arXivSub Start free trial

ICRA 2024 Papers — Page 10

IEEE International Conference on Robotics and Automation · 1760 papers

Learning-Aided Control of Robotic Tether-Net with Maneuverable Nodes to Capture Large Space Debris

Achira Boonrath, Souma Chowdhury

Robotic IntelligenceReinforcement Learning

🎯 What it does: A hierarchical decentralized trajectory planning and control method is proposed for a maneuverable net system launched from an unmanned space station, enabling the capture of large space debris.

Learning-Aided Warmstart of Model Predictive Control in Uncertain Fast-Changing Traffic

Mohamed-Khalil Bouzidi, Joerg Reichardt

Autonomous DrivingOptimization

🎯 What it does: Proposes a learning-based MPC warmstart framework that uses a neural network multi-modal predictor to generate multiple trajectory proposals and further optimizes them through sampling techniques to identify multiple local optima and provide better initial guesses.

Learning-based Inverse Perception Contracts and Applications

Dawei Sun, S. Mitra

Autonomous DrivingRobotic Intelligence

🎯 What it does: Proposed a learning-based inverse perception contract method and applied it to the drone visual pipeline to achieve safe landing

Learning-based Model Predictive Control for an Autonomous Formula Student Racing Car

D. R. Gomes, Pedro U. Lima

Autonomous DrivingOptimizationRecurrent Neural Network

🎯 What it does: Proposed and implemented a learning-based model predictive control (LMPC) for automated Formula Student racing cars, combining MPCC with neural networks to correct first-principle model errors.

Learning-Based Motion Planning with Mixture Density Networks

Yinghan Wang, Jianping He

Autonomous DrivingOptimizationPoint Cloud

🎯 What it does: Propose a multi-modal neural planner (MNP) based on a hybrid density network for motion planning in point cloud environments. First, point clouds are compressed into latent vectors, and then a multi-modal planning network learns and predicts multiple optimal paths.

Legible and Proactive Robot Planning for Prosocial Human-Robot Interactions

Jasper Geldenbott, Karen Leung

Robotic Intelligence

🎯 What it does: A generic robot trajectory planning framework is proposed to synthesize readable and proactive behaviors, thereby naturally guiding prosocial interactions

Lens Capsule Tearing in Cataract Surgery using Reinforcement Learning

R. Peter, F. Mathis-Ullrich

Robotic IntelligenceReinforcement LearningMesh

🎯 What it does: Developed an interactive, physically realistic simulation based on the finite element method (FEM) to simulate the continuous curve capsulotomy (CCC) process, and used reinforcement learning (RL) to train control strategies in this simulation environment;

Less is More: Physical-Enhanced Radar-Inertial Odometry

Qiucan Huang, Huan Yin

Autonomous DrivingSimultaneous Localization and MappingPoint CloudPhysics Related

🎯 What it does: Designed a physics-enhanced localization system that integrates radar with inertial measurement units (IMUs), leveraging Doppler velocity and radar cross-section (RCS) information.

LESS-Map: Lightweight and Evolving Semantic Map in Parking Lots for Long-term Self-Localization

Mingrui Liu, Liang Li

Autonomous DrivingComputational EfficiencySimultaneous Localization and MappingImage

🎯 What it does: Proposes a lightweight, evolvable semantic mapping, localization, and map updating system based on ground semantic features and low-cost cameras.

LET-3D-AP: Longitudinal Error Tolerant 3D Average Precision for Camera-Only 3D Detection

Wei-Chih Hung, Drago Anguelov

Object DetectionAutonomous DrivingImageBenchmark

🎯 What it does: Proposed longitudinal error-tolerant variants of the 3D average precision (3DAP) metric, LET-3D-AP and LET-3D-APL, aiming to be more tolerant of depth errors in camera-only detection

Leveraging Compliant Tactile Perception for Haptic Blind Surface Reconstruction

L. Y. E. R. Cheret, T. E. D. Oliveira

Robotic Intelligence

🎯 What it does: Using compliant tactile perception and robot kinematic data to blindly measure and reconstruct unknown non-planar surfaces.

Leveraging Cycle-Consistent Anchor Points for Self-Supervised RGB-D Registration

Siddharth Tourani, Dinesh Reddy Narapureddy

Pose EstimationDepth EstimationRecurrent Neural NetworkSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Utilize cyclic consistency keypoints and a novel pose module to enhance RGB-D registration accuracy.

Leveraging Neural Radiance Fields for Uncertainty-Aware Visual Localization

Le Chen, Marc Pollefeys

Pose EstimationDepth EstimationNeural Radiance Field

🎯 What it does: Generate training samples using NeRF, design NeRF to predict uncertainties in rendered color and depth, formulate scene coordinate regression (SCR) as deep evidential learning with appearance uncertainty, and construct a novel view selection strategy based on three types of uncertainty.

Leveraging Opportunism in Sample-Based Motion Planning

Michael W. Lanighan, Oscar Youngquist

OptimizationRobotic Intelligence

🎯 What it does: Proposed an opportunistic RRT* method (ORRT*) aimed at quickly finding solutions, reducing computational waste, and improving data efficiency

Leveraging Tethers for Distributed Formation Control of Simple Robots

Sadie Cutler, Kirstin Petersen

Robotic IntelligencePhysics Related

🎯 What it does: By connecting robots with non-powered, flexible, fixed-length ropes in a simulated environment, and using the angles and tension (strain) of the ropes for distributed formation control, the transition and stabilization of formation shapes were achieved.

Leveraging the efficiency of multi-task robot manipulation via task-evoked planner and reinforcement learning

Haofu Qian, Shiqiang Zhu

Robotic IntelligenceReinforcement LearningBenchmark

🎯 What it does: Propose a planning-guided reinforcement learning method that utilizes a Task-Induced Planner (TEP) to enhance learning efficiency and performance in multi-task robot manipulation.

LHMap-loc: Cross-Modal Monocular Localization Using LiDAR Point Cloud Heat Map

Xinrui Wu, Hesheng Wang

Pose EstimationCompressionAutonomous DrivingOptical FlowMultimodalityPoint Cloud

🎯 What it does: Propose the LHMap-loc pipeline to achieve accurate and efficient localization of a monocular camera in pre-built LiDAR point cloud maps

LiDAR Data Synthesis with Denoising Diffusion Probabilistic Models

Kazuto Nakashima, Ryo Kurazume

Data SynthesisDiffusion modelPoint Cloud

🎯 What it does: Generate diverse and high-fidelity 3D LiDAR point clouds using a denoising diffusion probabilistic model and provide a LiDAR completion pipeline based on this model

LiDAR-based Robot Transplanter

Masaki Asano, Takanori Fukao

OptimizationRobotic IntelligenceSimultaneous Localization and MappingPoint CloudAgriculture Related

🎯 What it does: Developed a LiDAR-based self-localization method and a robust control scheme to achieve automated transplanting

LiDAR-camera Calibration using Intensity Variance Cost

Ryoichi Ishikawa, Katsushi Ikeuchi

Pose EstimationAutonomous DrivingOptimizationImagePoint Cloud

🎯 What it does: Proposes a method for extrinsic calibration by utilizing intensity variations from projecting camera images onto LiDAR point clouds;

LiDAR-Camera Extrinsic Calibration with Hierachical and Iterative Feature Matching

Xuzhong Hu, Jie Ma

Autonomous DrivingConvolutional Neural NetworkPoint Cloud

🎯 What it does: Proposed a deep learning-based LiDAR-Camera extrinsic calibration network called HIFM-Net, which achieves precise alignment between the camera and LiDAR using global and hierarchical feature matching.

LiDAR-CS Dataset: LiDAR Point Cloud Dataset with Cross-Sensors for 3D Object Detection

Jin Fang, Liangjun Zhang

Object DetectionData SynthesisDomain AdaptationPoint CloudBenchmark

🎯 What it does: Created the LiDAR-CS dataset and evaluated and analyzed the baseline detectors

LiDARFormer: A Unified Transformer-based Multi-task Network for LiDAR Perception

Zixiang Zhou, H. Foroosh

Object DetectionSegmentationAutonomous DrivingTransformerPoint Cloud

🎯 What it does: Proposes LiDARFormer, a unified Transformer-based multi-task network for simultaneously performing LiDAR 3D detection and semantic segmentation.

Lifelong LERF: Local 3D Semantic Inventory Monitoring Using FogROS2

Adam Rashid, Kenneth Y. Goldberg

OptimizationRobotic IntelligenceNeural Radiance FieldSimultaneous Localization and MappingImageText

🎯 What it does: Propose a system named Lifelong LERF that jointly optimizes dense language and geometric representations under low computational resources using mobile robots, maintains environmental representations through semantic change detection, allows users to retrieve 3D heatmaps of target objects via natural language queries, and offloads compute-intensive tasks to the Fog-ROS2 cloud platform. The system obtains poses via monocular RGBD SLAM and achieves semantic monitoring using Language Embedded Radiance Field (LERF).

Lifelong Robot Learning with Human Assisted Language Planners

Meenal Parakh, Pulkit Agrawal

Robotic IntelligenceTransformerLarge Language ModelAgentic AI

🎯 What it does: Proposes a method that utilizes a large language model (LLM) planner to query and learn new skills, enabling robots to efficiently acquire skills for handling rigid objects in terms of data and time, and reusing the newly acquired skills for subsequent tasks, demonstrating potential in open-world and lifelong learning.

Lifelong Robot Library Learning: Bootstrapping Composable and Generalizable Skills for Embodied Control with Language Models

Georgios Tziafas, H. Kasaei

Robotic IntelligenceTransformerLarge Language ModelText

🎯 What it does: This paper proposes a lifelong learning robot library (LRLL) based on large language models, which continuously expands the robot's skill library to handle increasingly complex manipulation tasks through a soft memory module, automatic exploration strategy, skill abstractor, and human-robot interactive lifelong learning algorithm.

Light-weight approach for safe landing in populated areas

Tilemachos Mitroudas, A. Gasteratos

Object DetectionComputational EfficiencySimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposes a lightweight safe landing pipeline based on the state-of-the-art object detector and OctoMap, first generating point clouds of surface obstacles and inserting them into OctoMap, then identifying unoccupied regions and counting safe landing points; this method achieves low processing time and is suitable for lightweight embedded systems.

Lightning NeRF: Efficient Hybrid Scene Representation for Autonomous Driving

Junyi Cao, Chao Ma

Autonomous DrivingNeural Radiance FieldPoint Cloud

🎯 What it does: Propose Lightning NeRF, which uses an efficient hybrid scene representation and leverages LiDAR geometric priors to enhance the quality of novel view synthesis in autonomous driving environments, while reducing computational costs for training and rendering.

Lightweight and Compliant Bilateral Teleoperation System with Anthropomorphic Arms for Aerial and Ground Service Operations

Alejandro Suárez, A. Ollero

Robotic Intelligence

🎯 What it does: Proposes a dual-arm teleoperation system based on intelligent servo, aimed at achieving flexible manipulation tasks in aerial or ground service operations.

Lightweight Event-based Optical Flow Estimation via Iterative Deblurring

Yilun Wu, G. D. Croon

Computational EfficiencyOptical FlowTime Series

🎯 What it does: Proposes a lightweight event-based optical flow estimation network, IDNet, which directly estimates optical flow using event trajectories without constructing correlation volumes.

Lightweight Ground Texture Localization

Aaron Wilhelm, Nils Napp

Computational EfficiencyImage

🎯 What it does: Propose a lightweight ground texture localization algorithm (L-GROUT) that can run in real-time on a single-board computer without GPU acceleration, achieving millimeter-level positioning accuracy;

Lightweight Untethered Soft Robotic Fish

Xiangxing Wang, Taogang Hou

Robotic Intelligence

🎯 What it does: Developed a small untethered soft robotic fish

LIKO: LiDAR, Inertial, and Kinematic Odometry for Bipedal Robots

Qingrui Zhao, Qiang Huang

Robotic IntelligenceSimultaneous Localization and MappingPoint CloudTime Series

🎯 What it does: Propose a tightly coupled LiDAR-IMU-kinematic odometry (LIKO), achieving state estimation for biped robots through an iterative extended Kalman filter, while modeling and estimating foot contact positions.

LIO-EKF: High Frequency LiDAR-Inertial Odometry using Extended Kalman Filters

Yibin Wu, H. Kuhlmann

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a tightly-coupled LiDAR-IMU odometry system LIO-EKF based on point-to-point registration and extended Kalman filter.

Liquids Identification and Manipulation via Digitally Fabricated Impedance Sensors

Junyi Zhu, Wojciech Matusik

ClassificationRobotic Intelligence

🎯 What it does: Impedance sensors and digital embroidery electrode arrays are integrated into a robotic gripper to collect liquid manipulation data and train a learning model for liquid classification and state estimation.

LiRaFusion: Deep Adaptive LiDAR-Radar Fusion for 3D Object Detection

Jingyu Song, Katherine A. Skinner

Object DetectionAutonomous DrivingMultimodalityPoint Cloud

🎯 What it does: Proposed LiRaFusion to address the fusion of LiDAR and radar for 3D object detection.

Lissajous Curve-Based Vibrational Orbit Control of a Flexible Vibrational Actuator with a Structural Anisotropy

Yuto Miyazaki, Mitsuru Higashimori

Robotic IntelligencePhysics Related

🎯 What it does: This paper proposes a flexible vibration actuator with structural anisotropy and its control method. By analyzing models and theoretical derivations, it demonstrates how laser point trajectory control can be achieved using synthetic wave inputs based on resonance frequencies. Subsequently, the control scheme is validated through prototype experiments and applied to an underactuated walking robot.

LiSTA: Geometric Object-Based Change Detection in Cluttered Environments

Joseph Rowell, Maurice F. Fallon

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed the LiSTA (LiDAR Spatio-Temporal Analysis) system for detecting object-level probabilistic changes in multi-task LiDAR SLAM scenarios, applicable to semi-static environments where objects are added, removed, or displaced over weeks or months.

Lite-SVO: Towards A Lightweight Self-Supervised Semantic Visual Odometry Exploiting Multi-Feature Sharing Architecture

Wenhui Wei, Yangfan Zhou

Pose EstimationAutonomous DrivingSimultaneous Localization and Mapping

🎯 What it does: Propose Lite-SVO lightweight multi-feature sharing architecture to enhance the efficiency and accuracy of self-supervised semantic visual odometry on edge devices

LiteTrack: Layer Pruning with Asynchronous Feature Extraction for Lightweight and Efficient Visual Tracking

Qingmao Wei, Guotian Zeng

Object TrackingComputational EfficiencyTransformerVideo

🎯 What it does: Proposed a lightweight Transformer-based visual tracking model called LiteTrack, which is efficiently optimized for real-time robots and edge devices.

LLM-Assisted Multi-Teacher Continual Learning for Visual Question Answering in Robotic Surgery

Kexin Chen, P. Heng

Robotic IntelligenceTransformerLarge Language ModelMixture of ExpertsVision Language ModelMultimodality

🎯 What it does: Proposes a visual question answering framework based on multi-teacher continual learning, utilizing multi-modal large language models (LLMs) as additional teachers. It designs a method to convert LLM embeddings into log probabilities and introduces an adaptive weight allocation mechanism to balance the generalization ability of LLMs with the specialized knowledge of older models, while constructing a new surgical visual question answering dataset.

LLM-BT: Performing Robotic Adaptive Tasks based on Large Language Models and Behavior Trees

Haotian Zhou, Huasong Min

Robotic IntelligenceTransformerLarge Language Model

🎯 What it does: Implementing adaptive robotic tasks based on LLM and behavior trees, utilizing ChatGPT to infer task steps and construct semantic maps.

LLM-Grounder: Open-Vocabulary 3D Visual Grounding with Large Language Model as an Agent

Jianing Yang, Joyce Chai

RetrievalLarge Language ModelAgentic AI

🎯 What it does: Proposes LLM-Grounder, a zero-shot, open-vocabulary LLM-driven 3D visual localization pipeline.

Log Loading Automation for Timber-Harvesting Industry

Elie Ayoub, Inna Sharf

Robotic IntelligenceImageAgriculture Related

🎯 What it does: Implement a complete autonomous loading pipeline for fixed-base robotic arms, enabling the loading of logs from forests to sawmills.

Long-HOT: A Modular Hierarchical Approach for Long-Horizon Object Transport

S. Narayanan, Manmohan Chandraker

Robotic IntelligenceBenchmark

🎯 What it does: Designed the Long-HOT task and proposed a modular topology graph-driven long-term object transportation strategy (HTP), achieving efficient object picking and transportation through hierarchical motion planning and goal navigation.

Long-Tailed 3D Semantic Segmentation with Adaptive Weight Constraint and Sampling

Jean Lahoud, Salman H. Khan

SegmentationPoint Cloud

🎯 What it does: To address the long-tailed distribution in 3D semantic segmentation, we propose an adaptive regularization and sampling regularization method based on classifier weights.

Longitudinal Control Volumes: A Novel Centralized Estimation and Control Framework for Distributed Multi-Agent Sorting Systems

James Maier, Matthew Travers

Optimization

🎯 What it does: Propose a framework named Longitudinal Control Volume (LCV) for centralized estimation and control in multi-agent classification systems; construct a material flow model, use a Kalman filter to fuse local measurements into global estimates, and employ model predictive control algorithms to optimize material flow rates in real time.

Looking Beneath More: A Sequence-based Localizing Ground Penetrating Radar Framework

Pengyu Zhang, Liang Shen

RecognitionImageSequential

🎯 What it does: Proposed a sequence-based underground penetrating radar localization framework, introducing a trainable strategy to extract robust underground features under multiple weather conditions; leveraging sequence information to observe richer underground scene contexts, thereby enhancing underground location recognition performance.

Looking Inside Out: Anticipating Driver Intent From Videos

Yung-chi Kung, Joydeep Biswas

Autonomous DrivingRecurrent Neural NetworkTransformerVideo

🎯 What it does: Propose a method to predict future driver actions by utilizing in-vehicle and external camera data, combined with manually extracted object and road-level features.

LORIS: A Lightweight Free-Climbing Robot for Extreme Terrain Exploration

Paul Nadan, Aaron M. Johnson

OptimizationRobotic Intelligence

🎯 What it does: Designed and demonstrated a lightweight fully mobile climbing robot capable of vertical climbing on vertical flat and irregular rock surfaces.

LOTUS: Continual Imitation Learning for Robot Manipulation Through Unsupervised Skill Discovery

Weikang Wan, Yuke Zhu

Robotic Intelligence

🎯 What it does: Proposed the LOTUS algorithm, enabling robots to continuously learn new operational tasks through a few human demonstrations throughout their lifetime, achieving this by constructing an expanding skill library.

Low-to-High Resolution Path Planner for Robotic Gas Distribution Mapping

Rohit V. Nanavati, Cunjia Liu

OptimizationComputational EfficiencyRobotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Proposed a low-to-high resolution path planner that first generates a low-resolution gas distribution map using sparse sampling, then uses this map as a prior for high-resolution sampling to enhance task efficiency.

LPFormer: LiDAR Pose Estimation Transformer with Multi-Task Network

Dongqiangzi Ye, H. Foroosh

Pose EstimationAutonomous DrivingTransformerPoint Cloud

🎯 What it does: Proposes the first end-to-end 3D human pose estimation framework using only LiDAR, named LPFormer

LPS-Net: Lightweight Parameter-shared Network for Point Cloud-based Place Recognition

Chengxin Liu, Ran Song

RetrievalConvolutional Neural NetworkPoint Cloud

🎯 What it does: Propose a lightweight parameter-sharing network (LPS-Net) for point cloud scene recognition, which includes multi-scale bidirectional perception units and a parameter-shared NetVLAD aggregation module;

LSSAttn: Towards Dense and Accurate View Transformation for Multi-modal 3D Object Detection

Qi Jiang, Hao Sun

Autonomous DrivingTransformerMultimodalityPoint Cloud

🎯 What it does: In multi-modal 3D detection, an optimized view transformation module is proposed, leveraging the LSS mechanism and attention to achieve dense association between perspective pixels and BEV grids. Additionally, cross-attention and multi-scale feature fusion are integrated into the BEVFusion framework to enhance view transformation performance.

Lumped Drag Model Identification and Real-Time External Force Detection for Rotary-Wing Micro Aerial Vehicles

Lucas Wälti, A. Martinoli

OptimizationRobotic IntelligenceTime Series

🎯 What it does: Proposed an integrated drag model and identified parameters using an offline gradient method, relying only on onboard accelerometers, attitude estimation, and throttle commands to achieve real-time prediction of aerodynamic forces generated by the vehicle's own movements, thereby distinguishing external disturbances.

Lumped Parameter Dynamic Model of an Eversion Growing Robot: Analysis, Simulation and Experimental Validation

Panagiotis Vartholomeos, C. Bergeles

Robotic IntelligencePhysics RelatedOrdinary Differential Equation

🎯 What it does: Built an aggregated parameter dynamic model of a pressure-driven rotary soft robot (with conduit) and developed a MATLAB simulation framework to verify its physical principles and operational areas.

MAC-ID: Multi-Agent Reinforcement Learning with Local Coordination for Individual Diversity

Hojun Chung, Songhwai Oh

Reinforcement LearningBenchmark

🎯 What it does: Propose MAC-ID, which generates diverse pedestrian movements by utilizing local coordination factors; create the BSON benchmark and introduce the SNS metric to evaluate social navigation methods.

Machine Learning-Driven Burrowing with a Snake-Like Robot

Sean Even, Yasemin Ozkan-Aydin

Robotic IntelligenceTime Series

🎯 What it does: Designed and implemented a snake-like robot equipped with an IMU and two sets of triaxial magnetometers, and proposed a deep learning control strategy based on simulating depth using magnetic field intensity to achieve vertical excavation.

MAexp: A Generic Platform for RL-based Multi-Agent Exploration

Shaohao Zhu, Jinming Xu

Robotic IntelligenceReinforcement LearningPoint CloudBenchmark

🎯 What it does: Proposed the MAexp platform, which represents multi-robot exploration scenarios using point clouds, integrates various state-of-the-art MARL algorithms, and is equipped with an attention mechanism-based multi-robot target generator and single-robot motion planner, supporting any number of robots and continuous action control; built the first MARL benchmark covering typical scenarios through large-scale experiments;

MagicTac: A Novel High-Resolution 3D Multi-layer Grid-Based Tactile Sensor

Wen Fan, Dandan Zhang

Robotic IntelligenceOptical Flow

🎯 What it does: Designed and manufactured the MagicTac high-resolution 3D multi-layer grid tactile sensor, and evaluated its performance in tactile reconstruction tasks.

Magnetic Mobile Micro-Gripping MicroRobots (MMµGRs) with Two Independent Magnetic Actuation Modes

Aaron C. Davis, D. Cappelleri

Robotic Intelligence

🎯 What it does: Developed a magnetic mobile micro-grasping micro-robot with two independent magnetic drive modes, achieving precise control of the robot's position and orientation by adjusting the magnetic moment orientation of two magnets, and completing the grasping action by generating internal stress through opposite torque.

Magnetic-Guided Flexible Origami Robot toward Long-Term Phototherapy of H. pylori in the Stomach

Sishen Yuan, Hongliang Ren

Robotic Intelligence

🎯 What it does: Developed and implemented a magnetically controlled flexible origami robot that uses flexible printed circuit units to provide long-term photodynamic therapy for H. pylori.

MAL: Motion-Aware Loss with Temporal and Distillation Hints for Self-Supervised Depth Estimation

Yue-Jiang Dong, Song-Hai Zhang

Depth EstimationKnowledge DistillationVideo

🎯 What it does: Propose Motion-Aware Loss (MAL), leveraging the temporal relationship between consecutive frames and the spatial position of moving objects to improve multi-frame self-supervised depth estimation methods for dynamic environments

Manipulator as a Tail: Promoting Dynamic Stability for Legged Locomotion

Huang Huang, J. Malik

Robotic IntelligenceReinforcement Learning

🎯 What it does: Studying the dynamic stability role of a robotic arm in multi-legged robots during high-speed locomotion and under external disturbances, proposing to use the robotic arm as a 'tail' to enhance turning performance and anti-disturbance capabilities, and developing an incremental training process that gradually releases degrees of freedom with the assistance of behavior cloning.

ManyQuadrupeds: Learning a Single Locomotion Policy for Diverse Quadruped Robots

M. Shafiee, A. Ijspeert

Robotic IntelligenceReinforcement Learning

🎯 What it does: Learned and implemented a single motion strategy capable of controlling quadruped robots with varying shapes, masses, and sizes, drawing inspiration from the central pattern generator (CPG) and pattern formation (PF) mechanisms in animal locomotion.

Mapping High-level Semantic Regions in Indoor Environments without Object Recognition

Roberto Bigazzi, Neural Mapper

Representation LearningRobotic IntelligenceVision Language ModelSimultaneous Localization and MappingMultimodality

🎯 What it does: Propose a method for semantic region mapping in indoor environments through embodied navigation, generating high-level agent knowledge representation.

Markerless Ultrasnd Probe Pose Estimation in Mini-Invasive Surgery

Mohammad Mahdi Kalantari, Adrien Bartoli

Pose EstimationImageUltrasound

🎯 What it does: Propose a method that does not require markers or additional sensors, using a single standard laparoscope monocular RGB image to estimate the pose of the ultrasound probe relative to the laparoscope.

Marrying NeRF with Feature Matching for One-step Pose Estimation

Ronghan Chen, Yu Ren

Pose EstimationNeural Radiance Field

🎯 What it does: A one-step real-time pose estimation method is achieved by integrating image matching with NeRF;

Mask4Former: Mask Transformer for 4D Panoptic Segmentation

Kadir Yilmaz, Bastian Leibe

Object TrackingSegmentationAutonomous DrivingTransformerPoint Cloud

🎯 What it does: Design Mask4Former to achieve 4D panoptic segmentation, unifying semantic instance segmentation with temporal tracking;

Masked Local-Global Representation Learning for 3D Point Cloud Domain Adaptation

Bowei Xing, Ruibin Wang

Domain AdaptationRepresentation LearningPoint Cloud

🎯 What it does: Proposes a self-supervised method that utilizes masked representation learning, combining masked feature prediction and masked sample consistency to learn cross-domain invariant point cloud representations, and enhances knowledge transfer effectiveness by learning domain-specific representations through prototype-based self-training.

Masked Visual-Tactile Pre-training for Robot Manipulation

Qingtao Liu, Jiming Chen

Robotic IntelligenceTransformerReinforcement LearningVision-Language-Action ModelMultimodality

🎯 What it does: Developed a visuo-tactile data acquisition system, collected a bottle cap rotation dataset, and proposed the M2VTP visuo-tactile fusion network for pre-training. The pre-trained model was subsequently embedded into a reinforcement learning framework to complete downstream robotic manipulation tasks.

Masked γ-SSL: Learning Uncertainty Estimation via Masked Image Modeling

David S. W. Williams, Daniele de Martini

SegmentationRepresentation LearningImageBenchmark

🎯 What it does: Proposes a semantic segmentation network that can generate high-quality uncertainty estimates in a single forward pass.

Mastering Stacking of Diverse Shapes with Large-Scale Iterative Reinforcement Learning on Real Robots

Thomas Lampe, M. Riedmiller

Robotic IntelligenceReinforcement LearningImageBenchmark

🎯 What it does: By embedding offline reinforcement learning methods into an iterative online/offline 'collect-reasoning' framework, leveraging all collected experiences, training with only real robot data, learning directly from pixels without requiring a simulator or demonstrations; significantly improving performance on real robot control benchmarks.

MateRobot: Material Recognition in Wearable Robotics for People with Visual Impairments

Junwei Zheng, R. Stiefelhagen

RecognitionSegmentationRobotic IntelligenceTransformerImage

🎯 What it does: Developed a wearable visual robotic system called MATERobot to help visually impaired individuals identify object categories and materials through vision before contact, and proposed a lightweight MATEViT model to achieve pixel-level semantic segmentation;

MATRIX: Multi-Agent Trajectory Generation with Diverse Contexts

Zhuo Xu, Jiachen Li

GenerationSequential

🎯 What it does: Proposes a learning-based multi-agent trajectory generation model called MATRIX, designed to generate diverse interactive trajectories in multi-human or human-robot interaction scenarios.

MAVIS: Multi-Camera Augmented Visual-Inertial SLAM using SE2(3) Based Exact IMU Pre-integration

Yifu Wang, Hongdong Li

Autonomous DrivingOptimizationSimultaneous Localization and MappingImageMultimodality

🎯 What it does: Proposes an optimization-based visual-inertial SLAM system called MAVIS for multi-camera systems, utilizing wide field-of-view and IMU metric scales.

Maximizing Quadruped Velocity by Minimizing Energy

Srinath Mahankali, Pulkit Agrawal

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposed and implemented the Extrinsic-Intrinsic Policy Optimization (EIPO) framework, which enhances task performance and energy consumption simultaneously in quadruped robot speed maximization tasks through constrained optimization.

MBFusion: A New Multi-modal BEV Feature Fusion Method for HD Map Construction

Xiaoshuai Hao, ByungIn Yoo

Autonomous DrivingImageMultimodalityPoint Cloud

🎯 What it does: Proposed a multi-modal BEV feature fusion method called MBFusion for high-precision map construction, addressing the semantic misalignment between camera and LiDAR features and achieving adaptive information fusion.

MBot: A Modular Ecosystem for Scalable Robotics Education

Peter Gaskell, O. C. Jenkins

Robotic Intelligence

🎯 What it does: Designed and promoted a low-cost, modular MBot mobile robot platform for training over 1400 students in robotics courses at the University of Michigan and its collaborating institutions

Mean Shift Mask Transformer for Unseen Object Instance Segmentation

Ya Lu, Yu Xiang

SegmentationTransformerImage

🎯 What it does: Proposed a differentiable Mean Shift Mask Transformer (MSMFormer) for unseen object instance segmentation

Measurement-limited Multi-Agent, Relative Pose Estimation for On-Orbit Inspection

Mark Mercier, Clark N. Taylor

Pose EstimationAutonomous DrivingOptimizationRobotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Presents a multi-agent relative pose estimation method that maintains robustness under sensor occlusion and dynamic uncertainty.

Measuring Ball Joint Faults in Parabolic-Trough Solar Plants with Data Augmentation and Deep Learning

M. Pérez-Cutiño, J. Valverde

Object DetectionSegmentationAnomaly DetectionConvolutional Neural NetworkImage

🎯 What it does: A method for automatically detecting ball joints in parabolic trough collector systems is proposed. The method uses images collected by drones, employs deep learning for segmentation to extract ball joint components, generates rich training data through data augmentation, detects liquid leaks via image color filtering, and measures robotic arm angles to analyze geometric abnormalities, providing corresponding evaluation metrics.

Mechanical Design and Kinematics of a Multimodal Two-wheeled Robot

Botian Sun, Xuefeng Wang

Robotic Intelligence

🎯 What it does: This paper designs a multi-modal two-wheeled robot with additional structural degrees of freedom to balance control inputs and degrees of freedom, avoiding over-constrained planar motion; proposes a transitional mode based on diagonal vehicle motion, constructs a generic kinematic model, and treats bicycle and self-balancing modes as special constraint cases; designs a control law for structural degrees of freedom and verifies the feasibility of smooth multi-modal transitions through experiments on a prototype robot.

Mechanism Design for New Sensors Field Deployment by LineRanger Powerline Robot

P. Richard, N. Pouliot

OptimizationRobotic Intelligence

🎯 What it does: Implemented three new asset management tasks on the LineRanger power line robot: measuring adjacent conductors using a micro-ohmmeter, installing/retrieving multi-day monitoring sensors, and assessing conductor aging surface characteristics to improve thermal models and optimize line capacity during heatwaves.

MEDL-U: Uncertainty-aware 3D Automatic Annotation based on Evidential Deep Learning

Helbert Paat, Tong Zhang

Autonomous DrivingExplainability and InterpretabilityData-Centric LearningTransformerPoint Cloud

🎯 What it does: Proposed a 3D automatic annotation framework named MEDL-U based on Evidential Deep Learning (EDL), which can quantify uncertainty while generating pseudo-labels.

MegaParticles: Range-based 6-DoF Monte Carlo Localization with GPU-Accelerated Stein Particle Filter

Kenji Koide, A. Banno

OptimizationSimultaneous Localization and Mapping

🎯 What it does: Proposed a 6-DoF range-based Monte Carlo localization method using a GPU-accelerated Stein particle filter.

Merging Decision Transformers: Weight Averaging for Forming Multi-Task Policies

Daniel Lawson, A. H. Qureshi

Robotic IntelligenceTransformer

🎯 What it does: Construct a multi-task model by averaging the parameter space of Decision Transformers trained on different MuJoCo locomotion tasks, without requiring centralized training.

Meta-Reinforcement Learning Based Cooperative Surface Inspection of 3D Uncertain Structures using Multi-robot Systems

Junfeng Chen, Tin Lun Lam

Meta LearningReinforcement Learning

🎯 What it does: Propose a decentralized collaborative motion planning method based on meta-learning for surface inspection of uncertain 3D structures

Metrically Scaled Monocular Depth Estimation through Sparse Priors for Underwater Robots

Luca Ebner, Stefan Williams

Depth EstimationConvolutional Neural NetworkTransformerImage

🎯 What it does: Proposed a real-time dense depth estimation model that enhances monocular image depth prediction and resolves scale ambiguity by leveraging sparse depth priors generated from triangulated features.

MF-MOS: A Motion-Focused Model for Moving Object Segmentation

Jintao Cheng, Rui Fan

SegmentationAutonomous DrivingConvolutional Neural NetworkOptical FlowPoint CloudBenchmark

🎯 What it does: Proposed MF-MOS, a motion-focused model for LiDAR moving object segmentation, which employs a dual-branch structure to separate spatial-temporal information.

MiBOT: A head-worn robot that modulates cardiovascular responses through human-like soft massage

A. Mylaeus, Christian Holz

Robotic IntelligenceBiomedical Data

🎯 What it does: Propose and evaluate a head-mounted massage robot MiBOT, using two soft tentacles to perform tactile actions similar to human massage, and assess blood pressure and heart rate in subjects.

Microexpression to Macroexpression: Facial Expression Magnification by Single Input

Yaqi Song, Jianfeng Li

Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkOptical FlowImage

🎯 What it does: Using deep learning methods to upscale single micro-expression images into macro-expression images.

Microrobotic Flight Enabled by Ultralight Ion Thrusters with High Thrust-to-Weight Ratio and Low Fabrication Cost

Yang Gu, Wei Li

Robotic IntelligencePhysics Related

🎯 What it does: Designed and fabricated a low-weight, high thrust-to-weight ratio, and low-cost ion propulsion micro aerial vehicle.

MIM: Indoor and Outdoor Navigation in Complex Environments Using Multi-Layer Intensity Maps

A. Sathyamoorthy, Dinesh Manocha

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed and implemented a Multi-layer Intensity Map (MIM) as a 3D object representation for robot perception and autonomous navigation in complex indoor and outdoor environments, and developed an adaptive obstacle inflation strategy based on MIM to handle narrow passages and non-solid obstacles;

Mission Planning for Multiple Autonomous Underwater Vehicles with Constrained In Situ Recharging

Priti Singh, Geoffrey A. Hollinger

OptimizationRobotic Intelligence

🎯 What it does: Proposes a task planning framework for multiple autonomous underwater vehicles (AUVs) under limited on-site charging conditions, leveraging centralized evolutionary algorithms and decentralized Monte Carlo Tree Search (MCTS);

Mitigating Causal Confusion in Vector-Based Behavior Cloning for Safer Autonomous Planning

Jiayu Guo, Jian Pu

Autonomous DrivingMultimodality

🎯 What it does: Proposed an off-policy disentangled supervision method to alleviate causal confusion in vector-based behavioral cloning, and designed a decoder utilizing iterative path fusion to better capture environmental cues, followed by validation of its effectiveness through reactive and non-reactive closed-loop simulations on the nuPlan dataset.

Mixed Traffic Control and Coordination from Pixels

Michael Villarreal, Weizi Li

Autonomous DrivingReinforcement LearningImage

🎯 What it does: Studied the performance of robotic vehicles using image observations in mixed traffic control and coordination

MM4MM: Map Matching Framework for Multi-Session Mapping in Ambiguous and Perceptually-Degraded Environments

Zhenyu Wu, Danwei Wang

OptimizationSimultaneous Localization and Mapping

🎯 What it does: Proposed a new probabilistic magnetic sensing multi-session mapping framework called MM4MM, for estimating relative transformations of multi-session maps and constructing globally consistent maps in ambiguous and perception-degraded environments.

MMA-Net: Multiple Morphology-Aware Network for Automated Cobb Angle Measurement

Z. Qiu, Jiankun Wang

SegmentationConvolutional Neural NetworkImageBiomedical Data

🎯 What it does: Proposed a multi-modal perception network called MMA-Net, which achieves automated Cobb angle measurement by fusing various spinal morphological information.