arXivSub Start free trial

ICRA 2025 Papers — Page 8

IEEE International Conference on Robotics and Automation · 1604 papers

ICRT: In-Context Imitation Learning via Next-Token Prediction

Letian Fu, Kenneth Y. Goldberg

Robotic IntelligenceTransformerSequential

🎯 What it does: Propose ICRT, a causal transformer that uses autoregressive prediction for sensory motion trajectories, enabling scenario imitation learning without training and allowing flexible task execution in new environments.

iKap: Kinematics-Aware Planning with Imperative Learning

Qihang Li, Chen Wang

OptimizationRobotic Intelligence

🎯 What it does: Proposed and implemented a visual-planning system called iKap, which directly integrates the robot's kinematic model into the learning pipeline, achieving self-supervised learning and learning collision-safe and kinematically feasible trajectories through differentiable bilevel optimization.

Illumination Adaptation for SAM to Achieve Accurate Segmentation of Images Taken in Low-Light Scenes

Hongmin Mu, Zhengcai Cao

SegmentationDomain AdaptationTransformerImage

🎯 What it does: An adaptive method for the Segment Anything Model (SAM) is proposed to address low-light scenarios, incorporating self-training, low-light feature enhancement head, and domain shift compensation loss to achieve more accurate image segmentation.

Image-Based Compliance Control for Robotic Steering of a Ferromagnetic Guidewire

An Hu, Yu Sun

Robotic IntelligenceImageBiomedical Data

🎯 What it does: This paper utilizes only 2D perspective images as feedback, proposing a model-based external force observer to perceive unknown interactions between the guidewire and vascular wall, and designs a compliant controller to achieve safe and stable guidance of magnetic guidewires;

Image-Guided Surgical Planning for Percutaneous Nephrolithotomy Using CTRs: A Phantom-Based Study

Filipe C. Pedrosa, J. Jayender

Robotic IntelligenceBiomedical DataComputed Tomography

🎯 What it does: The optimal planning algorithm based on patient-specific coaxial tube robots (CTRs) for PCNL surgery planning was validated on a realistic right thoracic model.

Imitation Learning with Limited Actions via Diffusion Planners and Deep Koopman Controllers

Jianxin Bi, Harold Soh

Representation LearningRobotic IntelligenceDiffusion modelSequential

🎯 What it does: Proposes a plan-then-control framework that learns latent action representations in a Deep Koopman dynamics model using observational demonstration data, and maps them to high-dimensional continuous actions through a linear action decoder, significantly reducing the demand for labeled action data.

IMOST: Incremental Memory Mechanism with Online Self-Supervision for Continual Traversability Learning

Kehui Ma, Ling Pei

Autonomous DrivingComputational EfficiencyRepresentation Learning

🎯 What it does: Designed and implemented the IMOST framework, leveraging incremental dynamic memory and online self-supervised annotation to achieve continuous traversability learning.

Impedance Primitive-Augmented Hierarchical Reinforcement Learning for Sequential Tasks

Amin Berjaoui Tahmaz, J. Kober

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposes a hierarchical reinforcement learning framework enhanced with impedance primitives, capable of efficiently performing mechanical operations in continuous contact tasks;

Implicit Articulated Robot Morphology Modeling with Configuration Space Neural Signed Distance Functions

Yiting Chen, A. Billard

Computational EfficiencyRobotic Intelligence

🎯 What it does: Proposes an implicit robot morphology modeling method based on the configuration space signature distance function (Robot Neural Distance Function, RNDF), utilizing forward kinematics to achieve precise encoding and optimize the computational efficiency and accuracy of distance queries.

Implicit Contact Diffuser: Sequential Contact Reasoning With Latent Point Cloud Diffusion

Zixuan Huang, Dmitry Berenson

Robotic IntelligenceDiffusion modelPoint Cloud

🎯 What it does: Propose Implicit Contact Diffuser (ICD), a diffusion-based model that generates a series of neural descriptors specifying the sequence of contact relationships between objects and the environment, and uses this sequence as guidance for Model Predictive Control (MPC) methods to achieve long-horizon manipulation tasks with rich contact interactions.

Implicit Physics-aware Policy for Dynamic Manipulation of Rigid Objects via Soft Body Tools

Zixing Wang, A. H. Qureshi

Robotic IntelligenceWorld ModelPhysics Related

🎯 What it does: Proposed and verified an implicit physical awareness (IPA) strategy for dynamically moving rigid objects using flexible tools (e.g., ropes) in unknown environmental physical parameters with a single attempt;

Impossibility of Self-Organized Aggregation Without Computation

Roy Steinberg, Kiril Solovey

Robotic Intelligence

🎯 What it does: Prove that in robot systems without computational capability, it is impossible to achieve aggregation of any number of robots with any controller, and propose an alternative controller along with a rigorous proof of aggregation.

Improved Bag-of-Words Image Retrieval with Geometric Constraints for Ground Texture Localization

Aaron Wilhelm, Nils Napp

RetrievalSimultaneous Localization and MappingImage

🎯 What it does: Proposed an improved bag-of-words (BoW) based image retrieval system for ground texture localization, achieving higher global localization accuracy and improved loop closure detection precision and recall in SLAM, with two versions providing high precision and high speed respectively

Improving Coverage Performance of a Size-Reconfigurable Robot Based on Overlapping and Reconfiguration Reduction Criteria

M. V. J. Muthugala, M. R. Elara

OptimizationRobotic Intelligence

🎯 What it does: Propose a size-reconfigurable robot coverage path planning method based on overlap and reconfiguration reduction criteria

Improving Efficiency in Path Planning: Tangent Line Decomposition Algorithm

Yu Tian, Hongliang Ren

Autonomous DrivingOptimization

🎯 What it does: Propose a tangent decomposition algorithm (TLD) for efficiently finding near-optimal collision-free paths in 2D and 3D environments

Improving Generalization Ability for 3D Object Detection by Learning Sparsity-Invariant Features

Hsin-Cheng Lu, Winston H. Hsu

Object DetectionDomain AdaptationKnowledge DistillationPoint Cloud

🎯 What it does: This paper proposes a 3D object detection method for a single source domain, aiming to enhance the model's generalization capability in target domains with different sensor configurations and scene distributions.

Improving Grip Stability Using Passive Compliant Microspine Arrays for Soft Robots in Unstructured Terrain

Lauren Ervin, V. Vikas

Robotic Intelligence

🎯 What it does: Designed and implemented a passive compliant micro-ridge stacked array for soft robot legs to enhance grasping and walking performance on irregular terrain.

Improving Indoor Localization Accuracy by Using an Efficient Implicit Neural Map Representation

Haofei Kuang, C. Stachniss

Pose EstimationRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Propose an implicit neural map representation to capture position and orientation geometric features from 2D LiDAR scans, and combine a lightweight neural network with a traditional Monte Carlo localization framework to design an efficient observation model, achieving real-time robot pose estimation.

Improving Monocular Visual-Inertial Initialization with Structureless Visual-Inertial Bundle Adjustment

Junlin Song, M. Olivares-Méndez

Pose EstimationOptimizationSimultaneous Localization and MappingMultimodality

🎯 What it does: Propose a structureless visual-inertial bundle adjustment method to improve the initialization process of monocular visual-inertial odometry (VIO).

Improving Probe Localization for Freehand 3D Ultrasound Using Lightweight Cameras

Dianye Huang, Zhongliang Jiang

Pose EstimationDomain AdaptationConvolutional Neural NetworkAuto EncoderContrastive LearningImageBiomedical DataUltrasound

🎯 What it does: Propose a cost-effective and scalable method for handheld 3D ultrasound probe pose localization using two lightweight cameras.

Improving Vision-Language-Action Model with Online Reinforcement Learning

Yanjiang Guo, Jianyu Chen

Computational EfficiencySupervised Fine-TuningReinforcement LearningVision-Language-Action Model

🎯 What it does: Propose the iRe-VLA framework, which iteratively improves vision-language-action (VLA) models through alternating reinforcement learning and supervised learning.

Improving Zero-Shot ObjectNav with Generative Communication

Vishnu Sashank Dorbala, Dinesh Manocha

Robotic IntelligencePrompt EngineeringVision Language ModelMultimodality

🎯 What it does: Propose a method to enhance navigation performance in zero-shot ObjectNav by leveraging generative communication (GC), utilizing a vision-language model (VLM) for information exchange between a global-perspective upper auxiliary agent and a limited-perspective ground agent.

IMRL: Integrating Visual, Physical, Temporal, and Geometric Representations for Enhanced Food Acquisition

Rui Liu, Pratap Tokekar

Representation LearningRobotic IntelligenceMultimodality

🎯 What it does: Propose the IMRL method, integrating visual, physical, temporal, and geometric representations for imitation learning strategies in food acquisition, capturing food types, physical properties, action temporal dynamics, and geometric information to adapt scooping strategies under different scenarios.

In the Wild Ungraspable Object Picking with Bimanual Nonprehensile Manipulation

Albert Wu, Dan Kruse

Robotic IntelligenceImage

🎯 What it does: Utilizing dual-arm non-grasping operations to identify grasp points from visual information on compact shelves, gently pushing away obstacles when necessary, and subsequently using side-contact bimanual grasping to pick up objects that cannot be handled by traditional grippers.

In-Context Learning Enables Robot Action Prediction in LLMs

Yida Yin, Roei Herzig

Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: Propose the RoboPrompt framework, enabling text-based LLMs to directly predict robot actions through ICL without training.

In-Pipe Navigation Development Environment and a Smooth Path Planning Method on Pipeline Surface

Hao Liu, Mingquan Lu

OptimizationRobotic Intelligence

🎯 What it does: Proposed an open-source pipeline internal navigation development environment and designed a smooth path planning method based on the pipeline axis, subsequently verifying its usability in simulation and real environments.

In-Plane Manipulation of Soft Micro-Fiber with Ultrasonic Transducer Array and Microscope

Jie Zou, Song Liu

Robotic IntelligenceUltrasound

🎯 What it does: An automated ultrasonic manipulation system was constructed for non-contact manipulation of soft microfibers in-plane, along with the design of a real-time capture generation algorithm and theoretical analysis.

In-Vivo Cable-Driven Rodent Ankle Exoskeleton System for Sensorimotor Rehabilitation

Juwan Han, Keehoon Kim

Robotic IntelligenceBiomedical Data

🎯 What it does: Introduced a cable-driven ankle exoskeleton system for in vivo studies, experimentally evaluating its effects on gait and sensory-motor recovery under anesthetized and awake conditions.

Incremental Few-Shot Adaptation for Non-Prehensile Object Manipulation Using Parallelizable Physics Simulators

Fabian Baumeister, Joerg Stueckler

OptimizationRobotic IntelligenceMeta LearningPhysics Related

🎯 What it does: Proposed an incremental few-shot adaptation method that utilizes parallelizable rigid-body physics simulation and sampling optimization to calibrate the dynamics model for manipulating non-grasped objects.

Individual and Collective Behaviors in Soft Robot Worms Inspired by Living Worm Blobs

Carina Kaeser, Justin Werfel

Robotic Intelligence

🎯 What it does: Designed and demonstrated a set of pneumatic soft robots inspired by California black ants, studied their motion behavior and physical entanglement strength in individual and collective states, and compared them with live ants.

Indoor and Outdoor Multi-Terrain Stair-Climbing Robot Design

Wei-Ting Chen, Pei-Chun Lin

Robotic Intelligence

🎯 What it does: Proposed an indoor and outdoor multi-terrain stair-climbing robot named IOMT, and designed key functions such as four-wheel independent drive and steering, and control of stable pitch angle.

Indoor Localization of UAVs Using Only Few Measurements by Output-Sensitive Preimage Intersection

Michael M. Bilevich, Dan Halperin

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Propose a deterministic indoor UAV positioning method using only a small number of downward distance measurements and corresponding odometry, achieved through preimage intersection and spatial subdivision search.

Inducing Matrix Sparsity Bias for Improved Dynamic Identification of Parallel Kinematic Manipulators using Deep Learning

M. Lahoud, Ferdinando Cannella

Computational EfficiencyRobotic IntelligencePhysics Related

🎯 What it does: Proposed and validated a physics-informed neural network (PINN)-based dynamic model for the Delta parallel manipulator ABB IRB 360-6/1600, and introduced a sparsity constraint on the mass matrix;

Inference Based Multi-Object Reactive Search in a Partially Known Environment With Temporal Logic Specifications

Yaohui Kang, Zheng Kan

Robotic IntelligenceLarge Language Model

🎯 What it does: Proposes a reasoning-based multi-object reactive search framework that infers co-occurrence values between target objects and known landmarks using the COMET model, integrating the inferred results into LTL-constrained reactive temporal logic motion planning to enable dynamic multi-object search by robots in partially known environments.

Inference-Time Policy Steering Through Human Interactions

Yanwei Wang, Julie Shah

Reinforcement Learning from Human FeedbackDiffusion model

🎯 What it does: Propose an ITPS framework that guides generative strategies through human interaction during the inference phase.

Infield Self-Calibration of Intrinsic Parameters for Two Rigidly Connected IMUs

Can Huang, Kejian J. Wu

OptimizationRobotic Intelligence

🎯 What it does: Studied the intrinsic parameter self-calibration of two rigidly connected IMUs using only IMU data and known extrinsic parameters, focusing on observability analysis. It proved that gyroscope intrinsic parameters and partial accelerometer biases are observable, while identifying unobservable directions caused by degenerate motions. Numerical simulations verified the observability, and real data were used to evaluate calibration accuracy.

Informed Repurposing of Quadruped Legs for New Tasks

Fuchen Chen, Daniel M. Aukes

Robotic Intelligence

🎯 What it does: Studies how to evaluate and repurpose existing quadruped legs for new tasks, implementing the method on 15 robot designs generated from six preselected leg design combinations.

InsCMPR: Efficient Cross-Modal Place Recognition via Instance-Aware Hybrid Mamba-Transformer

Shuaifeng Jiao, Xieyuanli Chen

RetrievalAutonomous DrivingTransformerImagePoint Cloud

🎯 What it does: Proposed InsCMPR, an instance-aware cross-modal pose recognition method, which generates descriptors through pixel-level and instance-level modality alignment as well as a dual-branch hybrid Mamba-Transformer network;

Instance Segmentation-Based Hazard Detection with Lunar South Pole Lighting

Joseph M. Cloud, Jason M. Schuler

SegmentationData SynthesisConvolutional Neural NetworkTransformerImagePhysics Related

🎯 What it does: Study on using instance segmentation models to detect rock hazards in the visual environment of the Moon's south pole

Integrated Motion State Prediction for Sit-to-Stand and Stand-to-Sit Motions Toward Effective Power Assist Control

Kai Ren, Qi An

Robotic IntelligenceRecurrent Neural NetworkBiomedical Data

🎯 What it does: Proposed a sensing method utilizing electrophysiological measurements and deep neural networks to predict the timing of sit-to-stand and stand-to-sit movement initiation, for controlling robotic assistive devices.

Integrating Active Sensing and Rearrangement Planning for Efficient Object Retrieval from Unknown, Confined, Cluttered Environments

Junyong Kim, A. H. Qureshi

Robotic IntelligenceImage

🎯 What it does: Proposed a retrieval planning framework that combines heuristic active perception with Monte-Carlo tree search (MCTS), utilizing a robotic arm with a handheld camera to retrieve target objects in unknown constrained, cluttered environments.

Integrating Field of View in Human-Aware Collaborative Planning

Ya-Chuan Hsu, S. Nikolaidis

Autonomous DrivingReinforcement Learning from Human FeedbackWorld Model

🎯 What it does: A hierarchical online planner is proposed for subtask intent adaptation in human-robot collaboration, considering the limited human field of view (FOV), and its effectiveness is validated through user studies and VR kitchen environments.

Integrating Human-Robot Teaming Dynamics Into Mission Planning Tools for Transparent Tactics in Multi-Robot Human Integrated Teams

Audrey L. Aldridge, M. Novitzky

🎯 What it does: In small-scale experiments, the integration of human-robot collaboration dynamics with the task planning tool (PETAAR) was studied to evaluate the coordination ability of robot operators in multi-robot tasks, comparing two collaboration modes: manually inserting waypoints for each robot and using PETAAR to plan all waypoints at once.

Integrating Learning-Based Manipulation and Physics-Based Locomotion for Whole-Body Badminton Robot Control

Haochen Wang, Dong Xuan

Robotic IntelligenceReinforcement LearningPhysics Related

🎯 What it does: Proposed a hybrid control system named Hamlet for controlling full-body doubles badminton robots;

Integrating Model-Based Control and RL for Sim2Real Transfer of Tight Insertion Policies

Isidoros Marougkas, Kostas E. Bekris

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed and verified a strategy integrating model-based control with reinforcement learning for tight plug-and-play tasks, trained in simulation and achieving zero-shot transfer to real systems.

Integrating Multi-Robot Adaptive Sampling and Informative Path Planning for Spatiotemporal Natural Environment Prediction

Siva Kailas, Katia P. Sycara

OptimizationRobotic IntelligenceTime Series

🎯 What it does: Study the application of multi-robot information path planning in adaptive sampling, using spatiotemporal mixed Gaussian processes (STMGP) to determine the most informative sampling locations, and submodular function optimization to plan sampling paths; implement a decentralized two-stage sampling process.

Intelligence Evaluation Methods for Autonomous Vehicles

Jun Zhou, Xiaofan Wang

Autonomous DrivingOptimizationAdversarial Attack

🎯 What it does: Propose a Robust Training-based Comprehensive Evaluation System (RTCE) for quantitatively assessing the temporal intelligence level of autonomous vehicles.

Intelligent Self-Healing Artificial Muscle: Mechanisms for Damage Detection and Autonomous Repair of Puncture Damage in Soft Robotics

Ethan J. Krings, Eric J. Markvicka

Robotic Intelligence

🎯 What it does: Proposed a soft structure embedding liquid metal microdroplets into a silicone rubber elastomer, enabling soft robots to automatically detect damage under high pressure or puncture and achieve self-healing and functional recovery through the formation of a conductive network.

IntelliRMS: A Robotic Manipulation System for Domain-Specific Tasks Using Vision and Language Foundational Models

Chandan Kumar Singh, Rajesh Sinha

Robotic IntelligenceVision Language ModelVision-Language-Action ModelMultimodality

🎯 What it does: Designed and implemented the IntelliRMS system, utilizing vision and language foundation models to achieve instruction-following robot control for domain-specific tasks such as industrial pick-and-place.

Interaction-Driven Updates: 3D Scene Graph Maintenance During Robot Task Execution

Qingfeng Li, Jianwei Niu

Robotic IntelligenceLarge Language ModelWorld ModelGraph

🎯 What it does: Proposes an interaction-driven 3D scene graph maintenance method that integrates an observation point selection module and a dynamic scene maintenance module to continuously update scene information as the robot performs tasks.

Interactive Motion Planning for a 7-DOF Robot

Nicholas Greene, P. Kazanzides

Robotic IntelligenceReinforcement Learning from Human FeedbackImage

🎯 What it does: Proposes an extended Interactive Planning and Supervised Execution (IPSE) system to achieve full remote control of a 7-degree-of-freedom (7-DOF) robot, encoding redundant degrees of freedom using a shoulder-elbow-wrist (SEW) angle map, and embedding robot state information as a 2D image within this angle map to enable interactive motion planning.

Interactive OT Gym: A Reinforcement Learning-Based Interactive Optical Tweezer (OT)-Driven Microrobotics Simulation Platform

Zongcai Tan, Dandan Zhang

Robotic IntelligenceReinforcement LearningPhysics Related

🎯 What it does: Designed an Interactive OT Gym simulation platform based on reinforcement learning, supporting collaborative manipulation by optical tweezers-driven micro-robots, and integrated physical field simulation, tactile feedback, and context-aware shared control strategies

Interactive4D: Interactive 4D LiDAR Segmentation

Ilya Fradlin, Bastian Leibe

SegmentationPoint Cloud

🎯 What it does: Proposes a novel paradigm for interactive 4D LiDAR segmentation, and realizes the first model named Interactive4D that can simultaneously segment multiple objects in overlapping consecutive LiDAR scans in a single iteration

Interface Matters: Comparing First and Third-Person Perspective Interfaces for Bi-Manual Robot Behavioural Cloning

Haining Luo, Y. Demiris

Robotic IntelligenceVideo

🎯 What it does: Investigated the effectiveness of first-person and third-person perspective interfaces in robot shoelace behavior cloning, and explored the impact of interface design on the quality of expert demonstrations.

Internal-Stably Energy-Saving Cooperative Control of Articulated Wheeled Robot with Distributed Drive Units

Yi Yang, Shanshan Xie

OptimizationRobotic Intelligence

🎯 What it does: Proposed a coordinated control algorithm based on equivalent and distribution of driving force for multi-driving units to improve the foldable tracked robot with distributed drive.

Interpretable Active Inference Gait Control Learning

Rudolf J. Szadkowski, J. Faigl

Explainability and InterpretabilityRobotic Intelligence

🎯 What it does: Implement adaptive gait control for hexapod robots in adversarial environments using a self-learning gait dynamics model and the Free Energy Principle (FEP) framework;

Intraoperative 3D Shape Estimation of Magnetic Soft Guidewire

Yiting Zhao, Nan Xiao

Image

🎯 What it does: Introduces a technique utilizing a flexible magnetic tip guidewire to achieve intraoperative three-dimensional shape reconstruction during vascular interventional surgery.

Intraoperative Trocar-Based Eyeball Rotation Estimation Using Only 2D Microscope Images

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

Pose EstimationImageBiomedical Data

🎯 What it does: By using only 2D microscope images, leveraging the eye's geometric model and the position information of the microscope puncture holes, the rotation of the eye along the x and y axes is calculated to achieve intraoperative eye pose estimation.

Introducing Collaborative Robots as a First Step Towards Autonomous Reprocessing of Medical Equipment

Florian Voigt, Sami Haddadin

Robotic Intelligence

🎯 What it does: This study proposes a framework based on compliant collaborative robots for automated handling and storage of endoscopes after disinfection, addressing the challenge of high-dexterity operations in endoscope reprocessing.

Introducing KUGE: A Simultaneous Control Co-Design Architecture and its Application to Aerial Robotics Development

J. Wauters, G. Crevecoeur

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: Proposed and validated a synchronous control co-design (CCD) architecture KUGE for the dynamic design of tail-sitters

Introspective Loop Closure for SLAM with 4D Imaging Radar

Maximilian Hilger, Achim J. Lilienthal

Autonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Studied how to utilize 4D imaging radar for loop closure detection in SLAM, focusing on similar and opposite viewpoints.

Inverse Kinematics on Guiding Vector Fields for Robot Path Following

Yu Zhou, Héctor García de Marina

Robotic Intelligence

🎯 What it does: This paper applies inverse kinematics to a guidance vector field to achieve path tracking for autonomous mobile robots. It constructs an error signal using zero-level sets, enabling the robot to converge and move along the path under the guidance of the vector field. Furthermore, inverse kinematics is utilized to ensure the error signal forms a linear system, thereby enabling control of the robot's short-term motion, with a feedforward signal injected to precisely adjust motion behavior along the path. A theoretical and practical solution is proposed for achieving precise short-term control of a constant-speed unicycle on a 2D path. Finally, the theoretical results are validated through flight experiments with a fixed-wing unmanned aerial vehicle.

Inverse Mixed Strategy Games with Generative Trajectory Models

Muchen Sun, Todd Murphey

OptimizationReinforcement LearningAuto EncoderBenchmark

🎯 What it does: Propose an inverse game method that integrates a trajectory generation model with a differentiable hybrid strategy game framework, using CVAE to represent hybrid strategies, inferring high-dimensional multimodal behavior distributions from noisy measurements and adapting in real-time to new observations.

Investigating Security Threats in Multi-Tenant ROS 2 Systems

Lichen Xia, Weisong Shi

Safty and PrivacyRobotic Intelligence

🎯 What it does: In-depth study of security threats in a multi-tenant ROS 2 system, focusing on analyzing vulnerabilities in ROS nodes and topics, designing attack strategies to bypass isolation and security mechanisms, validating attack effectiveness through simulation, and proposing defensive practices.

IRef-VLA: A Benchmark for Interactive Referential Grounding with Imperfect Language in 3D Scenes

Haochen Zhang, Wenshan Wang

Graph Neural NetworkVision Language ModelPoint CloudGraphBenchmark

🎯 What it does: Constructed and released the IRef-VLA dataset for interactive referential localization tasks, containing natural language instructions and navigation goals in 3D scenes.

IROAM: Improving Roadside Monocular 3D Object Detection Learning from Autonomous Vehicle Data Domain

Zhe Wang, Yan Wang

Domain AdaptationAutonomous DrivingTransformerContrastive Learning

🎯 What it does: Propose the IROAM framework, leveraging data from vehicle-mounted and roadside units to enhance monocular 3D object detection at the roadside.

Is Discretization Fusion All You Need for Collaborative Perception?

Kang Yang, Deying Li

Object DetectionAutonomous DrivingVideoPoint Cloud

🎯 What it does: Proposed a anchor-based collaborative perception framework named ACCO for target detection in multi-agent systems; the framework includes anchor feature block (AFB), anchor confidence generator (ACG), and local-global fusion modules (LAAF and SACA)

Is Iteration Worth It? Revisit its Impact in Sliding-Window VIO

Chuchu Chen, Guoquan Huang

Pose EstimationOptimizationSimultaneous Localization and Mapping

🎯 What it does: The first comprehensive study on iterative algorithms in sliding window visual inertial odometry (VIO) was conducted, separately analyzing the impact of relinearization of IMU and camera measurements on system performance.

Is Linear Feedback on Smoothed Dynamics Sufficient for Stabilizing Contact-Rich Plans?

Yuki Shirai, Tao Pang

Robotic IntelligenceOrdinary Differential Equation

🎯 What it does: Analyzes the effectiveness of linear controller synthesis based on contact smoothing in contact-rich operations, and verifies its performance in whole-body experiments with robotic hands.

Iterative Volume Fusion for Asymmetric Stereo Matching

Yuanting Gao, Linghao Shen

Depth EstimationImageBenchmark

🎯 What it does: Propose a two-stage iterative volume fusion network, IVF-AStereo, for addressing visual asymmetry in stereo matching.

Jailbreaking LLM-Controlled Robots

Alexander Robey, George Pappas

Adversarial AttackRobotic IntelligenceTransformerLarge Language ModelText

🎯 What it does: Proposed and implemented the ROBOPAIR algorithm to attack and disable robots controlled by large language models (LLMs), enabling them to perform harmful physical actions.

Joint 3D Point Cloud Segmentation Using Real-Sim Loop: From Panels to Trees and Branches

Tian Qiu, Yu Jiang

SegmentationData SynthesisPoint Cloud

🎯 What it does: Proposes a training data generation method based on Real2Sim L-TreeGen and designs a joint model J-P2TB for joint point cloud segmentation from panels to trees and branches.

Joint Localization and Planning Using Diffusion

L. L. Beyer, S. Karaman

Autonomous DrivingDiffusion modelPoint Cloud

🎯 What it does: Propose an end-to-end localization and path planning framework based on diffusion models, which can generate globally collision-free paths in any known 2D environment by combining self-observed LiDAR scans and target positions.

Jointly Assigning Processes to Machines and Generating Plans for Autonomous Mobile Robots in a Smart Factory

Christopher Leet, Sven Koenig

OptimizationRobotic Intelligence

🎯 What it does: Proposed ACES, which first jointly optimizes the allocation of processes and machines as well as mobile robot path planning.

JORD: A Benchmark Dataset for Off-Road LiDAR Place Recognition and SLAM

Wei Zhou, Gang Wang

Simultaneous Localization and MappingPoint CloudBenchmark

🎯 What it does: Proposed and released the first benchmark dataset JORD specifically designed for airborne LiDAR SLAM, and conducted benchmark testing with multiple advanced methods.

JPG-SLAM: Joint Point-Gaussian Splatting Representation for Dense Dynamic SLAM

Kunrui Huang, Jian Yao

Gaussian SplattingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a dense dynamic SLAM system based on Joint Point-Gaussian Splatting, enabling precise pose estimation and dynamic scene reconstruction.

JRN-Geo: A Joint Perception Network Based on RGB and Normal Images for Cross-View Geo-Localization

Hongyu Zhou, Yizhong Zhang

Data SynthesisRetrievalConvolutional Neural NetworkMultimodality

🎯 What it does: This paper proposes a joint perception network called JRN-Geo based on RGB and normal maps, which achieves deep fusion through a dual-branch feature extraction framework combined with a Difference-Aware Fusion Module (DAFM) and a Joint-Constrained Interaction Aggregation (JCIA). It enhances the robustness of cross-perspective geolocation by generating perspective variation samples using 3D geographic augmentation technology.

Juzu Type Gripper That Can Change Both Shape and Firmness

Shunya Hara, Mitsuru Higashimori

Robotic Intelligence

🎯 What it does: Designed and developed a Juzu-type gripper capable of actively altering the shape and stiffness of its fingers, and experimentally validated its effectiveness in pre-shaping and grasping objects of different shapes and sizes.

KALIE: Fine-Tuning Vision-Language Models for Open-World Manipulation Without Robot Data

G. Tang, Kuan Fang

Data SynthesisRobotic IntelligenceSupervised Fine-TuningVision Language ModelImageText

🎯 What it does: KALIE fine-tunes pre-trained vision-language models to predict point-basis affordance representations using natural language instructions and visual observations, enabling robots to perform open-world manipulation tasks.

KALM: Keypoint Abstraction Using Large Models for Object-Relative Imitation Learning

Xiaolin Fang, L. Kaelbling

Large Language ModelVision Language ModelMultimodality

🎯 What it does: Leverage large-scale pre-trained vision-language models to automatically generate task-related and cross-instance consistent keypoints, and train keypoint-conditioned policy models based on these keypoints, enabling robots to generalize across different object poses, camera viewpoints, and object instances with similar functional shapes;

Kalman-Filter-Based Pose Estimation of Cable-Driven Parallel Robots Using Cable-Length Measurements with Colored Noise

Vinh Nguyen, Ryan J. Caverly

Pose Estimation

🎯 What it does: Proposes an Extended Kalman Filter (EKF) framework based on cable length for estimating the end-effector pose of cable-driven parallel robots.

KARMA: Augmenting Embodied AI Agents with Long-and-Short Term Memory Systems

Zixuan Wang, Yiming Gan

Robotic IntelligenceLarge Language ModelPrompt EngineeringRetrieval-Augmented Generation

🎯 What it does: Proposed the KARMA memory system, integrating long-term and short-term memory modules to enhance the planning capabilities of embodied AI agents when performing complex household tasks.

Key-Scan-Based Mobile Robot Navigation: Integrated Mapping, Planning, and Control Using Graphs of Scan Regions

Dharshan Bashkaran Latha, Ömür Arslan

Robotic IntelligenceSimultaneous Localization and MappingPoint CloudGraph

🎯 What it does: Propose a mobile robot navigation framework based on key scans, integrating mapping, planning, and control into a scan area map; verify its effectiveness in 2D cluttered environments through experiments.

Keypoint Detection and Description for Raw Bayer Images

Jiakai Lin, Guoyu Lu

Pose EstimationConvolutional Neural NetworkImage

🎯 What it does: Proposed a keypoint detection and description network that directly processes raw Bayer images, avoiding traditional ISP processing;

KFCalibNet: A KansFormer-Based Self-Calibration Network for Camera and LiDAR

Zejing Xu, Zhe Fu

Pose EstimationAutonomous DrivingOptimizationTransformerImageMultimodalityPoint Cloud

🎯 What it does: Proposed a self-calibration network called KFCalibNet based on KansFormer for extrinsic calibration between camera and LiDAR;

Kinematic-ICP: Enhancing LiDAR Odometry with Kinematic Constraints for Wheeled Mobile Robots Moving on Planar Surfaces

Tiziano Guadagnino, C. Stachniss

Autonomous DrivingOptimizationSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes a LiDAR odometry system that integrates vehicle kinematic constraints into traditional ICP optimization to improve localization accuracy for mobile robots on planar surfaces.

Kineto-Dynamical Planning and Accurate Execution of Minimum-Time Maneuvers on Three-Dimensional Circuits

Mattia Piccinini, F. Biral

Autonomous DrivingOptimization

🎯 What it does: Proposed an artificial racing driver (ARD) capable of online learning of vehicle dynamics, planning, and executing shortest-time actions on 3D tracks;

Kinodynamic Model Predictive Control for Energy Efficient Locomotion of Legged Robots with Parallel Elasticity

Yulun Zhuang, Yanran Ding

OptimizationRobotic Intelligence

🎯 What it does: Proposes a kinodynamic MPC framework that utilizes unidirectional parallel springs (UPS) to enhance the energy efficiency of dynamic legged robots, and adopts a hierarchical control structure: first using a simplified dynamic model MPC for warm-up, then employing full nonlinear centroid dynamics and kinematic constraints for kinodynamic MPC; reduces peak motor torque and energy consumption during the stance phase via UPS, achieving energy-efficient dynamic jumps.

KISS-Matcher: Fast and Robust Point Cloud Registration Revisited

Hyungtae Lim, L. Carlone

Pose EstimationComputational EfficiencyPoint Cloud

🎯 What it does: Developed an open-source C++ library called KISS-Matcher for point cloud registration, integrating a new feature detector Faster-PFH and a graph theory pruning algorithm based on k-core, forming a complete and user-friendly registration pipeline.

Knowledge-Driven Visual Target Navigation: Dual Graph Navigation

Shiyao Li, Feilong Wang

Autonomous DrivingComputational EfficiencyRepresentation LearningGraph Neural NetworkImageGraph

🎯 What it does: Propose a knowledge-driven, lightweight dual graph navigation framework (Dual Graph Navigation), achieving image instance navigation by constructing external and internal knowledge graphs;

Koopman Operator Based Linear Model Predictive Control for Quadruped Trotting

Chun-Ming Yang, Pranav A. Bhounsule

OptimizationRobotic Intelligence

🎯 What it does: Built a high-dimensional linear model based on the Koopman operator for quadruped robot gait control, and achieved high-fidelity tracking and disturbance suppression through linear model predictive control (LMPC).

KUDA: Keypoints to Unify Dynamics Learning and Visual Prompting for Open-Vocabulary Robotic Manipulation

Zixi Liu, Yunzhu Li

Robotic IntelligenceVision Language ModelImageText

🎯 What it does: Built an open-source lexical robot operating system named KUDA, integrating keypoint visual cues with dynamics learning to achieve automatic planning for diverse tasks;

LaB-CL: Localized and Balanced Contrastive Learning for Improving Parking Slot Detection

J. Jeong, I. Yong

Object DetectionData-Centric LearningContrastive Learning

🎯 What it does: Proposes a supervised contrastive learning framework named LaB-CL specifically designed for parking slot detection, aiming to address classification bias caused by data imbalance.

Lab2Car: A Versatile Wrapper for Deploying Experimental Planners in Complex Real-World Environments

Marc Heim, Momchil S. Tomov

Autonomous DrivingOptimization

🎯 What it does: Developed Lab2Car, an expandable wrapper capable of converting trajectory sketches generated by any motion planner into safe, comfortable, and dynamically feasible driving trajectories, enabling originally unsafe planners to be safely tested and optimized in real-world environments.

Label Anything: An Interpretable, High-Fidelity and Prompt-Free Annotator

Wei-Bin Kou, Yik-Chung Wu

SegmentationAutonomous DrivingOptimizationExplainability and InterpretabilityTransformerImage

🎯 What it does: Proposed a Label Anything Model (LAM) that leverages pre-trained Vision Transformer for feature extraction, and adds Semantic Class Adapter (SCA) and Optimization-oriented Unrolling Algorithm (OptOU) to achieve interpretable, high-fidelity, prompt-free data annotator.

LACNS: Language-Assisted Continuous Navigation in Structured Spaces

Rutong Peng, Mengyin Fu

Autonomous DrivingLarge Language ModelVision Language ModelImageTextMultimodality

🎯 What it does: Proposed the LACNS system, which generates BEV maps using vehicle front-facing cameras, detects intersections with visual language models (VLM), selects navigation frontiers with language models (LLM), achieving continuous space autonomous driving navigation without relying on HD maps

LAFNET: Lightweight Aerial Fire Detection Model for Onboard Edge Computing

Haozhou Zhai, Tianjiang Hu

Object DetectionComputational EfficiencyConvolutional Neural NetworkImage

🎯 What it does: Proposed and implemented a lightweight aerial fire detection model called LAFNET

LaMMA-P: Generalizable Multi-Agent Long-Horizon Task Allocation and Planning with LM-Driven PDDL Planner

Xiaopan Zhang, Jiachen Li

OptimizationLarge Language ModelAgentic AIMultimodalityBenchmark

🎯 What it does: Proposed LaMMA-P, a language model-driven multi-agent PDDL planning framework that combines LM reasoning with traditional heuristic search planning to address long-term tasks, and constructed the MAT-THOR benchmark for multi-level complexity home tasks based on AI2-THOR.

LaMOT: Language-Guided Multi-Object Tracking

Yunhao Li, Libo Zhang

Object TrackingVision Language ModelVideoBenchmark

🎯 What it does: Proposed a language-oriented unified task framework for multi-object tracking, created the LaMOT benchmark dataset, and introduced a concise and effective tracker called LaMOTer.

LamPro: Multi-Prototype Representation Learning for Enhanced Visual Pattern Recognition

Ji-rong Qi, Yang Wang

RecognitionRepresentation LearningContrastive LearningImage

🎯 What it does: Proposed the Label-aware multi-prototype learning method LamPro, which utilizes label information during prototype formation and updating to enhance the quality of visual pattern recognition representations.