networks to represent objects and scenes, supporting classification, segmentation, and depth estimation, and achieving training and inference without post-processing through bilateral matching.

EffLoc: Lightweight Vision Transformer for Efficient 6-DOF Camera Relocalization

Zhendong Xiao, Wu Wei

Pose EstimationComputational EfficiencyTransformerImage

🎯 What it does: Propose EffLoc, a lightweight visual Transformer for single-image camera relocalization.

Effort Level Search in Infinite Completion Trees with Application to Task-and-Motion Planning

Marc Toussaint, Wolfgang Hönig

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: Propose an infinite completion tree framework to address computational resource allocation problems in task and motion planning, and present a heuristic method for balancing branch width and computational depth.

EgoPAT3Dv2: Predicting 3D Action Target from 2D Egocentric Vision for Human-Robot Interaction

Irving Fang, Chen Feng

Pose EstimationRobotic IntelligenceVideo

🎯 What it does: Expanded the scale and diversity of the EgoPAT3D dataset, improved baseline algorithms, introduced large pre-trained models and human prior knowledge, achieved 3D coordinate prediction of hand motion targets using only RGB first-person videos, and verified its practicality in simple HRI tasks on a real robot platform.

Elasto-Static Modelling and Identification of a Deployable Cable-Driven Parallel Robot with Compliant Masts*

Zane Zake, St´ephane Caro

Robotic IntelligencePhysics Related

🎯 What it does: This paper proposes a deployable cable-driven parallel robot named Rocaspect, which uses four deformable masts. The robot's behavior and accuracy were experimentally evaluated, and three different mast modeling schemes were proposed.

Eliminating Cross-modal Conflicts in BEV Space for LiDAR-Camera 3D Object Detection

Jiahui Fu, Si Liu

Object DetectionAutonomous DrivingImageMultimodalityPoint Cloud

🎯 What it does: Proposed the ECFusion method to eliminate cross-modal conflicts in the BEV space

Elliptical torus-based Six-axis FBG Force Sensor with In-situ Calibration for Condition Monitoring of Orthopedic Surgical Robot*

Tianliang Li, Zude Zhou

Robotic IntelligenceRecurrent Neural NetworkBiomedical DataMagnetic Resonance Imaging

🎯 What it does: A six-axis fiber Bragg grating (FBG) force/torque sensor based on an elliptical toroidal structure was proposed, along with an in-situ calibration method for force perception in orthopedic surgical robots. A multi-channel one-dimensional convolutional gated recurrent unit (M1-DCGRU) was combined to achieve rapid and accurate identification of seven drilling stages.

Embedded air channels transform soft lattices into sensorized grippers

Annan Zhang, Daniela Rus

Robotic Intelligence

🎯 What it does: Designed and manufactured a soft robotic gripper composed of two cubic lattices, with airways embedded inside the lattice to achieve sensing

EMIFF: Enhanced Multi-scale Image Feature Fusion for Vehicle-Infrastructure Cooperative 3D Object Detection

Zhe Wang, Ya-Qin Zhang

Object DetectionAutonomous DrivingTransformerImage

🎯 What it does: Proposed a camera-based 3D object detection framework called EMIFF for vehicle-infrastructure collaborative detection, addressing pose errors and information transmission loss issues.

Empirical Study of Ground Proximity Effects for Small-scale Electroaerodynamic Thrusters

Grant Nations, Daniel S. Drew

Physics Related

🎯 What it does: First empirical study on small electric air动力 propulsion devices (EAD), investigating performance changes when approaching the ground, including single and quadruple thruster arrays.

Enabling Large-scale Heterogeneous Collaboration with Opportunistic Communications

Fernando Cladera, Vijay Kumar

Robotic Intelligence

🎯 What it does: Developed and evaluated the MOCHA framework, which supports heterogeneous collaboration among multiple robots in large-scale environments under conditions of lack of continuous communication, and integrated with a drone planning stack.

Enabling passivity for Cartesian workspace restrictions

Sebastian Hjorth, Dimitrios Chrysostomou

Robotic Intelligence

🎯 What it does: Designed and implemented an energy-aware Cartesian impedance controller with a virtual workspace constraint to ensure the robot's passivity during collaborative disassembly processes.

Enabling the Deployment of Any-Scale Robotic Applications in Microservice Architectures through Automated Containerization*

Jean-Pierre Busch, Lutz Eckstein

Robotic Intelligence

🎯 What it does: This paper proposes and releases a set of automated tools for microservices-based development of robot applications based on ROS, including minimized containerization, machine learning-enabled base images, and CLI tools that simplify container image interactions.

End-to-end Reinforcement Learning for Time-Optimal Quadcopter Flight

Robin Ferede, G. D. de Croon

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed and implemented an end-to-end reinforcement learning (E2E RL) framework to achieve high-speed time-optimal flight for quadrotors and directly output motor commands.

End-to-end Semantic Segmentation Network for Low-Light Scenes

Hongmin Mu, Zhengcai Cao

SegmentationAutonomous DrivingConvolutional Neural NetworkImage

🎯 What it does: This paper proposes an end-to-end semantic segmentation network specifically designed for low-light scenarios, incorporating hierarchical gated convolutional units and a dual-loop dual-part graph matching algorithm;

End-to-End Semi-Supervised 3D Instance Segmentation with PCTeacher

Linfeng Li, Na Zhao

SegmentationPoint Cloud

🎯 What it does: Proposes an end-to-end semi-supervised 3D instance segmentation framework called PCTeacher based on the Mean Teacher paradigm, leveraging point-level and cluster-level pseudo labels to mine knowledge from unlabeled data.

End-to-End Thermal Updraft Detection and Estimation for Autonomous Soaring Using Temporal Convolutional Networks

Christian Gall, Aamir Ahmad

Autonomous DrivingConvolutional Neural NetworkTime Series

🎯 What it does: Propose an end-to-end deep learning method that can simultaneously detect multiple thermal updrafts and estimate their positions, intensities, and diffusion ranges.

Energy Consumption Modelling of Coaxial-Rotor in Vortex Ring State for Controllable High-speed Descending

Jiawei Sun, Shuang Feng

OptimizationRobotic IntelligencePhysics Related

🎯 What it does: Design a high-speed vertical descent coaxial rotor UAV for mountain rescue, propose a power management pipeline, and achieve a controlled vertical descent speed of 8 m/s.

Energy-Aware Ergodic Search: Continuous Exploration for Multi-Agent Systems with Battery Constraints

Adam Seewald, Ian Abraham

Optimization

🎯 What it does: Conduct energy-aware traversal search to maintain continuous exploration by at least one agent, balancing energy consumption with coverage quality.

Enhanced Human-Robot Collaboration with Intent Prediction using Deep Inverse Reinforcement Learning

Mukund Mitra, Pradipta Biswas

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a hand motion and target intent prediction system based on inverse reinforcement learning

Enhanced multifunctional interface for reconfigurability of robotic teams in planetary applications

Mehmed Yüksel, Frank Kirchner

Robotic Intelligence

🎯 What it does: Further improvements to the reliable electromechanical interface (EMI) in the TransTerrA project, verifying its docking and disassembly performance under various loads and tilt angles in field tests; simultaneously testing the power and data transmission reliability of the new contact block under water pressure and environmental factors.

Enhanced Robust Motion Control based on Unknown System Dynamics Estimator for Robot Manipulators

Xinyu Jia, Haoyong Yu

Robotic Intelligence

🎯 What it does: Proposed and implemented two robust motion control schemes for high-dimensional robot manipulators based on the Unknown System Dynamics Estimator (USDE).

Enhancement on Target-Gripper Alignment: A Tomato Harvesting Robot with Dual-Camera Image-Based Visual Servoing

Lu-Ching Wang, Feng-Li Lian

Robotic IntelligenceImageAgriculture Related

🎯 What it does: Developed a dual-camera image visual servo controller to improve the precise alignment between the tomato picking robot's gripper and the target tomato, and reduce picking time through cumulative error compensation.

Enhancing Inland Water Safety: The Lake Constance Obstacle Detection Benchmark

Dennis Grießer, G. Umlauf

Object DetectionImagePoint CloudBenchmark

🎯 What it does: Collected and annotated 1,974 stereo images and LiDAR point clouds using a LiDAR scanner and stereo camera on a research vessel, generating a new inland water obstacle detection dataset and providing preliminary methods and applicable evaluation metrics;

Enhancing mmWave Radar Point Cloud via Visual-inertial Supervision

Cong Fan, Wei Wang

Autonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Enhance millimeter-wave radar point clouds using low-cost cameras and inertial measurement units (IMU) through supervised learning.

Enhancing motion trajectory segmentation of rigid bodies using a novel screw-based trajectory-shape representation

Arno Verduyn, J. Schutter

SegmentationVideoTime Series

🎯 What it does: Proposed a new representation for rigid body motion trajectories that considers translation and rotation, with time and reference point invariance, and verified the effectiveness of the self-supervised segmentation method based on this representation in simulations and real human pouring actions.

Enhancing Tactile Sensing in Robotics: Dual-Modal Force and Shape Perception with EIT-based Sensors and MM-CNN

Haofeng Chen, Xiaojie Wang

Robotic IntelligenceConvolutional Neural NetworkBiomedical Data

🎯 What it does: Proposed a multi-modal convolutional neural network (MM-CNN) based on electrical impedance tomography (EIT) to achieve joint perception of force and shape.

Enhancing Task Performance of Learned Simplified Models via Reinforcement Learning

Hien Bui, Michael Posa

Robotic IntelligenceReinforcement Learning

🎯 What it does: Learn a simplified dynamic model using policy gradient algorithms to explicitly maximize task performance, applicable to robot hands with rich contact manipulating unknown objects.

Enhancing the Tracking Performance of Passivity-based High-Frequency Robot Cloud Control

Fabian Jakob, Sami Haddadin

Robotic Intelligence

🎯 What it does: This paper studies the migration of high-frequency robot controllers to remote computing services, ensuring the stability of networked systems by guaranteeing the passivity of subsystems and the time-domain passivity of communication channels (TDPA); reducing the conservatism of TDPA by leveraging bilateral model knowledge, identifying passivity excess; avoiding excessive energy dissipation by enhancing allowable passivity gaps; eliminating tracking offset and providing convergence guarantees through a position drift compensation algorithm, and experimentally validating on a 7-degree-of-freedom Franka Research 3 robot, showing significant improvements in tracking performance under high communication latency scenarios.

Enhancing Visual Inertial SLAM with Magnetic Measurements

Bharat Joshi, Ioannis Rekleitis

OptimizationSimultaneous Localization and MappingMultimodality

🎯 What it does: Introduce tightly coupled fusion of magnetometer measurements in Visual-Inertial Odometry (VIO), optimize keyframes via sliding window optimization to minimize reprojection error, relative inertial error, and relative magnetometer orientation error. Utilize IMU orientation propagation to transform magnetometer measurements across frames, generating relative orientation constraints between adjacent frames; calibrate soft and hard iron effects using elliptical fitting algorithms.

Enhancing Visual Place Recognition with Multi-modal Features and Time-constrained Graph Attention Aggregation

Zhuo Wang, Meiqi Pei

RecognitionGraph Neural NetworkTransformerMultimodality

🎯 What it does: Proposes a multi-modal visual place recognition method that leverages depth information, employing dual-branch feature extraction, shared multi-modal feature fusion based on Transformer (SFFM), and time-constrained graph attention aggregation (TC-GAT)

Ensemble Latent Space Roadmap for Improved Robustness in Visual Action Planning

M. Lippi, Danica Kragic

Representation LearningRobotic IntelligenceGraph Neural NetworkImage

🎯 What it does: Enhancing the robustness of learning-based latent space planning systems through an ensemble approach, by constructing and combining multiple Latent Space Roadmap (LSR) instances for visual action planning;

Envibroscope: Real-Time Monitoring and Prediction of Environmental Motion for Enhancing Safety in Robot-Assisted Microsurgery

Alireza Alikhani, M. Nasseri

Safty and PrivacyRobotic Intelligence

🎯 What it does: Proposed and implemented an environment motion analysis system based on grid-distributed sensor nodes for real-time monitoring and prediction of unexpected movements in minimally invasive surgery environments.

Environment-Modulated Self-Assembly by Changes in Modules’ Buoyancy

Xiao Chen, Shuhei Miyashita

Physics Related

🎯 What it does: A self-assembly method that automatically selects and provides modules using environmental signals, changing module configurations through different environments (flat, low-density saline, saturated saline).

Environmental Awareness Dynamic 5G QoS for Retaining Real Time Constraints in Robotic Applications

G. Damigos, G. Nikolakopoulos

Robotic Intelligence

🎯 What it does: By constructing a dynamic selection framework based on 5G QoS functions, autonomous QoS data stream switching for drones in time-critical applications is achieved using a probabilistic finite state machine (PFSM), improving performance.

EnYOLO: A Real-Time Framework for Domain-Adaptive Underwater Object Detection with Image Enhancement

Junjie Wen, Benwen Chen

Object DetectionDomain AdaptationConvolutional Neural NetworkImage

🎯 What it does: Propose the EnYOLO framework, which integrates real-time underwater image enhancement and object detection while incorporating domain adaptation capabilities.

EquivAct: SIM(3)-Equivariant Visuomotor Policies beyond Rigid Object Manipulation

Jingyun Yang, Jeannette Bohg

Representation LearningRobotic IntelligenceContrastive LearningPoint Cloud

🎯 What it does: Proposes EquivAct, a method that constructs visual representations and control strategies using SIM(3)-equivariant network architectures. It first performs contrastive pre-training on point clouds in simulated scenarios, then fine-tunes the equivariant visuomotor policy with a small number of source task demonstrations, achieving zero-shot transfer to new objects with significant differences in scale, position, and pose.

ERASOR++: Height Coding Plus Egocentric Ratio Based Dynamic Object Removal for Static Point Cloud Mapping

Jiabao Zhang, Yu Zhang

SegmentationPoint Cloud

🎯 What it does: Proposes ERASOR++, a dynamic object removal method based on the egocentric ratio of pseudo occupancy, and introduces the height coding descriptor as well as the height stack test, ground layer test, and surrounding point test to accurately and effectively identify dynamic object boxes in point clouds.

ESP: Extro-Spective Prediction for Long-term Behavior Reasoning in Emergency Scenarios

Dingrui Wang, Wei Li

Autonomous DrivingLarge Language Model

🎯 What it does: Constructed a new long-term prediction dataset and proposed a pluggable feature encoder along with a new evaluation metric.

Estimating 3D Uncertainty Field: Quantifying Uncertainty for Neural Radiance Fields

Jianxiong Shen, Francesc Moreno-Noguer

Neural Radiance Field

🎯 What it does: Proposes a 3D uncertainty field estimation method based on NeRF to quantify uncertainty in unseen spaces (including occluded and external regions of the scene).

Estimating Material Properties of Interacting Objects Using Sum-GP-UCB

M. Seker, Oliver Kroemer

OptimizationPhysics Related

🎯 What it does: Proposes a method based on Bayesian optimization for identifying object material property parameters by utilizing observed interaction scenarios to estimate the material and dynamic parameters of objects.

Ethically Compliant Autonomous Systems under Partial Observability

Qingyuan Lu, Stuart Russell

OptimizationRobotic Intelligence

🎯 What it does: Proposes a Partially Observable Ethical Compliance Autonomous System (PO-ECAS) framework and approximately solves the optimal ethical compliance strategy using MILP; extends the existing Prima Facie duties ethics framework to the belief space and introduces a virtue ethics framework inspired by Aristotle's golden mean; validates the effectiveness of the method in a simulated campus patrol robot scenario.

Evaluating Robustness of Visual Representations for Object Assembly Task Requiring Spatio-Geometrical Reasoning

Chahyon Ku, K. Desingh

Representation LearningRobotic IntelligenceTransformerReinforcement LearningImageBenchmark

🎯 What it does: Evaluate and benchmark the robustness of visual representations in object assembly tasks (e.g., peg-in-hole tasks), investigate their grasping variability in dual-arm operations, and conduct experiments using a vision-motion policy learning framework with visual pre-trained models as visual encoders; meanwhile, explore the improvement of rotation representations and associated loss functions on policy learning, and propose new task scenarios to measure the robustness of complex assembly tasks.

Everyday finger: a robotic finger that meets the needs of everyday interactive manipulation

R. Ornelas, Pulkit Agrawal

Robotic Intelligence

🎯 What it does: Designed a robot finger called Everyday Finger capable of performing various daily tasks, and proposed a novel actuator and finger structure.

eWand: An extrinsic calibration framework for wide baseline frame-based and event-based camera systems

T. Gossard, Andreas Zell

Pose EstimationImageMultimodality

🎯 What it does: Propose the eWand framework, which uses flickering LEDs placed inside an opaque sphere to replace traditional 2D patterns, achieving external calibration of wide-baseline multi-camera systems.

Excitation Trajectory Optimization for Dynamic Parameter Identification Using Virtual Constraints in Hands-on Robotic System

Huanyu Tian, Christos Bergeles

OptimizationRobotic Intelligence

🎯 What it does: Propose a more efficient robot trajectory optimization method for dynamic parameter identification, with emphasis on self-collision avoidance.

ExoRecovery: Push Recovery with a Lower-Limb Exoskeleton Based on Stepping Strategy

Zeynep Ozge Orhan, Mohamed Bouri

OptimizationRobotic Intelligence

🎯 What it does: This study proposes an omnidirectional balance recovery gait planning framework based on online optimization, which real-time optimizes step length and position, generates foot and joint trajectories through inverse kinematics, and subsequently executes them via a damping controller to achieve collaborative recovery across feet;

Exoskeleton-Mediated Physical Human-Human Interaction for a Sit-to-Stand Rehabilitation Task

Lorenzo Vianello, José Luis Pons Rovira

Robotic Intelligence

🎯 What it does: Proposed and designed a sit-up rehabilitation framework utilizing two lower-limb exoskeletons for physical interaction between therapists and patients

Experience Consistency Distillation Continual Reinforcement Learning for Robotic Manipulation Tasks

Chao Zhao, Xuguang Lan

Knowledge DistillationRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed an experience consistency distillation method for robotic continuous reinforcement learning to improve the compression rate and information content of old task experiences, while ensuring data distribution consistency through Fréchet Inception Distance (FID) regularization.

Experimental comparison of pinwheel and non-pinwheel designs of 3D-printed cycloidal gearing for robotics

Wesley Roozing, G. Roozing

Robotic Intelligence

🎯 What it does: Conducted an experimental comparative study on two designs of 3D-printed cycloidal gears: pinwheel and non-pinwheel.

Expert Composer Policy: Scalable Skill Repertoire for Quadruped Robots

Guilherme Christmann, Wei-Chao Chen

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposes an expert combination strategy to reliably expand the skill library of quadruped robots and achieve non-interfering sequential combination among experts.

Exploitation-Guided Exploration for Semantic Embodied Navigation

Justin Wasserman, Unnat Jain

Reinforcement Learning

🎯 What it does: Propose the XgX method that combines exploration and exploitation modules, where the exploitation module takes over the navigation task when the goal is visible, and the strategy of the exploration module is enhanced through teacher forcing.

Exploring the Effect of Base Compliance on Physical Human-Robot Collaboration

Ziqi Wang, Marc G. Carmichael

Robotic Intelligence

🎯 What it does: Investigated the impact of base flexibility on mobile physical human-robot collaboration by varying the base stiffness of collaborative robots and measuring metrics such as physical exertion, task speed, and task error in human operators.

Exploring the Impact of Narrator Type on Response Latency and Utterance Length During Interactive Storytelling

Iman Bakhoda, W. Louie

🎯 What it does: In an inter-subject experiment involving 28 participants, researchers compared the effects of human narrators and robot narrators on participants' response latency and speech length during interactive storytelling.

Exploring the Needle Tip Interaction Force with Retinal Tissue Deformation in Vitreoretinal Surgery

Simon Pannek, Nassir Navab

Robotic IntelligenceImageBiomedical Data

🎯 What it does: Collected and synchronized interaction forces between the needle tip and retinal tissue measured by FBG sensors with OCT B-scan images, constructed a specialized dataset, and proposed a neural network model for force estimation based on images.

Exploring Transformers and Visual Transformers for Force Prediction in Human-Robot Collaborative Transportation Tasks

J. E. Domínguez-Vidal, Alberto Sanfeliu

Robotic IntelligenceTransformer

🎯 What it does: Using Transformers and Visual Transformers to predict human effort in human-robot collaborative transportation tasks, modifying the model architecture to achieve dual outputs (force prediction and velocity prediction), and evaluating performance on test sets and real experiments.

Extending Guiding Vector Field to track unbounded UAV paths

Mael Feurgard, Simon Lacroix

Robotic Intelligence

🎯 What it does: Propose a dynamic step size adaptation strategy to improve the parametric guiding vector field method for tracking unbounded paths.

Extending the Cooperative Dual-Task Space in Conformal Geometric Algebra

Tobias Löw, Sylvain Calinon

Robotic Intelligence

🎯 What it does: Studied methods to extend the collaborative dual-task space (CDTS) under the conformal geometric algebra framework, and conducted experimental validation on a dual-arm Franka Emika robot.

Extreme Parkour with Legged Robots

Xuxin Cheng, Deepak Pathak

Robotic IntelligenceReinforcement LearningImage

🎯 What it does: On small, low-cost robots, using a single front depth camera and a trained single neural network policy, the robot performs extreme parkour actions such as high jumps, overcoming obstacles twice its own length, handstands, slope running, and generalizes to new obstacle scenarios with different physical properties.

Extremum-Seeking Action Selection for Accelerating Policy Optimization

Ya-Chien Chang, Sicun Gao

OptimizationReinforcement Learning

🎯 What it does: Propose to adaptively improve actions sampled from a high-entropy random policy through extreme value tracking control (ESC), aiming to accelerate policy optimization in model-free reinforcement learning.

F3DMP: Foresighted 3D Motion Planning of Mobile Robots in Wild Environments

Andong Yang, Yu Hu

OptimizationRobotic IntelligenceReinforcement LearningImage

🎯 What it does: Proposes F3DMP—a proactive 3D motion planning method for mobile robots in outdoor environments, integrating 3D spatial planning with a time allocation function based on offline reinforcement learning;

Facile Integration of Robots into Experimental Orchestration at Scientific User Facilities

Chandima Fernando, Phillip M. Maffettone

Robotic Intelligence

🎯 What it does: Developed and demonstrated a framework based on ROS2 and Bluesky, integrating robots into scientific user facilities to enable scalable robotic applications.

Fall Prediction for Bipedal Robots: The Standing Phase

M. E. Mungai, J. Grizzle

Anomaly DetectionRobotic IntelligenceConvolutional Neural Network

🎯 What it does: A 1D convolutional neural network (CNN) was developed to create an algorithm for fall prediction in bipedal robots during standing, capable of detecting sudden, latent, and intermittent faults and estimating the time-to-fall before a fall occurs.