arXivSub Start free trial

ICRA 2025 Papers — Page 9

IEEE International Conference on Robotics and Automation · 1604 papers

Language-Conditioned Offline RL for Multi-Robot Navigation

Steven D. Morad, Amanda Prorok

Robotic IntelligenceTransformerLarge Language ModelReinforcement Learning

🎯 What it does: Synthesize navigation strategies for multi-robot teams that can explain and follow natural language instructions.

Language-Guided Object Search in Agricultural Environments

Advaith Balaji, Dmitry Berenson

Robotic IntelligenceLarge Language ModelAgriculture Related

🎯 What it does: Proposed a language-guided target object search method in farm environments based on large language models (LLMs), which infers the location of unseen targets using the semantic relationships of known objects and plans paths accordingly.

Language-Guided Object-Centric Diffusion Policy for Generalizable and Collision-Aware Manipulation

Hang Li, Alois Knoll

OptimizationRobotic IntelligenceLarge Language ModelDiffusion modelPoint Cloud

🎯 What it does: Proposes a language-guided object-centric diffusion strategy framework called Lan-o3dp, which uses a 3D point cloud conditioned diffusion model to predict robotic arm end-effector trajectories, and achieves training-agnostic collision avoidance through cost optimization during inference.

Large Language Model Based Autonomous Task Planning for Abstract Commands

Seokjoon Kwon, D. Chang

Robotic IntelligenceTransformerLarge Language ModelAgentic AIVision Language ModelText

🎯 What it does: Proposes an autonomous task planning framework based on large language models for handling abstract natural language instructions, consisting of two stages: environment recognition and task planning.

Large-Expansion Bi-Layer Auxetics Create Compliant Cellular Motion

Lillian Chin, Daniela Rus

Robotic Intelligence

🎯 What it does: Proposed and implemented AuxSwarm, a scalable modular robot system composed of auxetic voxels with a dual-layer design, achieving a 1.57-fold size expansion within 0.2 seconds, and validated its versatility through 2D bending and 3D cube flipping experiments.

Large-Scale Gaussian Splatting SLAM

Zhe Xin, Guoquan Huang

Gaussian SplattingSimultaneous Localization and MappingImageVideo

🎯 What it does: Proposes a binocular visual SLAM system (LSG-SLAM) based on large-scale 3D Gaussian distribution (Gaussian Splatting), which estimates prior poses under large disparities through a multi-modal strategy, achieves infinite scene scalability via continuous Gaussian subgraphs, and optimizes relative poses through loop detection and rendering + feature deformation loss, ultimately enhancing reconstruction quality through global optimization and structural refinement.

Large-Scale UWB Anchor Calibration and One-Shot Localization Using Gaussian Process

Shenghai Yuan, Lihua Xie

Autonomous DrivingGaussian SplattingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes a UWB-LiDAR fusion-based single-time calibration and localization framework, which uses Gaussian processes to estimate anchor positions and significantly narrows the LiDAR loop closure search range through UWB range filtering, thereby improving localization accuracy and speed.

Latent Embedding Adaptation for Human Preference Alignment in Diffusion Planners

Wen Zheng Terence Ng, Tianwei Zhang

GenerationReinforcement Learning from Human FeedbackDiffusion model

🎯 What it does: Proposed a resource-efficient method that achieves trajectory personalization for individual user preferences by using Preference Latent Embeddings (PLE) in pre-trained conditional diffusion models and leveraging preference inversion for fast adaptation

LCSPose: Efficient, Accurate and Scalable Markerless 6-DoF Pose Estimation of a Quay Crane Spreader Based on LiDAR and Camera

Yichen Zhou, Danwei Wang

Pose EstimationImagePoint Cloud

🎯 What it does: Propose a markerless 6-DoF pose estimation method LCSPose based on LiDAR and camera fusion

LDM-ISP: Enhancing Neural ISP for Low Light with Latent Diffusion Models

Qiang Wen, Qifeng Chen

RestorationConvolutional Neural NetworkDiffusion modelImage

🎯 What it does: Proposed a method utilizing a pre-trained latent diffusion model for neural ISP processing of low-light RAW images

LE-Object: Language Embedded Object-Level Neural Radiance Fields for Open-Vocabulary Scene

Mengting Wang, Zhiteng Li

SegmentationGenerationRepresentation LearningVision Language ModelNeural Radiance FieldPoint Cloud

🎯 What it does: Propose LE-Object, an object-level neural implicit radiance field designed for open-world scenarios, to achieve fine-grained semantic segmentation and high-fidelity object reconstruction.

Leader-Follower Formation Enabled by Pressure Sensing in Free-Swimming Undulatory Robotic Fish

Kundan Panta, Bo Cheng

Robotic IntelligenceRecurrent Neural Network

🎯 What it does: A fish-shaped biomimetic robot leader-follower formation using flow differential pressure sensors for seamless swimming.

LeAP: Consistent multi-domain 3D labeling using Foundation Models

Simon Gebraad, Holger Caesar

SegmentationRepresentation LearningTransformerPoint Cloud

🎯 What it does: Propose Label Any Pointcloud (LeAP), which automatically performs consistent semantic annotation on multi-frame 3D point clouds using 2D Vision Foundation Models, aggregates point labels into voxels through Bayesian updates, and further improves label quality using a 3D Consistency Network.

Learn to Swim: Data-Driven LSTM Hydrodynamic Model for Quadruped Robot Gait Optimization

Fei Han, Dixia Fan

OptimizationRobotic IntelligenceRecurrent Neural NetworkTime SeriesSequential

🎯 What it does: Developed a fluid experiment data-driven model based on Long Short-Term Memory (LSTM) networks, named FEDLSTM, for predicting the non-steady nonlinear hydrodynamic forces of the underwater quadruped robot we constructed.

Learning a High-Quality Robotic Wiping Policy Using Systematic Reward Analysis and Visual-Language Model Based Curriculum

Yihong Liu, Sehoon Ha

Robotic IntelligenceReinforcement LearningVision Language ModelMultimodality

🎯 What it does: Proposes a reward analysis and curriculum learning method based on visual language models for high-quality robotic wiping

Learning Active Tactile Perception Through Belief-Space Control

J. Tremblay, Gregory Dudek

Robotic IntelligenceWorld Model

🎯 What it does: Propose a method for autonomously learning tactile exploration strategies through generative world models, differentiable Bayesian filtering, and information acquisition model predictive control (MPC), enabling estimation of unknown object properties via physical interaction.

Learning Adversarial Policies for Swarm Leader Identification Using a Probing Agent

Stergios E. Bachoumas, Panagiotis K. Artemiadis

Robotic IntelligenceReinforcement LearningAgentic AI

🎯 What it does: Train a probe agent to identify the leader of a multi-agent robot group through physical interactions.

Learning and Online Replication of Grasp Forces from Electromyography Signals for Prosthetic Finger Control

Robin Arbaud, Arash Ajoudani

Robotic IntelligenceBiomedical Data

🎯 What it does: A force-controlled prosthetic finger prototype based on electromyography (EMG) signals was developed, utilizing a neural network model to estimate fingertip force from EMG inputs for online adjustment of grip strength; the effectiveness of the force estimation model was validated on 10 subjects, and precise control was demonstrated in online experiments with 4 wearers;

Learning Based MPC for Autonomous Driving Using a Low Dimensional Residual Model

Yaoyu Li, Jun Li

Autonomous DrivingOptimization

🎯 What it does: Propose a learning-based model predictive control (MPC) that uses a low-dimensional residual model to improve the accuracy of autonomous vehicle dynamics models.

Learning Better Representations for Crowded Pedestrians in Offboard LiDAR-Camera 3D Tracking-by-detection

Shichao Li, Xiaozhi Chen

Object TrackingRepresentation LearningImageMultimodalityPoint CloudBenchmark

🎯 What it does: This study addresses the problem of pedestrian 3D tracking in highly crowded urban environments by constructing an offline automatic annotation system and proposing a high-resolution, density-aware, and relation-aware representation learning method;

Learning Coordinated Bimanual Manipulation Policies Using State Diffusion and Inverse Dynamics Models

Haonan Chen, K. Driggs-Campbell

Robotic IntelligenceDiffusion model

🎯 What it does: Using a state diffusion model to predict future states and combining it with an inverse dynamics model to generate actions for robotic bimanual manipulation

Learning Dexterous Bimanual Catch Skills Through Adversarial-Cooperative Heterogeneous-Agent Reinforcement Learning

Taewoo Kim, Jaehong Kim

Robotic IntelligenceReinforcement LearningAgentic AI

🎯 What it does: Proposes an adversarial-collaborative framework using heterogeneous agent reinforcement learning (HARL) to learn bimanual grasping skills.

Learning Direct Solution in Moving Horizon Estimation with Deep Learning Methods

Fabien Lionti, Philippe Martinet

OptimizationComputational Efficiency

🎯 What it does: Use deep learning models to learn and directly output the state estimation results of moving horizon estimation (MHE), avoiding online optimization calculations

Learning Diverse Robot Striking Motions with Diffusion Models and Kinematically Constrained Gradient Guidance

Kin Man Lee, Matthew C. Gombolay

Robotic IntelligenceDiffusion model

🎯 What it does: Propose an offline, constrained diffusion model for generating diverse and agile robotic attack actions.

Learning Dual-Arm Coordination for Grasping Large Flat Objects

Yongliang Wang, H. Kasaei

Robotic IntelligenceConvolutional Neural NetworkReinforcement Learning

🎯 What it does: Proposes a dual-arm collaborative grasping framework based on deep reinforcement learning that can complete grasping tasks without requiring additional displacement or pushing actions

Learning Dynamic Weight Adjustment for Spatial-Temporal Trajectory Planning in Crowd Navigation

Muqing Cao, Lihua Xie

Autonomous DrivingOptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Use neural networks to predict the optimal weights of each objective in the optimization planner, achieving real-time dynamic weight adjustment in crowded environments, thereby enhancing the safety and efficiency of robot navigation.

Learning Dynamics of a Ball with Differentiable Factor Graph and Roto-Translational Invariant Representations

Qingyu Xiao, Matthew C. Gombolay

Representation LearningRobotic IntelligenceGraph Neural NetworkPhysics Related

🎯 What it does: Propose an end-to-end learning framework that jointly trains dynamic models and factor graph estimators to predict the motion trajectory of a ball in dynamic environments.

Learning From Imperfect Demonstrations With Self-Supervision for Robotic Manipulation

Kun Wu, Jian Tang

Data-Centric LearningRobotic IntelligenceBenchmark

🎯 What it does: Proposed a self-supervised data filtering framework (SSDF) that computes the quality score of failure trajectory segments by combining expert demonstrations with imperfect demonstrations, thereby fully utilizing imperfect data in offline learning to enhance robot manipulation performance.

Learning Humanoid Locomotion with Perceptive Internal Model

Junfeng Long, Jiangmiao Pang

OptimizationRobotic IntelligenceReinforcement LearningWorld Model

🎯 What it does: Proposed and implemented a Perceptive Internal Model (PIM) for prone robot locomotion, which utilizes continuously updated terrain elevation maps around the robot for perception, and trains strategies in simulation using real obstacle heights, optimized based on Hybrid Internal Model (HIM), with inference conducted using sampled elevation heights from constructed elevation maps;

Learning IMU Bias with Diffusion Model

Shenghao Zhou, Guoquan Huang

Diffusion modelTime SeriesSequential

🎯 What it does: Propose an IMU bias learning method based on conditional diffusion models, using IMU measurements to predict time-varying random bias distributions

Learning In-Hand Translation Using Tactile Skin with Shear and Normal Force Sensing

Jessica Yin, T. Hellebrekers

Domain AdaptationRobotic IntelligenceReinforcement Learning

🎯 What it does: Developed a tactile skin sensor model capable of achieving zero-shot simulation-to-real transfer, and trained reinforcement learning strategies based on this model to realize hand-internal translation control through sliding contact; meanwhile, extensive real-world experiments were conducted to verify the adaptability of this strategy.

Learning Multi-Agent Loco-Manipulation for Long-Horizon Quadrupedal Pushing

Yuming Feng, Ding Zhao

Robotic IntelligenceReinforcement Learning

🎯 What it does: This paper studies the collaborative manipulation of multi-legged robots in obstacle perception and long-term pushing tasks, and proposes a three-level hierarchical multi-agent reinforcement learning framework;

Learning Multiple Probabilistic Decisions from Latent World Model in Autonomous Driving

Lingyu Xiao, Jingdong Wang

Autonomous DrivingWorld ModelBenchmark

🎯 What it does: Propose the LatentDriver framework, which uses an autoregressive world model to infer a mixed distribution of the next environmental state and possible vehicle actions through multiple probabilistic hypotheses, and generates deterministic control signals from it

Learning Object Properties Using Robot Proprioception via Differentiable Robot-Object Interaction

P. Y. Chen, Wojciech Matusik

Robotic IntelligenceTime Series

🎯 What it does: Construct a differentiable robot-object interaction model using the robot's own joint encoder information, inferring object properties (such as inertia and softness) through the robot's reaction to manipulated objects.

Learning Optimal Design Manifolds to Design More Practical Robotic Systems

J. Baumgärtner, J. Fleischer

OptimizationRobotic Intelligence

🎯 What it does: Propose and learn the 'optimal design manifold' to systematically explore the robot system design space, thereby selecting the most practically viable optimal designs; and apply it to tasks such as robot unit layout optimization, robot design optimization, and multi-camera arrangements.

Learning Quadrotor Control from Visual Features Using Differentiable Simulation

Johannes Heeg, Davide Scaramuzza

Autonomous DrivingReinforcement LearningWorld ModelImage

🎯 What it does: Learning control for quadrotors using differentiable simulation, comparing with model-agnostic RL, and finding that differentiable simulation significantly outperforms traditional RL in sample efficiency and training time, capable of achieving recovery within seconds using vehicle states and accomplishing control within minutes relying solely on visual features.

Learning Quiet Walking for a Small Home Robot

Ryo Watanabe, Marco Hutter

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposes a sim-to-real reinforcement learning method aimed at minimizing the foot contact velocity of quadrupedal home robots to reduce walking noise.

Learning Robust Policies via Interpretable Hamilton-Jacobi Reachability-Guided Disturbances

Hanyang Hu, Mo Chen

Explainability and InterpretabilityRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposes a robust strategy training framework that combines model-based control principles with adversarial RL training to enhance robustness and avoid using external black-box adversaries.

Learning Task Specifications from Demonstrations as Probabilistic Automata

Mattijs Baert, Pieter Simoens

Explainability and InterpretabilityComputational EfficiencyReinforcement Learning from Human Feedback

🎯 What it does: Propose a computationally efficient method that utilizes probabilistic deterministic finite automata (PDFA) to learn task structure and expert preferences from demonstration data, automatically inferring subgoals and their temporal dependencies to generate interpretable task specifications.

Learning Three-Dimensional Bin Packing with Adjustable-Order Semi-Online Setting

Hao Yin, Hongjie He

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: A new adjustable order semi-online 3D bin packing setting is proposed, and a strategy network and guided bottom-up packing reward function that can adapt to variable observed item quantities are designed using reinforcement learning methods to address the problem of the robot arm being hindered by already packed items.

Learning Time-Optimal Online Replanning for Distributed Model Predictive Contouring Control of Quadrotors

Xin Guan, Shuo Li

OptimizationRobotic Intelligence

🎯 What it does: Propose a time-optimal online replanning framework for multi-robot systems, combining neural networks to learn optimal time allocation for polynomial trajectories, and integrating with Model Predictive Contour Control (MPC) to achieve 100 Hz offline replanning;

Learning to Double Guess: An Active Perception Approach for Estimating the Center of Mass of Arbitrary Objects

Shengmiao Jin, Wenzhen Yuan

Hyperparameter SearchRobotic Intelligence

🎯 What it does: Achieve high-precision estimation of the center of gravity for any object through multiple active perception interactions combined with Bayesian neural networks;

Learning to Predict the Future from Monocular Vision for Efficient Human-Aware Navigation

Yushuang Huang, Zhaoqi Wang

Autonomous DrivingComputational EfficiencyRepresentation LearningReinforcement Learning from Human FeedbackConvolutional Neural NetworkSupervised Fine-TuningImage

🎯 What it does: Proposed an end-to-end model called ODVEEM based on the occlusion distance vector (ODV), which achieves human-aware navigation using monocular vision, and designed a neural network that treats ODV estimation as a classification problem with auxiliary tasks.

Learning to Prune Branches in Modern Tree-Fruit Orchards

Abhinav Jain, Stefan Lee

Robotic IntelligenceWorld ModelOptical FlowImageAgriculture Related

🎯 What it does: Proposed and implemented a closed-loop visual-motion controller for robotic pruning operations in non-productive trees, guiding the blade to precisely reach designated cut points and maintain perpendicularity to the branches.

Learning to Refine Input Constrained Control Barrier Functions via Uncertainty-Aware Online Parameter Adaptation

Taekyung Kim, Dimitra Panagou

Robotic Intelligence

🎯 What it does: Proposed a learning-based online adaptive method to dynamically adjust the parameters of input-constrained control barrier functions (ICCBF) in discrete-time nonlinear systems;

Learning to Swim: Reinforcement Learning for 6-DOF Control of Thruster-Driven Autonomous Underwater Vehicles

Levi Cai, Y. Girdhar

Domain AdaptationRobotic IntelligenceReinforcement Learning

🎯 What it does: A control strategy that directly maps 6-DOF commands to thruster outputs is trained using reinforcement learning within minutes, achieving zero-shot transfer from simulation to real AUV without parameter tuning.

Learning Visuotactile Skills With Two Multifingered Hands

Toru Lin, Jitendra Malik

Data-Centric LearningRobotic IntelligenceMultimodality

🎯 What it does: Training dual-hand manipulation skills for long-term high-precision tasks using visuotactile data collected through human demonstrations with a low-cost HATO remote operation system and modified prosthetic hands equipped with tactile sensing.

Learning Wheelchair Tennis Navigation from Broadcast Videos with Domain Knowledge Transfer and Diffusion Motion Planning

Zixuan Wu, Matthew C. Gombolay

Domain AdaptationRobotic IntelligenceReinforcement LearningDiffusion modelVideo

🎯 What it does: Proposed a zero-shot knowledge transfer framework that transfers motion navigation strategies from web videos to robotic systems, applied to wheelchair tennis navigation.

Learning-Based Adaptive Navigation for Scalar Field Mapping and Feature Tracking

Jose Fuentes, L. Bobadilla

Autonomous DrivingRobotic IntelligenceOrdinary Differential Equation

🎯 What it does: Proposed a learning-based adaptive navigation framework to enhance the efficiency and effectiveness of exploration for scalar field feature mapping and tracking.

Learning-Based Dynamic Robot-to-Human Handover

Hyeonseong Kim, Sungjoon Choi

Robotic IntelligenceVideoSequential

🎯 What it does: Proposed a learning-based dynamic robot-to-human handoff method, achieving nonparametric generation of continuous handoff motions.

Learning-Based Tip Contact Force Estimation for FBG-Embedded Continuum Robots

Majid Roshanfar, T. Looi

Robotic IntelligenceTime Series

🎯 What it does: A learning-based method for estimating the tip contact force of a continuous robot is proposed, which maps the curvature and bending angles measured by multi-core fiber Bragg grating (FBG) sensors embedded in titanium alloy tubes to three-dimensional tip forces.

Leg Exoskeleton Odometry using a Limited FOV Depth Sensor

Fabio Elnecave Xavier, Franccois Goulette

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes an odometry algorithm that combines proprioceptive data and depth camera point clouds for legged exoskeletons to generate accurate elevation maps under limited field-of-view conditions.

Legged Robot State Estimation with Invariant Extended Kalman Filter Using Neural Measurement Network

D. Youm, Jemin Hwangbo

Robotic Intelligence

🎯 What it does: Developed a self-perception state estimator for quadruped robots using a neural measurement network and an invariant extended Kalman filter.

LEMMo-Plan: LLM-Enhanced Learning from Multi-Modal Demonstration for Planning Sequential Contact-Rich Manipulation Tasks

Kejia Chen, A. Knoll

Robotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelMultimodality

🎯 What it does: Proposes the LEMMo-Plan framework, which integrates multimodal demonstrations combining tactile and torque information to enhance the LLM's ability to generate new task planning

LEVA: A High-Mobility Logistic Vehicle with Legged Suspension

M. Arnold, Marco Hutter

Robotic IntelligenceReinforcement Learning

🎯 What it does: Developed a high-load, high-mobility legged suspension logistics robot named LEVA, capable of autonomously transporting and loading cargo on diverse rugged terrains.

Leveraging LLMs for Mission Planning in Precision Agriculture

Marcos Abel Zuzu'arregui, Stefano Carpin

TransformerLarge Language ModelTextAgriculture Related

🎯 What it does: Develop an end-to-end system that enables users to assign complex data collection tasks to autonomous robots through natural language, executed via ROS2.

Leveraging Semantic and Geometric Information for Zero-Shot Robot-to-Human Handover

Jiangshan Liu, M. Meng

Robotic IntelligenceTransformerPrompt EngineeringVision Language Model

🎯 What it does: Propose a zero-shot robot-to-human handover system that generates optimal grasp poses by leveraging semantic and geometric information

Leveraging Surgical Activity Grammar for Primary Intention Prediction in Laparoscopy Procedures

Jie Zhang, Han Ding

RecognitionVision-Language-Action ModelVideoBiomedical Data

🎯 What it does: Proposed a framework that combines syntactic structure with visual cues for identifying primary intentions in laparoscopic surgery teaching videos

Leveraging Symmetry to Accelerate Learning of Trajectory Tracking Controllers for Free-Flying Robotic Systems

Jake Welde, Vijay Kumar

Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement Learning

🎯 What it does: Leveraging the Lie group symmetry of floating-base robots, an MDP homomorphism is constructed, enabling strategies trained on a low-dimensional 'quotient' MDP to be lifted as optimal trajectory tracking controllers for the original system, thereby accelerating reinforcement learning training and reducing tracking errors.

LiDAR Inertial Odometry and Mapping Using Learned Registration-Relevant Features

Zihao Dong, Michael Everett

Autonomous DrivingOptimizationComputational EfficiencySimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a learning-based LiDAR SLAM system called DFLIOM, which achieves localization and mapping by learning to select feature points related to point cloud registration, using only about 20% of the points.

LiDAR-EDIT: LiDAR Data Generation by Editing the Object Layouts in Real-World Scenes

Shing-Hei Ho, Minghan Zhu

GenerationData SynthesisPoint Cloud

🎯 What it does: Based on real-world LiDAR scans, synthetic LiDAR data is generated by editing object layouts (quantity, type, pose), maintaining the realism of the background environment and providing object labels for the generated data.

LiDAR-enhanced 3D Gaussian Splatting Mapping

Jian Shen, Gui-Song Xia

Pose EstimationGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Proposes a LiDAR-enhanced 3D Gaussian Splatting (LiGSM) mapping framework that improves the accuracy and robustness of 3D scene mapping using LiDAR data; constructs a joint loss combining images and LiDAR point clouds to estimate pose and optimize extrinsic parameters, dynamically adapting sensor alignment; initializes 3DGS with LiDAR point clouds, providing a denser and more reliable starting point than sparse SfM points; integrates LiDAR-projected depth maps during rendering to achieve geometric and photometric dual-precision scene representation.

LidarDM: Generative LiDAR Simulation in a Generated World

Vlas Zyrianov, Shenlong Wang

Data SynthesisAutonomous DrivingDiffusion modelWorld ModelPoint CloudTime Series

🎯 What it does: Proposes LidarDM, a generative model capable of generating realistic, layout-aware, physically feasible, and temporally consistent LiDAR videos and 4D point cloud sequences; the model can guide LiDAR generation based on driving scenarios.

LiDARDustX: A LiDAR Dataset for Dusty Unstructured Road Environments

Chenfeng Wei, Shenhong Wang

Autonomous DrivingPoint CloudBenchmark

🎯 What it does: This paper proposes the LiDARDustX dataset, which collects 30,000 frames of LiDAR point clouds in dusty environments, equipped with 3D bounding box annotations and semantic segmentation from six types of sensors. Based on this dataset, a benchmark experiment for 3D detection and segmentation is established, further analyzing the impact of dust on perception accuracy.

LiftFeat: 3D Geometry-Aware Local Feature Matching

Yepeng Liu, Yongchao Xu

Pose EstimationDepth EstimationRetrievalImage

🎯 What it does: Proposes a lightweight network called LiftFeat, which enhances the robustness of 2D descriptors by leveraging 3D geometric features.

Lightstereo: Channel Boost is All You Need for Efficient 2D Cost Aggregation

Xianda Guo, Long Chen

Depth EstimationImage

🎯 What it does: Propose the LightStereo network, which accelerates stereo matching by focusing on the channel dimension of the 3D cost volume.

Lightweight Yet High-Performance Defect Detector for Uav-Based Large-Scale Infrastructure Real-Time Inspection

Benyun Zhao, Ben M. Chen

Object DetectionAnomaly DetectionComputational EfficiencyConvolutional Neural NetworkImage

🎯 What it does: Proposed a lightweight and hardware-friendly UAV large-scale infrastructure defect detection model called CUPID, experimentally validated on the newly constructed CUBIT2024 dataset, and deployed on UAVs to test its real-time inference performance.

LiLoc: Lifelong Localization Using Adaptive Submap Joining and Egocentric Factor Graph

Yixin Fang, Gim Hee Lee

Autonomous DrivingOptimizationSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes a lifelong LiDAR-based localization framework named LiLoc, which utilizes adaptive subgraph connections between central sessions and sub-sessions, coarse-to-fine pose initialization, and joint optimization with self-centered factor graphs, supporting switching between relocalization and incremental localization modes.

Limits of Specifiability for Sensor-Based Robotic Planning Tasks

Basak Sakçak, J. O’Kane

Robotic Intelligence

🎯 What it does: Explore the limits of task specifiability in sensor-based robot planning tasks, and investigate the impact of specification foundations (states, actions, observations, knowledge, etc.) on specifiability

LIMT: Language-Informed Multi-Task Visual World Models

Elie Aljalbout, Nutan Chen

Representation LearningTransformerLarge Language ModelReinforcement LearningWorld ModelImageText

🎯 What it does: Propose a model-based multi-task visual world model method, leveraging pre-trained language models to extract semantic task representations, which help the world model and policy reasoning tasks in dynamics and behavioral similarity.

LiteVLoc: Map-Lite Visual Localization for Image Goal Navigation

Jianhao Jiao, Dimitrios Kanoulas

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Proposes the LiteVLoc framework, which achieves visual localization in image goal navigation using a lightweight topological-metric map, employing a three-stage coarse-to-fine camera pose estimation.

LLGS: Unsupervised Gaussian Splatting for Image Enhancement and Reconstruction in Pure Dark Environment

Haoran Wang, Mingrui Li

RestorationGaussian SplattingImage

🎯 What it does: Proposes an unsupervised multi-view stereo system named LLGS for low-light environments, used for image enhancement and scene reconstruction.

LLM-as-BT-Planner: Leveraging LLMs for Behavior Tree Generation in Robot Task Planning

Jicong Ao, Sami Haddadin

Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningPrompt Engineering

🎯 What it does: Developed a framework that utilizes LLM to generate behavior trees, named LLM-as-BT-Planner, for planning in robot assembly tasks.

LMH-MOT : A Light Multiple Hypothesis Framework for 3D Multi-Object Tracking

Tanghu Yuan, Meng Yang

Object TrackingAutonomous DrivingPoint Cloud

🎯 What it does: Proposes LMH-MOT, a lightweight multi-hypothesis framework for 3D multi-object tracking.

Local Policies Enable Zero-Shot Long-Horizon Manipulation

Murtaza Dalal, Ruslan Salakhutdinov

Domain AdaptationRobotic IntelligenceVision-Language-Action ModelMultimodalityBenchmark

🎯 What it does: Proposed ManipGen, achieving sim2real transfer for zero-shot long-horizon mechanical manipulation using local policies.

LoFSORT: Sample Online and Real-time Tracking in Low Frame Rate Scenarios

Jiabao Wang, D. Chang

Object TrackingOptical FlowVideo

🎯 What it does: Proposes a tracker based on a motion model for multi-person tracking in low frame rate scenarios.

Logic-RAG: Augmenting Large Multimodal Models with Visual-Spatial Knowledge for Road Scene Understanding

Imran Kabir, S. Billah

Autonomous DrivingLarge Language ModelVision Language ModelVideoRetrieval-Augmented Generation

🎯 What it does: Propose the Logic-RAG framework, applying visual-spatial knowledge retrieval-augmented generation (RAG) technology to large-scale multimodal models to enhance road scene understanding.

LoGS: Visual Localization via Gaussian Splatting with Fewer Training Images

Yuzhou Cheng, Dimitrios Kanoulas

Pose EstimationRetrievalGaussian SplattingImage

🎯 What it does: Proposed the LoGS visual localization pipeline, utilizing 3D Gaussian Splatting as the scene representation to achieve high-quality view synthesis and accomplish localization.

Loopy Movements: Emergence of Rotation in a Multicellular Robot

Trevor Smith, Yu Gu

Robotic Intelligence

🎯 What it does: Studied decentralized rotational motion of Loopy multi-cellular robots through local chemical signal interactions, and analyzed morphological effects and robustness to actuator failures

LoRD: Adapting Differentiable Driving Policies to Distribution Shifts

Christopher Diehl, Torsten Bertram

Domain AdaptationAutonomous DrivingSupervised Fine-Tuning

🎯 What it does: Study the performance of differentiable driving policies under distribution shift, propose the LoRD low-rank residual decoder and multi-task fine-tuning, and improve model performance in closed-loop evaluation.

Low-Rank Adaptation-Based All-Weather Removal for Autonomous Navigation

Sudarshan Rajagopalan, Vishal M. Patel

RestorationAutonomous DrivingImage

🎯 What it does: Propose LoRA and LoRA-Align methods to efficiently adapt all-weather image restoration models while maintaining original task performance.

LUMOS: Language-Conditioned Imitation Learning with World Models

Iman Nematollahi, Ingmar Posner

Robotic IntelligenceReinforcement LearningVision-Language-Action ModelWorld ModelBenchmark

🎯 What it does: Proposes LUMOS, a language-based multi-task imitation learning framework that conducts long-horizon replay training in the latent space of a learned world model, achieving zero-shot transfer on real robots.

LuVo: Lunar Visual Odometry Using Homography-Based Image Feature Matching

Ryan Soussan, M. Deans

Pose EstimationAutonomous DrivingRobotic IntelligenceImage

🎯 What it does: Developed an uninitialized stereo visual odometry (LuVo) method for the VIPER lunar rover, utilizing LightGlue for image feature matching in a distorted local plane space to improve matching robustness in large-baseline stereo sequences and repetitive terrain; and enhanced robustness against stereo correspondence failures by expanding the matching region through techniques such as estimating horizon truncation areas and Manhattan distance search.

LVBA: LiDAR-Visual Bundle Adjustment for RGB Point Cloud Mapping

Rundong Li, Fu Zhang

Pose EstimationAutonomous DrivingOptimizationSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Proposes a global LiDAR-visual bundle adjustment method called LVBA for generating high-quality RGB point cloud maps.

Lyapunov-Certified Trajectory Tracking for Mobile Robot With a Tail Wheel: Differential-Flatness and Adaptive Backstepping Design

Yuta Nishizawa, Yuji Yasui

Robotic IntelligenceOrdinary Differential Equation

🎯 What it does: Proposes a trajectory tracking control method for a two-front differential wheel mobile robot with a tail wheel. First, a trajectory tracking control law based on the differential flatness model is constructed, and the inverse staircase method is used to handle the tail wheel actuator dynamics. Then, the effectiveness is verified through hardware experiments, and finally, it is extended to adaptive tracking control under parameter uncertainty.

M2Distill: Multi-Modal Distillation for Lifelong Imitation Learning

Kaushik Roy, Peyman Moghadam

Knowledge DistillationRepresentation LearningReinforcement LearningVision-Language-Action ModelMultimodalityBenchmark

🎯 What it does: Proposes the M2Distill method, which utilizes multi-modal distillation to achieve lifelong imitation learning while maintaining latent space consistency across vision, language, and action modalities.

M3DSS: A Multi-Platform, Multi-Sensor, and Multi-Scenario Dataset for SLAM System

Shulei Huang, Xiaoguang Ma

Simultaneous Localization and MappingImageMultimodalityPoint CloudBenchmark

🎯 What it does: Proposed M3DSS, a multi-platform, multi-sensor, multi-scenario SLAM dataset;

MAC-VO: Metrics-Aware Covariance for Learning-Based Stereo Visual Odometry mac-vo.github.io

Yuheng Qiu, Sebastian A. Scherer

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Proposes MAC-VO, a learning-based stereo visual odometry framework, which trains a metric-aware uncertainty model for keypoint selection and residual weighting in pose graph optimization.

Magnetic Programming of Soft Materials Using Digitally Processed Laser Heating

Fatih Kocabas, Yunus Alapan

Physics Related

🎯 What it does: Proposes a fast parallel programming strategy for magnetic soft materials based on digitized laser heating, achieving precise control of magnetization direction by heating to the Curie temperature in an external magnetic field.

Magnetometer-Calibrated Hybrid Transformer for Robust Inertial Tracking in Robotics

Xinzhe Zheng, Chenshu Wu

Robotic IntelligenceRecurrent Neural NetworkTransformerTime SeriesSequential

🎯 What it does: Proposes NeurIT, a hybrid transformer integrating magnetometer calibration for robust inertial tracking.

Maintaining Strong $r$-Robustness in Reconfigurable Multi-Robot Networks Using Control Barrier Functions

Haejoon Lee, Dimitra Panagou

Robotic Intelligence

🎯 What it does: Propose a control barrier function to ensure sufficient strong $r$-robustness in multi-robot networks during reconfiguration, enabling leader-follower consensus.

MambaGlue: Fast and Robust Local Feature Matching with Mamba

Kihwan Ryoo, Hyun Myung

Image

🎯 What it does: Proposed a Mamba-based local feature matching method called MambaGlue, and designed two modules: MambaAttention mixer and depth confidence score regressor.

Manipulability Transfer and Tracking Control: Bridging Domain Adaptation with Predictive Feasibility

Yuhe Gong, Luis F. C. Figueredo

Domain AdaptationOptimizationRobotic Intelligence

🎯 What it does: Proposed a new framework to improve human-robot manipulability transfer and tracking in learning by demonstration.

Manual, Semi or Fully Autonomous Flipper Control? A Framework for Fair Comparison

Valentýn Cíhala, Karel Zimmermann

Robotic IntelligenceBenchmark

🎯 What it does: Studied and re-implemented existing semi-autonomous and fully autonomous control methods for landslide-type tracked robots; proposed a new semi-autonomous control strategy, introduced new metrics for evaluating cognitive load and navigation quality, and developed a benchmark interface for generating quality-load diagrams.

Map-SemNav: Advancing Zero-Shot Continuous Vision-and-Language Navigation Through Visual Semantics and Map Integration

Shuai Wu, Zhibo Pang

Vision Language ModelSimultaneous Localization and MappingMultimodality

🎯 What it does: Propose a zero-shot visual language navigation method called Map-SemNav that does not rely on large models, leveraging semantic cues such as direction, objects, and scenes to achieve generalization to unseen data categories.

MapEx: Indoor Structure Exploration with Probabilistic Information Gain from Global Map Predictions

Cherie Ho, Sebastian A. Scherer

Robotic IntelligenceSimultaneous Localization and MappingWorld ModelVideo

🎯 What it does: Propose the MapEx framework, which utilizes predictive maps to compute probability information gain for exploration in indoor structured environments.

MARF: Cooperative Multi-Agent Path Finding with Reinforcement Learning and Frenet Lattice in Dynamic Environments

Tianyang Hu, Yong Liu

Autonomous DrivingReinforcement Learning

🎯 What it does: Proposed a multi-agent path planning algorithm called MARF, designed to generate smooth paths that comply with vehicle kinematic and dynamic constraints in dynamic and complex environments.

Marginalizing and Conditioning Gaussians onto Linear Approximations of Smooth Manifolds with Applications in Robotics

Z. Guo, Tim D. Barfoot

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Proposes closed-form expressions for marginalization and conditioning of Gaussian distributions on linear manifolds, and applies them to smooth nonlinear manifolds via linearization;

MARLadona - Towards Cooperative Team Play Using Multi-Agent Reinforcement Learning

Zichong Li, Marco Hutter

Reinforcement Learning

🎯 What it does: Proposed and implemented the MARLadona decentralized multi-agent reinforcement learning training process, and created an open-source multi-agent soccer environment.

MARVEL: Multi-Agent Reinforcement Learning for Constrained Field-of-View Multi-Robot Exploration in Large-Scale Environments

Jimmy Chiun, G. Sartoretti

Robotic IntelligenceGraph Neural NetworkReinforcement Learning

🎯 What it does: Propose the MARVEL framework, utilizing Graph Attention Networks and Multi-Agent Reinforcement Learning to achieve cooperative exploration for multi-robot systems with limited visibility.