arXivSub Start free trial

IROS 2024 Papers — Page 14

IEEE/RSJ International Conference on Intelligent Robots and Systems · 1581 papers

Self-Supervised Monocular Depth Estimation with Effective Feature Fusion and Self Distillation

Zhenfei Liu, Xiaoyang Wang

Depth EstimationKnowledge DistillationTransformerImage

🎯 What it does: Proposed a self-supervised monocular depth estimation model, including an internal fusion Transformer backbone, a bidirectional attention module, and adopted three data augmentation techniques for self-distillation: cropping-resizing, cropping-shuffling, and mirroring

Self-Supervised Motion Segmentation with Confidence-Aware Loss Functions for Handling Occluded Pixels and Uncertain Optical Flow Predictions

Chun-Yu Chen, Chieh-Chih Wang

SegmentationAutonomous DrivingOptical FlowImage

🎯 What it does: Proposed two loss functions to improve self-supervised motion segmentation, mitigating the impact of occluded pixels and optical flow errors.

Semantic Belief Behavior Graph: Enabling Autonomous Robot Inspection in Unknown Environments

M. Ginting, A. Agha-mohammadi

Robotic IntelligenceGraph Neural NetworkReinforcement Learning

🎯 What it does: Proposes a Semantic Belief Behavior Graph (SB2G) framework for autonomous robot inspection in unknown environments, utilizing semantic behavior nodes and proactive semantic search behaviors to generate control strategies.

Semantic Layering in Room Segmentation via LLMs

Taehyeon Kim, Byung-Cheol Min

SegmentationRobotic IntelligenceTransformerLarge Language Model

🎯 What it does: Proposes a semantic room segmentation method (SeLRoS) that integrates large language models (LLMs) with traditional 2D map segmentation, enhancing robot navigation capabilities by adding semantic data such as object recognition and spatial relationships to the segmented map.

Semantic SLAM Fusing Moving Constraint for Dynamic Objects under Indoor Environments

Zhenyuan Yang, M. R. Elara

Object DetectionSimultaneous Localization and MappingPoint CloudBenchmark

🎯 What it does: Proposed a real-time RGB-D SLAM that integrates point, line, and plane features with object detection to enhance robustness in dynamic environments

Semantics from Space: Satellite-Guided Thermal Semantic Segmentation Annotation for Aerial Field Robots

Connor T. Lee, Soon-Jo Chung

SegmentationTransformerSimultaneous Localization and MappingImageAgriculture Related

🎯 What it does: Automatically generate semantic segmentation annotations for aerial thermal imaging using satellite data and onboard positioning information.

Sensor-agnostic Visuo-Tactile Robot Calibration Exploiting Assembly-Precision Model Geometries

Manuel Gomes, Jianwei Zhang

OptimizationRobotic IntelligenceMultimodalityMesh

🎯 What it does: Proposes a general tactile-based robot calibration method that supports self-touch throughout the system's kinematics and seamlessly integrates tactile and visual information within the ATOM calibration framework.

Sensorimotor Attention and Language-based Regressions in Shared Latent Variables for Integrating Robot Motion Learning and LLM

Kanata Suzuki, Tetsuya Ogata

Robotic IntelligenceTransformerLarge Language Model

🎯 What it does: Proposed an integrated method that connects robot motion learning models with large language models (LLM) by leveraging shared latent variables, and updates shared parameters during robot motion generation based on predictive errors from sensorimotor attention points and task language instructions.

SePaint: Semantic Map Inpainting via Multinomial Diffusion

Zheng Chen, Lantao Liu

RestorationGenerationDiffusion modelImage

🎯 What it does: Proposed SePaint, a filling model for semantic Bird’s-Eye-View maps, which utilizes generative multinomial diffusion to generate semantic information for missing regions.

Sequential Convex Programming for Time-optimal Quadrotor Waypoint Flight

Zhipeng Shen, Hailong Huang

Optimization

🎯 What it does: Convert the non-convex optimal control problem of six-degree-of-freedom quadrotor into convex subproblems using the Sequential Convex Programming (SCP) algorithm, and simultaneously optimize waypoint timing and trajectory through state-triggered constraints to achieve time-optimal waypoint flight;

Sequential Discrete Action Selection via Blocking Conditions and Resolutions

Liam Merz Hoffmeister, Daniel Rakita

Robotic IntelligenceGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringTextGraph

🎯 What it does: Propose a framework for addressing the problem of robot sequential action selection by resolving blocking conditions, implemented through a state transition graph and a zero-shot LLM; this strategy generates prompts to enable the LLM to make individual action decisions at each step, updating the state graph after execution until the goal is achieved or terminated.

Seven Benefits of Using Series Elastic Actuators in the Design of an Affordable, Simple Controlled, and Functional Prosthetic Hand

Erfan Koochakzadeh, Rezvan Nasiri

Robotic Intelligence

🎯 What it does: Designed and implemented a 3D-printed prosthetic hand, using a series of elastic actuators (SEA) composed of two antagonistic motors, pulleys, ropes, and springs to achieve synchronized finger extension and retraction. Synchronized motion of fingers was realized through optimization of pulley diameters, and the thumb is adjustable to adapt to different tasks. Experimental validation demonstrated its functionality and controllability.

SFTrack: A Robust Scale and Motion Adaptive Algorithm for Tracking Small and Fast Moving Objects

Inpyo Song, Jangwon Lee

Object TrackingVideo

🎯 What it does: A more concise and efficient method for multi-target tracking in UAV video is proposed, adopting a tracking strategy initiated with low-confidence detections and revisiting traditional appearance matching algorithms.

SGNet: Salient Geometric Network for Point Cloud Registration

Qianliang Wu, Jian Yang

Pose EstimationTransformerPoint Cloud

🎯 What it does: Proposes the SGNet framework, which includes a semantic-aware geometric encoder, prior knowledge-based selection of salient points using intrinsic shape signatures, an innovative high-order Transformer, an anchor node selection strategy, and a Sinkhorn matching module for point cloud registration.

SGOR: Outlier Removal by Leveraging Semantic and Geometric Information for Robust Point Cloud Registration

Guiyu Zhao, Hongbin Ma

Anomaly DetectionPoint Cloud

🎯 What it does: Proposed a new method that utilizes geometric and semantic information for outlier removal, achieving robust point cloud registration.

Shadow Maintenance for Automatic Light-Probe Control in Ophthalmic Surgeries Using Only 2D information

Junjie Yang, I. M. A. N. Fellow

Robotic IntelligenceBiomedical Data

🎯 What it does: Proposes a method to automatically control the illumination probe using only two-dimensional microscope image information, limiting shadow positions around the probe tip and incorporating an intensity balance submodule to ensure normal light intensity distribution and safe probe tip depth;

Shape-prior Free Space-time Neural Radiance Field for 4D Semantic Reconstruction of Dynamic Scene from Sparse-View RGB Videos

Sandika Biswas, H. Rezatofighi

SegmentationGenerationNeural Radiance FieldImageVideo

🎯 What it does: Propose a 3D neural radiance field technique without shape priors for detailed geometric reconstruction of human-object interactions in dynamic scenes under sparse-view RGB videos, generating explicit surfaces and semantic labels.

ShapeGrasp: Zero-Shot Task-Oriented Grasping with Large Language Models through Geometric Decomposition

Samuel Li, Simon Stepputtis

Robotic IntelligenceGraph Neural NetworkTransformerLarge Language ModelGraph

🎯 What it does: Propose a zero-shot task-oriented grasping method called ShapeGrasp based on geometric decomposition, which can achieve task-oriented grasping of unknown objects without requiring training data by decomposing the target object's geometry and representing it with a graph structure.

Sharing Attention Mechanism in V-SLAM: Relative Pose Estimation with Messenger Tokens on Small Datasets

Dun Dai, Kai-Yuan Cai

Pose EstimationTransformerSimultaneous Localization and MappingImage

🎯 What it does: Proposed and implemented an end-to-end relative pose estimation method based on shared attention mechanism, messenger tokens, and dual embeddings for V-SLAM in small indoor scenes.

SiCP: Simultaneous Individual and Cooperative Perception for 3D Object Detection in Connected and Automated Vehicles

Deyuan Qu, Qing Yang

Object DetectionAutonomous DrivingPoint Cloud

🎯 What it does: Proposes the Simultaneous Individual and Cooperative Perception (SiCP) framework, and employs a lightweight Dual-Perception Network (DP-Net) to simultaneously achieve individual and collaborative 3D object detection.

Side-scan sonar based landmark detection for underwater vehicles

Simon A. Hoff, Damiano Varagnolo

Autonomous DrivingComputational EfficiencySimultaneous Localization and Mapping

🎯 What it does: A pipeline is proposed and analyzed to convert raw sidescan sonar data from underwater vehicles into executable information in real-time for SLAM; the pipeline includes slice processing for blind spot elimination and noise suppression, conversion of slices into probability maps, and detection and classification of underwater landmarks from probability maps.

Signal Temporal Logic-Guided Apprenticeship Learning

Aniruddh Gopinath Puranic, S. Nikolaidis

Robotic IntelligenceGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: Propose a framework that encodes temporal logic specifications into graphs, defining a temporal-based metric to evaluate the behaviors of demonstrators and learning agents, thereby improving the quality of reward and control strategy learning.

Sim-to-Real Domain Shift in Online Action Detection

Constantin Patsch, Eckehard G. Steinbach

Object DetectionDomain AdaptationVideoBenchmark

🎯 What it does: This paper introduces the Human Kitchen Interaction (HKI) dataset to study the domain transfer problem from synthetic to real environments in online action detection, and evaluates existing state-of-the-art models;

Sim2real Cattle Joint Estimation in 3D point clouds

Mohammad Okour, A. Alempijevic

Pose EstimationDomain AdaptationConvolutional Neural NetworkPoint CloudAgriculture Related

🎯 What it does: Constructed a cow body posture dataset based on a single animated 3D model, using surface curvature to train a deep learning framework for estimating internal joints, and extracting joints through geodesic distances on the surface manifold combined with multilateration, followed by predicting the cow's hip height using all proportional relationships.

Sim2Real Transfer for Audio-Visual Navigation with Frequency-Adaptive Acoustic Field Prediction

Changan Chen, Kristen Grauman

Domain AdaptationRobotic IntelligenceMultimodalityAudio

🎯 What it does: Defined the Acoustic Field Prediction (AFP) task, trained frequency-specific AFP models, evaluated prediction errors on real data, proposed a frequency-adaptive selection strategy based on prior measurements and audio energy distribution, and combined it with landmark navigation to validate its transfer effectiveness on simulated and real robots.

Similarity Distance-Based Label Assignment for Tiny Object Detection

Shuohao Shi, Xin Xu

Object DetectionHyperparameter SearchImage

🎯 What it does: Proposed a simple and effective strategy called Similarity Distance (SimD) for evaluating the similarity of bounding boxes and assigning labels in small object detection.

Simulation-Assisted Learning for Efficient Bin-Packing of Deformable Packages in a Bimanual Robotic Cell

O. Manyar, Satyandra K. Gupta

OptimizationRobotic Intelligence

🎯 What it does: Studied and implemented a system for dual-arm robots to load deformable packages into boxes, and proposed a simulation-assisted learning framework for action prediction to optimize loading efficiency.

Simultaneous Super-resolution and Depth Estimation for Satellite Images Based on Diffusion Model

Yuwei Zhou, Yangming Lee

Depth EstimationSuper ResolutionDiffusion modelImage

🎯 What it does: This paper proposes a method that first uses a diffusion model to enhance satellite images to super-resolution, then performs depth estimation and 3D reconstruction based on the enhanced images, ultimately generating a 3D surface model with detailed landscape information.

Single Actuator Undulation Soft-bodied Robots Using A Precompressed Variable Thickness Flexible Beam

Tung D. Ta

Robotic Intelligence

🎯 What it does: Proposed a tendon-driven flexible beam using a single actuator, which forms an S-shape through pre-compression to generate mechanical transmission waves, enabling undulatory motion in soft robots. Experimental verification demonstrated the relationship between tendon pre-tension and motor winding/unwinding.

Single Protoplasts Pickup System Combining Brightfield and Confocal Images

Daito Ando, Fumihito Arai

Image

🎯 What it does: This study developed a system combining confocal and bright view microscopy for precise measurement and picking of bud cells after root tip cell wall removal;

Single-Shot 6DoF Pose and 3D Size Estimation for Robotic Strawberry Harvesting

Lun Li, H. Kasaei

Pose EstimationRobotic IntelligenceImageAgriculture Related

🎯 What it does: Proposed a deep learning method for estimating the 6DoF pose and 3D dimensions of strawberries to improve robotic picking efficiency.

SiSCo: Signal Synthesis for Effective Human-Robot Communication Via Large Language Models

Shubham D. Sonawani, H. B. Amor

Robotic IntelligenceTransformerLarge Language Model

🎯 What it does: Propose the SiSCo framework, which automatically generates visual signals using large language models (LLM) and mixed reality technology to enhance human-robot collaboration communication efficiency

Skeleton-Based Human Action Recognition with Noisy Labels

Yi Xu, Rainer Stiefelhagen

RecognitionMixture of ExpertsGraph

🎯 What it does: This paper proposes and verifies a framework targeting label noise in skeleton action recognition, and on this basis, designs a new method called NoiseEraSAR

Skill Q-Network: Learning Adaptive Skill Ensemble for Mapless Navigation in Unknown Environments

Hyunki Seong, D. H. Shim

Autonomous DrivingReinforcement Learning

🎯 What it does: Propose Skill Q-Network (SQN) to learn adaptive skill sets for mapless navigation in unknown environments.

Skill Transfer and Discovery for Sim-to-Real Learning: A Representation-Based Viewpoint

Haitong Ma, Na Li

Domain AdaptationRepresentation LearningRobotic IntelligenceReinforcement Learning

🎯 What it does: Studied simulation-to-real skill transfer and discovery using representation learning, and proposed a skill representation and orthogonal constrained skill discovery method based on the spectral decomposition of Markov decision processes, demonstrating the transfer of skills from simulation to a real quadrotor and improving control performance.

Skin the sheep not only once: Reusing Various Depth Datasets to Drive the Learning of Optical Flow

Shengyu Huang, Wei-Chen Chiu

ClassificationData SynthesisOptical Flow

🎯 What it does: Utilize geometric relationships to reconstruct multiple real-world depth datasets into stereo images, generating supervised optical flow training data, and further enhance the learning of optical flow estimators through geometric augmentation and auxiliary classifiers.

SLIP Embodied Robust Quadruped Robot Control

J. Hong, Sehoon Oh

Robotic Intelligence

🎯 What it does: Proposes a control method that projects SLIP dynamics onto multi-joint legs, and constructs a complete quadruped robot motion control framework based on impedance control in the rotational workspace using a force observer (RWFOB), connecting trunk and leg movements through the Jacobian matrix.

Small Multi-Rotor UAV Oriented Direct Thrust Sensor Based on Lightweight Barometers

Han Jiang, Yuqing He

Robotic Intelligence

🎯 What it does: Developed an embedded barometric pressure sensor (BFS) to directly measure the thrust generated by rotors between the multi-rotor drone body and motors, completing the sensor's design, parameter modeling, temperature stability improvement, installation structure optimization, and thrust model reconstruction;

SMART-LLM: Smart Multi-Agent Robot Task Planning using Large Language Models

S. S. Kannan, Byung-Cheol Min

Robotic IntelligenceTransformerLarge Language ModelAgentic AIPrompt EngineeringTextBenchmark

🎯 What it does: Propose the SMART-LLM framework, which utilizes large language models (LLMs) to convert high-level task instructions into multi-robot task plans, achieved through stages such as task decomposition, coalition formation, and task allocation.

SmartKit: User-Friendly Robot with Multiple Operating Systems

Guanyu Chen, Pan Lv

Robotic Intelligence

🎯 What it does: Proposed and implemented SmartKit, a multi-operating system mixed-criticality system (MCS) leveraging virtualization technology for mobile robots, demonstrating its software and hardware architecture, and verifying the system's performance and functionality.

SmartPathfinder: Pushing the Limits of Heuristic Solutions for Vehicle Routing Problem with Drones Using Reinforcement Learning

N. Imran, Myounggyu Won

OptimizationReinforcement Learning

🎯 What it does: Developed a framework that integrates reinforcement learning (RL) with traditional heuristic methods to enhance the solution quality and computational speed for vehicle and drone path planning problems.

Smooth Invariant Interpolation on Lie groups with Prescribed Terminal Conditions for Robot Motion Planning and Modeling of Soft Robots

Andreas Müller, H. Gattringer

Robotic Intelligence

🎯 What it does: Studied a method for smooth invariant interpolation on Lie groups, capable of generating trajectories given terminal pose, velocity, and acceleration (as well as initial conditions), with examples in drone docking tasks and Cosserat beam deformation.

SMORE-SLAM: Semantic Monocular SLAM with Scale Correction and Reverse Loop Utilization in Outdoor Environments

Yushi Chen, Haiyong Luo

Autonomous DrivingOptimizationSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Developed a semantic monocular SLAM system called SMORE-SLAM, which utilizes scale correction and reverse loop closure modules to enhance localization accuracy.

SNF-Feat: Semantic-Guided Negative-Sample-Free Representation Learning for Local Feature Extraction

Xun Zhou, Qi Chen

Representation Learning

🎯 What it does: Proposed a semantics-guided negative-sample-free learning method called SNF-Feat for local feature extraction.

SNU-Avatar Haptic Glove: Novel Modularized Haptic Glove via Trigonometric Series Elastic Actuators

E. Sung, Jaeheung Park

Pose EstimationRobotic Intelligence

🎯 What it does: Developed a modular tactile glove based on triangular series elastic actuators, which can provide force feedback to the fingertips during remote operation and estimate hand posture through algorithms.

SOAR: Simultaneous Exploration and Photographing with Heterogeneous UAVs for Fast Autonomous Reconstruction

Mingjie Zhang, Boyu Zhou

OptimizationSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Designed and implemented a LiDAR-visual heterogeneous multi-UAV system SOAR for rapid autonomous reconstruction of complex environments; achieved rapid acquisition of scene surface geometry and high-quality image coverage through LiDAR detector's front exploration and camera image collection;

Social Navigation in Crowded Environments with Model Predictive Control and Deep Learning-Based Human Trajectory Prediction

Viet-Anh Le, Ehsan Moradi-Pari

Autonomous DrivingOptimizationRobotic IntelligenceRecurrent Neural NetworkSequential

🎯 What it does: Propose combining Social-LSTM with MPC to plan optimal control actions for robots using predicted pedestrian trajectories to achieve social navigation.

Socially Integrated Navigation: A Social Acting Robot with Deep Reinforcement Learning

Daniel Flögel, Sören Hohmann

ClassificationRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose a new socialized integrated navigation method that enables robots' social behaviors to be adaptive and naturally generated through interactions with humans.

SocialNav-FTI: Field-Theory-Inspired Social-aware Navigation Framework based on Human Behavior and Social Norms

Siyi Lu, Run Liu

Autonomous DrivingRobotic IntelligenceSupervised Fine-TuningReinforcement LearningVideoPhysics Related

🎯 What it does: Propose a navigation framework based on social norms, using a physics-informed neural network to predict pedestrian motion and combined with reinforcement learning to achieve navigation.

Soft finger rotational stability for precision grasps

Hun Jang, Kevin Haninger

Robotic Intelligence

🎯 What it does: Proposed a rotational stability analysis model for precise grasping with soft fingers, and conducted experimental verification on various objects, grasping conditions, as well as PneuNet and commercial soft fingers.

Soft Task Planning with Hierarchical Temporal Logic Specifications

Ziyang Chen, Zhen Kan

Optimization

🎯 What it does: Propose using soft constraints in linear temporal logic (LTL) for task planning, and generate feasible task plans that meet constraints and minimize cost through hierarchical temporal logic specifications and hierarchical iterative search (HIS) algorithm.

SoftMAC: Differentiable Soft Body Simulation with Forecast-based Contact Model and Two-way Coupling with Articulated Rigid Bodies and Clothes

Min Liu, Lin Shao

Physics Related

🎯 What it does: Proposed the SoftMAC framework to achieve differentiable coupling between soft bodies, joint rigid bodies, and clothing, and introduced predictive contact models and penetration tracking algorithms.

SoftNeRF: A Self-Modeling Soft Robot Plugin for Various Tasks

Jiwei Shan, Hesheng Wang

Robotic IntelligenceNeural Radiance FieldImage

🎯 What it does: Developed SoftNeRF, a self-modeling method for soft robots based on self-supervised vision, which can learn the geometry and nonlinear motion of soft robots from RGB images through differentiable rendering, thereby enabling simulation and prediction of its future states.

Solving Dynamic Cosserat Rods with Frictional Contact Using the Shooting Method and Implicit Surfaces

Radhouane Jilani, Erwan Kerrien

Biomedical DataPhysics RelatedOrdinary Differential Equation

🎯 What it does: Solved the strong form closed-free boundary value problem for dynamic Cosserat rods with frictional contact using the shooting method, and handled contact reactions with the penalty function method;

Solving Multi-Robot Task Allocation and Planning in Trans-media Scenarios

Virgile De La Rochefoucauld, Haruo Takemura

OptimizationRobotic Intelligence

🎯 What it does: Proposes a new method for task allocation and planning in multi-robot systems under cross-media environments, capable of addressing specific complexities and constraints by decomposing the overall planning process into manageable subproblems (coalition formation, path planning, and task scheduling).

SOS-Match: Segmentation for Open-Set Robust Correspondence Search and Robot Localization in Unstructured Environments

Annika Thomas, Jonathan P. How

SegmentationRobotic IntelligenceSimultaneous Localization and MappingImage

🎯 What it does: Proposed the SOS-Match framework for object detection and matching in unstructured environments, comprising a frontend that uses a zero-shot segmentation model to extract and track object masks, and a backend that leverages object geometric consistency for frame alignment to achieve localization.

Sparse Points to Dense Clouds: Enhancing 3D Detection with Limited LiDAR Data

Aakash Kumar, Mubarak Shah

GenerationAutonomous DrivingImageMultimodalityPoint Cloud

🎯 What it does: Propose a method to reconstruct a complete 3D point cloud using only 512 point cloud data (about 1% of a full LiDAR frame) and a single image, and combine the reconstructed results with a multi-modal detector to achieve 3D object detection.

SparseGTN: Human Trajectory Forecasting with Sparsely Represented Scene and Incomplete Trajectories

Jianbang Liu, Max Q.-H. Meng

Autonomous DrivingComputational EfficiencyRepresentation LearningGraph Neural NetworkTransformerGraphTime Series

🎯 What it does: Propose a sparse graph representation for scenes and incomplete trajectories, and design a hierarchical graph transformer network called SparseGTN to predict multiple possible future trajectories of the target pedestrian.

Spatial Spinal Fixation: A Transformative Approach Using a Unique Robot-Assisted Steerable Drilling System and Flexible Pedicle Screw

Susheela Sharma, F. Alambeigi

Robotic IntelligenceBiomedical DataComputed Tomography

🎯 What it does: Developed a spatial spinal fixation (SSF) method combining the CT-SDR robot and flexible pedicle screws (FPS), enabling FPS installation on both planar and non-planar trajectories within the vertebra

Spatio-Temporal Consistent Mapping of Growing Plants for Agricultural Robots in the Wild

Luca Lobefaro, C. Stachniss

Robotic IntelligenceSimultaneous Localization and MappingPoint CloudAgriculture Related

🎯 What it does: Propose a system based on 3D plant models and consumer RGB-D cameras to achieve robot localization and model adaptation in dynamically changing environments, using deep learning feature descriptors and geometric information for cross-time 3D point matching, and updating the 3D model through non-rigid registration.

Spatiotemporal Co-Design Enabling Prioritized Multi-Agent Motion Planning

Yunshen Huang, Shen Zeng

Autonomous DrivingOptimization

🎯 What it does: Proposes a priority-based multi-agent motion planner based on the integration of spatial and temporal sequences

SPDAGG-TransNet: Integrating Symmetric Positive Definite Networks with Transformers for UAV-Human Action Recognition*

Mohamed Sanim Akremi, H. Tabia

RecognitionTransformerVideo

🎯 What it does: Proposed and implemented the SPDAGG-TransNet network for human action recognition from a drone's perspective

SpectralWaste Dataset: Multimodal Data for Waste Sorting Automation

Sara Casao, A. C. Murillo

SegmentationComputational EfficiencyData-Centric LearningImageMultimodalityBenchmark

🎯 What it does: Collected and provided a multimodal dataset named SpectralWaste, containing synchronized hyperspectral and RGB images, investigated the application of multimodal perception in waste classification, and evaluated different object segmentation architectures.

Speeding up 6-DoF Grasp Sampling with Quality-Diversity

J. Huber, Stéphane Doncieux

Pose EstimationOptimizationRobotic Intelligence

🎯 What it does: By integrating the quality diversity (QD) algorithm with prior knowledge, accelerating the diverse sampling of 6-DoF grasping poses in simulation.

Speeding Up Path Planning via Reinforcement Learning in MCTS for Automated Parking

Xinlong Zheng, Donghao Xu

Autonomous DrivingReinforcement Learning

🎯 What it does: Combine reinforcement learning with Monte Carlo Tree Search (MCTS) for online path planning in autonomous parking tasks, and build a value estimator and policy generator by iteratively learning state value and optimal actions to accelerate the planning process

Spike-based high energy efficiency and accuracy tracker for Robot

Jinye Qu, Hong Qiao

Object TrackingComputational EfficiencyRobotic IntelligenceSpiking Neural NetworkVideo

🎯 What it does: Proposed a 'motion feature extractor' and an 'RGB-DVS fusion module,' converted the model into a lossless SNN version, conducted experiments on the VOT2016 dataset, and deployed the model on a robot for tracking experiments.

SPVSoAP3D: A Second-order Average Pooling Approach to enhance 3D Place Recognition in Horticultural Environments

T. Barros, U. Nunes

RecognitionPoint CloudAgriculture Related

🎯 What it does: Proposed the SPVSoAP3D model and added two new horticultural environment sequences to the HORTO-3DLM dataset

SR-LIO: LiDAR-Inertial Odometry with Sweep Reconstruction

Zikang Yuan, Xin Yang

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes the SR-LIO LiDAR-Inertial Odometry method and utilizes scan reconstruction technology to enhance the frequency and accuracy of state estimation.

SSAP: A Shape-Sensitive Adversarial Patch for Comprehensive Disruption of Monocular Depth Estimation in Autonomous Navigation Applications

Amira Guesmi, Muhammad Shafique

Depth EstimationAutonomous DrivingAdversarial AttackConvolutional Neural NetworkTransformerImage

🎯 What it does: Proposed a shape-sensitive adversarial patch (SSAP) to comprehensively degrade the performance of monocular depth estimation (MDE) models in autonomous driving and robot navigation.

SSCBench: A Large-Scale 3D Semantic Scene Completion Benchmark for Autonomous Driving

Yiming Li, Chen Feng

Autonomous DrivingImagePoint CloudBenchmark

🎯 What it does: Built a large-scale 3D semantic scene completion benchmark named SSCBench, integrating automotive datasets such as KITTI-360, nuScenes, and Waymo, and unified semantic labels to facilitate exploration of SSC methods in street view scenarios.

SSL-RGB2IR: Semi-supervised RGB-to-IR Image-to-Image Translation for Enhancing Visual Task Training in Semantic Segmentation and Object Detection

Aniruddh Sikdar, Suresh Sundaram

Image TranslationObject DetectionSegmentationData SynthesisGenerative Adversarial NetworkImageMultimodality

🎯 What it does: Proposed and implemented the SSL-RGB2IR semi-supervised RGB-to-IR image-to-image translation model for generating synthetic IR images from RGB images.

Stability of Tethered Ground Robots on Extreme Terrains

Rahul Kumar, Sze Zheng Yong

Robotic Intelligence

🎯 What it does: Proposed a systematic algorithm to check the stability of two robots connected by cables on extreme terrain under a given path, considering the actual generation of cable tension and support constraints.

Stable Object Placing using Curl and Diff Features of Vision-based Tactile Sensors

Kuniyuki Takahashi, Tadahiro Taniguchi

Robotic IntelligenceImage

🎯 What it does: Propose an algorithm that estimates the object's corrective rotation direction and achieves stable placement by calculating Curl and Diff features through the displacement of black dots using the GelSight disparity tactile sensor.

Stable Wheel Gait Generation for Planar X-shaped Walker with Telescopic Legs Based on Asymmetric Impact Posture

Fumihiko Asano, Yanqiu Zheng

Robotic IntelligencePhysics RelatedOrdinary Differential Equation

🎯 What it does: This paper designs a control method to generate a stable wheeled gait on the horizontal plane for a planar X-shaped walker with extendable legs, avoiding the unstable and hard-to-control zero dynamics problem.

StaccaToe: A Single-Leg Robot that Mimics the Human Leg and Toe

Nisal Perera, Donghyun Kim

Robotic Intelligence

🎯 What it does: Designed and fabricated a human-scale single-legged robot named StaccaToe, experimentally validating its performance in balance and explosive ground reaction forces, including achieving balance in a pointed-toe stance and dynamic jumping.

STAIR: Semantic-Targeted Active Implicit Reconstruction

Liren Jin, Marija Popovi'c

GenerationDepth EstimationNeural Radiance FieldImagePoint CloudMesh

🎯 What it does: Proposes a semantic target-based active implicit reconstruction framework that utilizes pose RGB-D measurements and 2D semantic labels for object-level understanding and reconstruction

State Estimation of an Adaptive 3-Finger Gripper using Recurrent Neural Networks

Yannick Jonetzko, Jianwei Zhang

Robotic IntelligenceRecurrent Neural Network

🎯 What it does: Use two deep learning methods based on recurrent neural networks (RNN) to estimate the joint states of the Robotiq 3-Finger Adaptive Gripper

State Estimation Transformers for Agile Legged Locomotion

Chengxiang Yu, Diyun Xiang

Robotic IntelligenceTransformer

🎯 What it does: Propose a Transformer-based state estimation method that can accurately predict privileged states, which are difficult for robots to directly obtain, such as body height and velocity;

Static Modeling of the Stiffness and Contact Forces of Rolling Element Eccentric Drives for Use in Robotic Drive Systems

Simon Fritsch, Karsten Stahl

Robotic IntelligencePhysics Related

🎯 What it does: Developed a contact-based static model to calculate the stiffness and contact force of rolling element eccentric actuators under load, and derived the corresponding mathematical expressions; subsequently, compared the stiffness curves and torque-torque curves of different manufacturing processes (tolerance levels, material selection) and 3D printed plastic actuators.

Steering Decision Transformers via Temporal Difference Learning

Hao-Lun Hsu, Miroslav Pajic

TransformerReinforcement Learning

🎯 What it does: Propose an improved Decision Transformer (DT) that uses temporal difference learning to estimate expected returns, guiding the DT towards high-reward regions and addressing the issue of increasing variance in cumulative returns in random environments.

Stein Movement Primitives for Adaptive Multi-Modal Trajectory Generation

Zeya Yin, Fabio Ramos

Robotic IntelligenceMultimodality

🎯 What it does: Proposed and implemented Stein Motion Primitives (SMPs), adapting robot motion primitives and capturing multi-modal features in human demonstrations through non-parametric probabilistic inference using Stein Variational Gradient Descent (SVGD).

StereoNavNet: Learning to Navigate using Stereo Cameras with Auxiliary Occupancy Voxels

Hongyu Li, Huaizu Jiang

Autonomous DrivingRepresentation LearningSimultaneous Localization and MappingImage

🎯 What it does: Propose StereoNavNet, which utilizes stereo RGB images to estimate an auxiliary 3D voxel occupancy grid and extract geometric features for visual navigation.

Stick Roller: Precise In-hand Stick Rolling with a Sample-Efficient Tactile Model

Yipai Du, Yu She

Data-Centric LearningRobotic Intelligence

🎯 What it does: Learning High-Resolution Tactile Dynamics for Precise In-Hand Rod Rolling

STL-SLAM: A Structured-Constrained RGB-D SLAM Approach to Texture-Limited Environments

Juan Dong, Jie Chen

Simultaneous Localization and MappingImage

🎯 What it does: Proposed a structure-constrained RGB-D SLAM method (STL-SLAM) for texture-limited environments, which evaluates pixel distribution complexity through information entropy, detects the Manhattan framework, and separates rotation and translation when detected. It estimates drift-free rotation using the Manhattan world coordinate system and estimates translation by minimizing reprojection errors of points, lines, and planes. For non-Manhattan frameworks, it performs overall 6-DoF pose estimation with structural constraints including parallel and perpendicular planes and lines.

Strain-based Modeling of Rod-driven Soft Continuum Robots with Co-located Embedded Sensors

Peiyi Wang, Cecilia Laschi

Robotic Intelligence

🎯 What it does: Developed a dynamic and static mechanics model for a rod-driven soft continuum robot (RDSR) based on the geometric variable strain (GVS) method, capable of estimating the robot's shape and predicting strain changes under internal and external interactions.

StratXplore: Strategic Novelty-seeking and Instruction-aligned Exploration for Vision and Language Navigation

Muraleekrishna Gopinathan, Martin Masek

Vision-Language-Action ModelMultimodality

🎯 What it does: Propose a memory-based and error-aware path planning strategy called StratXplore for vision-language navigation, which selects unvisited frontier perspectives that match the instruction and are novel by collecting all actions and perspective features during navigation, achieving local and global decision-making to correct path errors.

Streamlining Forest Wildfire Surveillance: AI-Enhanced UAVs Utilizing the FLAME Aerial Video Dataset for Lightweight and Efficient Monitoring

Lemeng Zhao, S. Koshimura

CompressionComputational EfficiencyReinforcement LearningVideoAgriculture Related

🎯 What it does: Designed a lightweight and efficient UAV video understanding method, using a policy network to identify and compress redundant frames in videos, while introducing the concept of station points to utilize future information and improve accuracy.

Streamlining Object Pushing: Behavior Tree-Based Coordination of Control and Planning

Filippo Bertoncelli, Lorenzo Sabattini

Robotic Intelligence

🎯 What it does: Optimized object pushing tasks in complex environments by integrating behavior trees with control and planning frameworks.

Strong Compliant Grasps Using a Cable-Driven Soft Gripper

Gregory Xie, Daniela Rus

ClassificationRobotic Intelligence

🎯 What it does: Proposed and designed a soft gripper called FROG that combines compliance and strength, describing its mechanical structure, flexible component characteristics, grasping force analysis, and developing a feedforward grasping controller and grasping type classifier based on the gripper structure. Subsequently, the performance was verified through grasping experiments and holding force tests.

Structural Optimization of Lightweight Bipedal Robot via SERL

Yi Cheng, Hang Liu

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Designed a lightweight bipedal robot and proposed the SERL (Structural Evolutionary Reinforcement Learning) algorithm

Structure-Invariant Range-Visual-Inertial Odometry

Ivan Alberico, Davide Scaramuzza

Simultaneous Localization and MappingImage

🎯 What it does: Proposed a range-visual-inertial odometry system for Mars' irregular terrain, fusing consistent range information with visual and inertial measurements to avoid scale drift and support landing on arbitrary terrain structures.

Subtle-Diff: A Dataset for Precise Recognition of Subtle Differences Among Visually Similar Objects

F. Matsuzawa, Yutaka Satoh

RecognitionData SynthesisLarge Language ModelVision Language ModelImage

🎯 What it does: Developed the Subtle-Diff dataset and introduced two novel tasks based on visual differences: image selection and conditional difference description; evaluated existing vision-language models on this dataset and proposed a new model based on image-text similarity

SuFIA: Language-Guided Augmented Dexterity for Robotic Surgical Assistants

M. Moghani, Animesh Garg

Safty and PrivacyRobotic IntelligenceTransformerLarge Language ModelText

🎯 What it does: Propose the SuFIA framework, leveraging large language models (LLMs) and perception modules to achieve natural language-guided enhanced manipulability, completing surgical subtasks through high-level planning and low-level control.

SuPerPM: A Surgical Perception Framework Based on Deep Point Matching Learned from Physical Constrained Simulation Data

Sha Lin, Michael C. Yip

Object TrackingData SynthesisPoint Cloud

🎯 What it does: Proposes the SuPerPM framework, which utilizes learning-based non-rigid point cloud matching for data association to reduce misalignment errors in endoscopic soft tissue tracking; physically constrained correspondences are generated as training data through position-based dynamics (PBD) simulations.

Supervised Articulation Angles Estimation for Multi-Articulated Vehicles Based on Panoramic Camera System

Weimin Liu, Zhaocong Sun

Pose EstimationImage

🎯 What it does: Proposes a supervised multi-link vehicle joint angle estimation method based on a panoramic camera system, constructing a neural network with surrounding environment images captured by adjacent cameras as input, while considering time dependence and data imbalance issues.

SURESTEP: An Uncertainty-Aware Trajectory Optimization Framework to Enhance Visual Tool Tracking for Robust Surgical Automation

N. Shinde, Michael C. Yip

OptimizationRobotic IntelligenceBiomedical Data

🎯 What it does: Developed a trajectory optimization framework called SURESTEP that considers uncertainty to enhance the robustness of visual tool tracking in surgical automation, and applied it to a needle regrasping task under a moving endoscopic camera.

SurrealDriver: Designing LLM-powered Generative Driver Agent Framework based on Human Drivers’ Driving-thinking Data

Ye Jin, Jiangtao Gong

Autonomous DrivingTransformerLarge Language ModelTextChain-of-Thought

🎯 What it does: Proposes an LLM-driven generative driving agent framework based on human driver driving thought data, collects high-quality natural language data through urban driving experiments, and validates its effectiveness in simulation and human evaluation.

SwarmPRM: Probabilistic Roadmap Motion Planning for Large-Scale Swarm Robotic Systems

Yunze Hu, Chang Liu

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: Proposes a hierarchical, scalable, computationally efficient, and risk-aware sample-based motion planning method for swarm robots called SwarmPRM.

SWCF-Net: Similarity-weighted Convolution and Local-global Fusion for Efficient Large-scale Point Cloud Semantic Segmentation

Zhenchao Lin, Hong Zhang

SegmentationConvolutional Neural NetworkTransformerPoint Cloud

🎯 What it does: Propose a network called SWCF-Net, which combines Similarity-Weighted Convolution and local-global Fusion for efficient semantic segmentation of large-scale point clouds.

SWIFT: Strategic Weather-informed Image-based Forecasting for Trajectories

Youya Xia, Mark E. Campbell

Autonomous DrivingGraph Neural NetworkImage

🎯 What it does: Proposed a graph-based trajectory prediction model that uses only images to model the environment without requiring expensive map information.