arXivSub Start free trial

IROS 2023 Papers — Page 7

IEEE/RSJ International Conference on Intelligent Robots and Systems · 1195 papers

Large Language Models as Zero-Shot Human Models for Human-Robot Interaction

Bowen Zhang, Harold Soh

Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: Explores and verifies the feasibility of using large language models (LLM) as zero-shot human models in robot interaction, conducting case studies on trust-related tasks such as desktop cleaning and dish passing.

Large Scale Pursuit-Evasion Under Collision Avoidance Using Deep Reinforcement Learning

Helei Yang, Yong Liu

TransformerReinforcement Learning

🎯 What it does: Studied multi-catcher-escapee games, proposed a hybrid attention-based state processing method, and combined it with Independent Proximal Policy Optimization (IPPO) to achieve collaborative capture and collision avoidance among multiple catchers and escapees.

Latent Emission-Augmented Perspective-Taking (LEAPT) for Human-Robot Interaction

Kaiqi Chen, Harold Soh

Robotic IntelligenceWorld Model

🎯 What it does: Propose a deep world model that enables robots to infer human perceptions and beliefs, achieving perspective transfer.

Lateral-Direction Localization Attack in High-Level Autonomous Driving: Domain-Specific Defense Opportunity via Lane Detection

Junjie Shen, Qi Alfred Chen

Autonomous DrivingAdversarial Attack

🎯 What it does: Studied the use of lane detection technology to detect and correct localization drift caused by GPS spoofing attacks, and designed the LD3 defense system.

LB-L2L-Calib 2.0: A Novel Online Extrinsic Calibration Method for Multiple Long Baseline 3D LiDARs Using Objects

Jun Zhang, Danwei Wang

Pose EstimationAutonomous DrivingPoint Cloud

🎯 What it does: This study proposes a novel multi-sensor long-baseline 3D LiDAR external calibration method called LB-L2L-Calib 2.0, which utilizes vehicles on the road as perspective-invariant features to achieve calibration.

Leader-Follower Formation Control of a Large-Scale Swarm of Satellite System Using the State-Dependent Riccati Equation: Orbit-to-Orbit and In-Same-Orbit Regulation

S. R. Nekoo, A. Ollero

OptimizationPhysics RelatedStochastic Differential Equation

🎯 What it does: Designed and implemented a soft constraint control method based on the state-dependent Riccati equation (SDRE) for master-slave inter-orbit and co-orbit control in large-scale (>1000) satellite constellations.

Learn from Incomplete Tactile Data: Tactile Representation Learning with Masked Autoencoders

Guanqun Cao, Shan Luo

Representation LearningAuto EncoderImage

🎯 What it does: Proposes a tactile representation learning method called TacMAE based on masked autoencoders, which simulates missing regions in tactile images and reconstructs surface geometry and tactile properties through reconstruction.

Learned Parameter Selection for Robotic Information Gathering

Chris Denniston, G. Sukhatme

Robotic IntelligenceReinforcement Learning

🎯 What it does: Train a reinforcement learning agent to select planner parameters during each information path planning iteration to automatically configure the information collection path planner.

Learning a Causal Transition Model for Object Cutting

Zeyu Zhang, Hangxin Liu

OptimizationRobotic Intelligence

🎯 What it does: Propose a causal transfer model based on attribute stochastic grammar for learning and planning object cutting tasks, leveraging human demonstrations to learn grammar and using MCTS to infer optimal cutting sequences.

Learning a Single Policy for Diverse Behaviors on a Quadrupedal Robot Using Scalable Motion Imitation

Arnaud Klipfel, Sehoon Ha

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose a single strategy based on deep reinforcement learning that learns multiple motor skills by imitating various reference actions including walking, turning, moving forward, jumping, sitting, and lying down.

Learning Bifunctional Push-Grasping Synergistic Strategy for Goal-Agnostic and Goal-Oriented Tasks

Dafa Ren, Xiaoqiang Ren

Robotic IntelligenceReinforcement LearningImage

🎯 What it does: A dual-functional push-grasp collaborative strategy is proposed, combining push and grasp actions to achieve efficient grasping of all objects in the workspace (target-agnostic) and predefined target objects (target-specific). The strategy uses visual observations as input, employs a dual-functional network to generate pixel-level Q-value dense maps for push and grasp primitives, and builds a hierarchical reinforcement learning framework. It treats target-agnostic tasks as combinations of multiple target-specific tasks, and adopts a two-stage training method to separately train the two types of tasks. Finally, the method directly transfers from simulation environments to the real world without requiring additional fine-tuning.

Learning Compliant Stiffness by Impedance Control-Aware Task Segmentation and Multi-Objective Bayesian Optimization with Priors

Masashi Okada, T. Taniguchi

SegmentationOptimizationHyperparameter SearchRobotic Intelligence

🎯 What it does: A variable stiffness learning method is proposed that simultaneously meets task performance and compliance requirements through multi-objective Bayesian optimization. The method defines the search space by segmenting demonstrations into task phases and assigning constant stiffness to each phase; impedance control-aware switching linear dynamics (IC-SLD) is employed for segmentation, and the stiffness obtained from IC-SLD is used as a prior to enhance optimization efficiency.

Learning Constraints on Autonomous Behavior from Proactive Feedback

Connor Basich, S. Zilberstein

Reinforcement Learning from Human FeedbackReinforcement Learning

🎯 What it does: This paper proposes a framework that learns unknown constraints through sparse and possibly active human interventions, utilizing a known reward model for task planning without relying on complete or partial demonstration trajectories, assuming human responses are not entirely passive.

Learning Contact-Based State Estimation for Assembly Tasks

Johannes Pankert, Marco Hutter

Robotic IntelligenceReinforcement Learning

🎯 What it does: A state estimation system based on contact detection and precise forward kinematics is proposed, and a reinforcement learning-based exploration strategy is designed to reduce state uncertainty.

Learning Deep Sensorimotor Policies for Vision-Based Autonomous Drone Racing

Jiawei Fu, D. Scaramuzza

Robotic IntelligenceContrastive LearningImage

🎯 What it does: Learning deep sensorimotor policies for visual drone racing, combined with feature representation learning and a learning-cheating framework to directly infer control commands from raw images.

Learning Depth Vision-Based Personalized Robot Navigation From Dynamic Demonstrations in Virtual Reality

Jorge de Heuvel, Maren Bennewitz

Robotic IntelligenceAuto EncoderImage

🎯 What it does: Proposed a deep vision-based personalized robot navigation controller learning framework, collecting user dynamic demonstrations through a virtual reality interface.

Learning from Pixels with Expert Observations

M. Hoang, Hai V. Nguyen

Robotic IntelligenceReinforcement LearningImage

🎯 What it does: Using expert observations as intermediate visual targets, enabling goal-conditioned reinforcement learning agents to complete tasks by sequentially achieving a series of goals, primarily applied to sparse reward robotic manipulation under pixel observations;

Learning from Symmetry: Meta-Reinforcement Learning with Symmetrical Behaviors and Language Instructions

Xiangtong Yao, Alois Knoll

Meta LearningReinforcement LearningText

🎯 What it does: Propose a dual MDP meta-reinforcement learning method that integrates symmetric actions and language instructions to enhance generalization and learning efficiency in multi-task scenarios

Learning Human Motion Intention for pHRI Assistive Control

P. Franceschi, M. Beschi

RecognitionRobotic IntelligenceRecurrent Neural NetworkSupervised Fine-Tuning

🎯 What it does: In physical human-robot interaction tasks, recurrent neural networks are used to identify human motion intent, which is then integrated into an assistive controller;

Learning Joint Policies for Human-Robot Dialog and Co-Navigation

Yohei Hayamizu, Shiqi Zhang

Robotic IntelligenceReinforcement Learning

🎯 What it does: Developed a framework that learns joint strategies for human-robot dialogue and collaborative navigation to efficiently and accurately complete guidance and information delivery tasks.

Learning Joint Space Reference Manifold for Reliable Physical Assistance

Amirreza Razmjoo, S. Calinon

OptimizationRobotic Intelligence

🎯 What it does: Proposed and demonstrated a method for finding a spatial reference 1D manifold for the Talos humanoid robot to achieve assistive standing/sitting tasks, while controlling the robot's stability based on forces applied by the human body during the process.

Learning Models of Adversarial Agent Behavior Under Partial Observability

Sean Ye, M. Gombolay

Graph Neural NetworkGraph

🎯 What it does: Propose GrAMMI, a graph neural network-based adversarial modeling method, to predict the current and future states of opponents under partially observable conditions.

Learning Multimodal Bipedal Locomotion and Implicit Transitions: A Versatile Policy Approach

Lokesh Krishna, Quan Nguyen

Robotic IntelligenceAuto Encoder

🎯 What it does: Propose a single multi-modal control strategy framework capable of generating various gait patterns and their inherent transitional actions, enabling diverse movements in bipedal robots.

Learning Open-Loop Saccadic Control of a 3D Biomimetic Eye Using the Actor-Critic Algorithm

Henrique Granado, Alexandre Bernardino

Robotic IntelligenceReinforcement Learning

🎯 What it does: Using the actor-critic algorithm to learn open-loop saccadic control of a 3D biological simulation eye

Learning Reduced-Order Soft Robot Controller

Chen Liang, Zherong Pan

OptimizationRobotic Intelligence

🎯 What it does: Propose a two-stage algorithm that first identifies the low-dimensional simulation space of soft robots, then identifies the control space, and uses multi-precision Riemannian Bayesian bi-level optimization to determine task-specific control spaces, thereby achieving low-dimensional control of high-dimensional soft robots.

Learning Representation for Anomaly Detection of Vehicle Trajectories

Ruochen Jiao, Qiuhan Zhu

Anomaly DetectionAutonomous DrivingAuto EncoderContrastive LearningSequential

🎯 What it does: Propose two new methods for online vehicle trajectory anomaly detection, utilizing contrastive learning and semantic latent space reconstruction under supervised and unsupervised scenarios, respectively.

Learning Robotic Assembly by Leveraging Physical Softness and Tactile Sensing

Joaquín Royo-Miquel, Kazutoshi Tanaka

Anomaly DetectionRobotic IntelligenceAuto Encoder

🎯 What it does: By mounting visual tactile sensors on a flexible robotic arm and using a variational autoencoder for anomaly detection, the study achieves autonomous assembly tasks under uncertainties in target position and grasping posture errors.

Learning Robotic Powder Weighing from Simulation for Laboratory Automation

Y. Kadokawa, Kazutoshi Tanaka

Robotic IntelligenceRecurrent Neural NetworkReinforcement Learning

🎯 What it does: A robot powder weighing task for laboratory automation was studied, achieving zero-shot transfer using simulation-to-real domain randomization and completing weighing with a recurrent neural network (RNN) strategy.

Learning Soft Robot Dynamics Using Differentiable Kalman Filters and Spatio-Temporal Embeddings

Xinyu Liu, H. B. Amor

Robotic IntelligenceTime SeriesSequential

🎯 What it does: Using differentiable Kalman filters and spatiotemporal embedding methods, end-to-end training of soft robot dynamics models to learn system dynamics, noise characteristics, and temporal behavior.

Learning Terrain-Adaptive Locomotion with Agile Behaviors by Imitating Animals

Tingguang Li, Lei Han

Robotic IntelligenceVideoSequential

🎯 What it does: Developed a general learning framework enabling quadruped robots to mimic real animal behaviors and traverse challenging terrains.

Learning to Adapt the Parameters of Behavior Trees and Motion Generators (BTMGs) to Task Variations

Faseeh Ahmad, Volker Krueger

OptimizationRobotic IntelligenceMeta Learning

🎯 What it does: Proposed a model combining Gaussian processes and weighted support vector machines to predict parameters of the Behavior Tree and Motion Generator (BTMG), enabling rapid adaptation to new task variants; constructed a proxy reward function in the optimizer to maximize performance under the given task variant.

Learning to Efficiently Plan Robust Frictional Multi-Object Grasps

Wisdom C. Agboh, Ken Goldberg

OptimizationRobotic IntelligenceConvolutional Neural Network

🎯 What it does: This paper addresses the problem of cleaning multiple rigid convex polygonal objects randomly placed on a plane. It introduces friction effects, trains a robust multi-object grasping scheme using neural networks, and compares it with single-object grasping methods through experiments.

Learning to Grasp Clothing Structural Regions for Garment Manipulation Tasks

Wei Chen, Nicolás Rojas

SegmentationRobotic IntelligenceConvolutional Neural NetworkVideo

🎯 What it does: Built a neural network-based perception system that can segment the collar of a shirt in depth images, and proposed a grasping strategy based on the segmentation results to achieve clothing grasping and hanging.

Learning to Map Efficiently by Active Echolocation

Xixi Hu, Kristen Grauman

Robotic IntelligenceReinforcement LearningSimultaneous Localization and MappingImageMultimodalityAudio

🎯 What it does: Proposed an active audio-visual mapping agent that utilizes visual data and sound echoes to estimate maps of unseen areas.

Learning to Open Doors with an Aerial Manipulator

E. Cuniato, R. Siegwart

Robotic IntelligenceReinforcement Learning

🎯 What it does: Training an Omnidirectional Micro Aerial Vehicle (OMAV) to learn door-opening actions in physical simulations using reinforcement learning, and verifying its performance in the real world.

Learning to Schedule in Multi-Agent Pathfinding

Kyuree Ahn, Jinkyoo Park

OptimizationRobotic IntelligenceGraph Neural NetworkReinforcement Learning

🎯 What it does: Study the task allocation and path planning problem in multi-robot collaboration, proposing a dual-layer optimization model, utilizing congestion-aware graph neural networks for state representation, further obtaining joint scheduling strategies through multi-agent reinforcement learning, and introducing auxiliary loss to promote collaboration.

Learning to Solve Tasks with Exploring Prior Behaviours

Ruiqi Zhu, O. Çeliktutan

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose an intrinsic reward-driven example control method (IRDEC), enabling agents to complete tasks in sparse reward environments by exploring prior behaviors and connecting them with task-specific behaviors from examples.

Learning Type-Generalized Actions for Symbolic Planning

Daniel Tanneberg, M. Gienger

🎯 What it does: Proposes a new concept that generalizes symbolic actions by leveraging entity hierarchy and observed similar behaviors, demonstrating learning from few observations and generalizing to new scenarios in a simulated grid kitchen environment.

Learning Whom to Trust in Navigation: Dynamically Switching Between Classical and Neural Planning

Sombit Dey, Christian Wolf

Robotic Intelligence

🎯 What it does: Proposed a hierarchical method with a high-level planner that dynamically switches between classical and neural planners; the feasibility was verified through comprehensive training of the neural policy in simulation and experimental validation on the LoCoBot robot.

Learning-Augmented Model-Based Planning for Visual Exploration

Yimeng Li, J. Kosecka

OptimizationConvolutional Neural NetworkReinforcement LearningWorld ModelImage

🎯 What it does: Proposed a learning-enhanced model-driven planning method for visual exploration of unknown indoor environments under time constraints.

Learning-Based Real-Time Torque Prediction for Grasping Unknown Objects with a Multi-Fingered Hand

Dominik Winkelbauer, Rudolph Triebel

Computational EfficiencyRobotic Intelligence

🎯 What it does: We propose a learning-based real-time torque prediction method, which trains a neural network using end-to-end supervised learning to predict the torque required at each joint when a multi-fingered hand grasps unknown objects.

LEF: Late-to-Early Temporal Fusion for LiDAR 3D Object Detection

Tong He, Mingxing Tan

Object DetectionAutonomous DrivingTransformerPoint Cloud

🎯 What it does: Proposes a post-to-early recursive feature fusion scheme for LiDAR 3D object detection, injecting object-aware latent embeddings into the early stages of the detector.

Legged Locomotion Control of an Under-Actuated Eccentric Paddle Mechanism with Torso Stabilization

Yanqiu Zheng, Shugen Ma

Robotic Intelligence

🎯 What it does: Proposed an underactuated controller for the ePaddle mechanism, utilizing a single ePaddle with a free trunk to achieve more efficient and flexible gaits; verified zero dynamics stability through numerical simulations;

Less Than Human: How Different Users of Telepresence Robots Expect Different Social Norms

Cheng Lin, AJung Moon

Robotic Intelligence

🎯 What it does: In an online study (N=903), the authors simulated human-robot interaction scenarios to explore whether remote operators and on-site users expect the same social norms for mobile remote presence (MRP) robots.

Leveraging Cloud Computing to Make Autonomous Vehicles Safer

Peter Schafhalter, Ion Stoica

Autonomous DrivingSafty and Privacy

🎯 What it does: Propose a system design that leverages unreliable cloud computing to enhance the decision accuracy of autonomous vehicles while ensuring the ability to fall back to onboard computing.

Leveraging Multimodal Sensing and Topometric Mapping for Human-Like Autonomous Navigation in Complex Environments

Kosmas Tsiakas, D. Tzovaras

Autonomous DrivingSimultaneous Localization and MappingImageMultimodalityPoint Cloud

🎯 What it does: Proposed a method that utilizes RGB and LiDAR data combined with rough topometric maps for free space extraction, and achieves adaptive path planning for human driver behavior through local goals and a lattice planner.

Leveraging Saliency-Aware Gaze Heatmaps for Multiperspective Teaching of Unknown Objects

Daniel Weber, Enkelejda Kasneci

Object DetectionRobotic IntelligenceImageMultimodality

🎯 What it does: Integrate augmented reality with human gaze to generate saliency-based eye heatmaps, aiding robots in learning unknown objects.

Leveraging Single-Goal Predictions to Improve the Efficiency of Multi-Goal Motion Planning with Dynamics

Yuanjie Lu, E. Plaku

Autonomous DrivingOptimization

🎯 What it does: Proposes a method to enhance the efficiency of multi-objective motion planning by combining machine learning for distance and direction prediction, TSP solvers, and sampling-based motion planning.

LiDAR Meta Depth Completion

Wolfgang Boettcher, Dengxin Dai

Depth EstimationMeta LearningPoint Cloud

🎯 What it does: Proposed a depth completion network capable of dynamically adapting to different LiDAR scanning patterns, achieving adaptive depth completion with a single model across multiple sensors;

LiDAR Missing Measurement Detection for Autonomous Driving in Rain

Chen Zhang, Daniela Rus

Anomaly DetectionAutonomous DrivingPoint Cloud

🎯 What it does: Propose a method to detect missing LiDAR measurements in rainy weather for autonomous driving, utilizing a two-stage learning approach to generate anomaly scores for missing measurements.

Lidar Panoptic Segmentation and Tracking without Bells and Whistles

Abhinav Agarwalla, Deva Ramanan

Object TrackingSegmentationPoint Cloud

🎯 What it does: Propose a detection center network for LiDAR panoptic segmentation and tracking

Lidar-Based Multiple Object Tracking with Occlusion Handling

Ruo-Tsz Ho, Wen-Chieh Lin

Object TrackingPoint Cloud

🎯 What it does: Proposed an occlusion level indicator based on LiDAR geometric information for occlusion handling in multi-object tracking, integrating it into trajectory management and data association to prevent premature trajectory termination.

LiDAR-Inertial SLAM with Efficiently Extracted Planes

Chao Chen, Yong Liu

Autonomous DrivingOptimizationSimultaneous Localization and MappingMultimodalityPoint Cloud

🎯 What it does: Proposed an efficient LiDAR-Inertial SLAM system based on plane extraction, including point-to-line-to-plane extraction, tightly coupled odometry with plane information, and global plane-assisted mapping

LiDAR-SGMOS: Semantics-Guided Moving Object Segmentation with 3D LiDAR

Shuo Gu, Hui Kong

SegmentationPoint Cloud

🎯 What it does: Propose a semantics-guided moving object segmentation network LiDAR-SGMOS, combining semantic segmentation and residual map fusion to achieve efficient moving object segmentation

Light-Field Visual System for the Remote Robot Operation Interface

Tetsuro Morimoto, Satoru Tokuhisa

Robotic Intelligence

🎯 What it does: Developed a remote robotic operation interface, and evaluated the effectiveness of light field head-mounted display (LFHMD) in eliminating vergence-accommodation conflict (VAC), enhancing depth perception, and reducing eye fatigue through TransRay, using ophthalmological experiments.

Lightweight Neural Path Planning

Jinsong Li, Jun Yu

OptimizationRobotic IntelligenceGraph

🎯 What it does: Developed a lightweight neural path planning architecture using a dual-input network and hybrid sampler, suitable for resource-constrained robotic systems.

Lightweight Real-Time Detection Model for Multi-Sheep Abnormal Behaviour Based on Yolov7-Tiny

Haotian Zhang, Meili Wang

Object DetectionAnomaly DetectionConvolutional Neural NetworkImageAgriculture Related

🎯 What it does: Constructed the ABSB sheep abnormal behavior dataset and proposed a lightweight real-time multi-sheep abnormal behavior detection model YOLOv7-Lrab based on YOLOv7-tiny.

Lightweight Semantic Segmentation Network for Semantic Scene Understanding on Low-Compute Devices

H. Son, James Weiland

SegmentationComputational EfficiencyConvolutional Neural NetworkImage

🎯 What it does: Proposed a lightweight convolutional neural network for achieving semantic scene understanding on devices with low computational resources.

Lightweight, Uncertainty-Aware Conformalized Visual Odometry

A. Stutts, A. Trivedi

Pose EstimationAutonomous DrivingComputational EfficiencySimultaneous Localization and Mapping

🎯 What it does: A lightweight, statistically robust framework for uncertainty estimation in visual odometry is proposed, which leverages conformal inference to generate adjustable multidimensional prediction intervals and combines Monte Carlo dropout with an improved training loss to achieve more accurate multidimensional uncertainty band predictions.

LIO-PPF: Fast LiDAR-Inertial Odometry via Incremental Plane Pre-Fitting and Skeleton Tracking

Xingyu Chen, Thomas H. Li

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Propose a fast LiDAR-Inertial odometry method based on incremental plane pre-fitting (PPF) and skeleton tracking

Lip-Inspired Passive Jamming Gripper with Teeth Structure

Jooyoung Hong, Joohyung Kim

Robotic Intelligence

🎯 What it does: Developed a lip-shaped passive vacuum clamp based on the structure of a dog's oral cavity, incorporating a tooth-shaped structure to enhance gripping softness and force

LIWO: LiDAR-Inertial-Wheel Odometry

Zikang Yuan, Xin Yang

Autonomous DrivingOptimizationSimultaneous Localization and MappingMultimodalityPoint Cloud

🎯 What it does: Proposes the LIWO (LiDAR-Inertial-Wheel Odometry) system, which fuses LiDAR, IMU, and wheel encoder data under a Bundle Adjustment (BA) optimization framework to achieve precise localization.

Load Awareness: Sensorless Body Payload Sensing and Localization for Heavy Quadruped Robot

Shaoxun Liu, Rongrong Wang

Robotic IntelligenceOrdinary Differential Equation

🎯 What it does: This paper proposes a sensorless method for load perception and localization in heavy quadruped robots by decomposing whole-body dynamics into body dynamics and single-leg floating dynamics, and observing virtual coupling forces between the body and legs, achieving estimation of ground reaction forces of the supporting legs and real-time calculation of unknown loads.

Local and Global Information in Obstacle Detection on Railway Tracks

Matthias Brucker, César Cadena

Object DetectionSegmentationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage

🎯 What it does: Utilizes a shallow network to segment normal railway images and achieves obstacle detection by learning to generate obstacle-free images.

Local Non-Cooperative Games with Principled Player Selection for Scalable Motion Planning

Makram Chahine, Daniela Rus

Autonomous DrivingOptimization

🎯 What it does: Proposed a planning algorithm that utilizes local games and recursive horizons to achieve game-theoretic motion planning in large-scale multi-agent systems.

LocalViT: Analyzing Locality in Vision Transformers

Yawei Li, L. Gool

ClassificationTransformerImage

🎯 What it does: Study the locality mechanism in Vision Transformers and systematically validate it by incorporating a locality module into the feed-forward network.

Locking On: Leveraging Dynamic Vehicle-Imposed Motion Constraints to Improve Visual Localization

Stephen Hausler, Michael Milford

Pose EstimationAutonomous DrivingOptimization

🎯 What it does: Propose a visual localization method that utilizes dynamic vehicles to provide pose constraints, refines the initial pose through a motion model in the PnP-RANSAC localization pipeline, and predicts the quality of future pose estimates.

Locomotion and Obstacle Avoidance of a Worm-Like Soft Robot

Sean Even, Yasemin Ozkan-Aydin

Robotic Intelligence

🎯 What it does: Designed and implemented a soft soil worm robot with efficient movement and real-time obstacle avoidance capabilities.

Locomotion Planning of a Truss Robot on Irregular Terrain

Jangho Bae, Taewon Seo

OptimizationRobotic Intelligence

🎯 What it does: Proposed a variable topology truss robot motion planning algorithm for irregular terrains, extending the Polygon-based Random Tree search to consider friction and internal force constraints, and achieving path generation through first-order rolling motion primitives.

Long-Distance Gesture Recognition Using Dynamic Neural Networks

Shubhang Bhatnagar, Liu Ren

Recognition

🎯 What it does: Propose a long-distance gesture recognition method based on dynamic neural networks, which can select spatial region features containing gestures from input sensor data and further process them.

Long-Endurance Optical Seafloor Imaging Using Underwater Gliders: Concept, Development and Initial Trials

Daniel Gregorek, Ralf Bachmayer

Robotic IntelligenceImage

🎯 What it does: Developed a system utilizing underwater gliders equipped with downward-facing cameras for long-term, fully autonomous deep-sea bottom optical imaging.

Long-Range UAV Thermal Geo-Localization with Satellite Imagery

Jiuhong Xiao, Giuseppe Loianno

Domain AdaptationImage

🎯 What it does: Proposed a thermal imaging geolocation framework using satellite RGB images, and addressed the scarcity of paired data between thermal images and satellite images through multiple domain adaptation methods, verifying its effectiveness in long-range drone low-light environments

Long-Short Term Policy for Visual Object Navigation

Yubing Bai, Shuqiang Jiang

Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningImage

🎯 What it does: Proposed a long-term and short-term goal strategy framework, achieving visual goal navigation by classifying hidden states, separately rewarding them, and combining state memory, adjustment gates, and action enhancement gates.

Look Before You Drive: Boosting Trajectory Forecasting via Imagining Future

Yixuan Fan, Shengjin Wang

Autonomous DrivingTransformerSequential

🎯 What it does: Propose a two-stage trajectory prediction framework called LBYD, which first uses Transformer to generate a rough future estimate, and then improves the final prediction accuracy through collaborative training of two networks.

LQR-Trees with Sampling Based Exploration of the State Space

Jiří Fejlek, Stefan Ratschan

OptimizationRobotic Intelligence

🎯 What it does: Extended the LQR-tree algorithm by incorporating randomized motion planning to explore uncharted regions of the state space and generate new demonstration trajectories, thereby constructing a more comprehensive feedback control strategy.

Lunar Excavator Mission Operations Using Dynamic Movement Primitives

Joseph M. Cloud, Michael A. DuPuis

Robotic IntelligencePhysics Related

🎯 What it does: Utilizing Dynamic Movement Primitives (DMP) to enable lunar mining robots to execute circular trajectories on the lunar surface while avoiding rock obstacles;

Lyapunov Constrained Safe Reinforcement Learning for Multicopter Visual Servoing

Dafang Yu, Hesheng Wang

Autonomous DrivingRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose a residual reinforcement learning framework for training multirotor drones to complete visual servoing tasks under disturbances, guided by a system Lyapunov function for safety.

Machine Learning Best Practices for Soft Robot Proprioception

Annan Zhang, Daniela Rus

Data-Centric LearningRobotic IntelligenceTime SeriesSequential

🎯 What it does: Systematic analysis of different design choices in the machine learning pipeline for soft robot proprioception, evaluating their impact on the performance of neural networks in predicting the state of soft robots.

MagHT: A Magnetic Hough Transform for Fast Indoor Place Recognition

Iad Abdul Raouf, Alexis Paljic

Pose EstimationRetrievalTime SeriesPhysics Related

🎯 What it does: Proposed a fast indoor localization algorithm based on magnetic fields called MagHT, which uses an improved Hough Transform to match magnetic measurement sequences and recover global transformations;

Magnet Array-Actuated Steerable Flexible Robot with Beacon- TfmUltrasonic Position Sensing for Robotic Neurosurgery

Xinfeng Gu, Feng Ju

Robotic IntelligenceBiomedical DataUltrasound

🎯 What it does: Introduces a magnetically controlled steerable flexible robot for flexible navigation, intraoperative ultrasound imaging, and position sensing in neurosurgery; employs piezoelectric transducers as ultrasound imaging beacons and uses permanent magnets for actuation; proposes enhancing magnetic force for flexible navigation through small magnet arrays; validates improved magnetic field distribution via simulation and models the transmission and reception functions of piezoelectric transducers; fabricates and tests a prototype to verify its feasibility in soft tissue.

Magnetic Navigation Using Attitude-Invariant Magnetic Field Information for Loop Closure Detection

Natalia Pavlasek, J. R. Forbes

Robotic IntelligenceSimultaneous Localization and MappingPhysics Related

🎯 What it does: Estimating the pose of ground robots and detecting loop closure points using magnetic field information

Magnetically Controlled Cell Robots with Immune-Enhancing Potential

Hongyan Sun, Lin Feng

Robotic IntelligenceDrug DiscoveryBiomedical Data

🎯 What it does: Proposed magnet-controlled cell robots (MCRs) based on macrophages, driven by rotating magnetic fields, loaded with immunomodulatory factors such as IL-12, CCL-5, and CXCL-10, and demonstrated precise navigation and targeting of cancer cells in vitro.

Maintaining Visibility of Dynamic Objects in Cluttered Environments Using Mobile Manipulators and Vector Field Inequalities

Fatih Dursun, Wei Pan

Robotic Intelligence

🎯 What it does: Proposes a method that utilizes a constrained kinematic controller combined with Vector Field Inequality (VFI) to ensure dynamic targets remain within the camera's field of view, achieved by representing the field of view as four infinite planes and maintaining the distance from the target to each plane above a preset threshold;

Manipulation of Center of Pressure for Bipedal Locomotion by Passive Twisting of Viscoelastic Trunk Joint and Asymmetrical Arm Swinging

Takashi Takuma, Shinya Aoi

Robotic Intelligence

🎯 What it does: Investigated the impact of the robot's upper body (including torso rotation and arm swinging) on the center of pressure (CoP), and validated three important findings through theoretical derivations, robot experiments, and simulations.

Manipulation of Optical Force-Induced Micro-Assemblies at the Air-Liquid Interface

Nicholas Carlisle, E. Avci

Robotic IntelligencePhysics Related

🎯 What it does: Dynamic assembly and manipulation of particles at the air-liquid interface are achieved through holographic optical tweezers, enabling the construction of semi-autonomous dynamic and static assemblies for reconfigurable swarm manipulation.

Mapless Urban Robot Navigation by Following Pedestrians

Sophie Buckeridge, Wesley P. Chan

Robotic Intelligence

🎯 What it does: Propose a mapless global planning method based on pedestrian following, enabling robots to navigate safely and effectively in unknown urban environments.

MapNeRF: Incorporating Map Priors into Neural Radiance Fields for Driving View Simulation

Chenming Wu, Liangjun Zhang

Autonomous DrivingNeural Radiance Field

🎯 What it does: Synthesize driving views that deviate from the trajectory by combining map priors with neural radiance fields (NeRF), while maintaining semantic road consistency.

Mapping Unknown Environments through Passive Deformation of Soft, Growing Robots

Francesco Fuentes, Laura H. Blumenschein

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Construct an environment mapping method based on passive deformation by collecting information through a soft-growing robot during collision and deformation.

Marine Vessel Attitude Estimation from Coastline and Horizon

Shobhit Singhal, Stefano Maranò

Pose EstimationConvolutional Neural NetworkImage

🎯 What it does: Proposes a method for estimating ship attitude (pitch, roll) by utilizing shoreline and horizon image streams along with known world features.

Marker-Based Visual SLAM Leveraging Hierarchical Representations

Ali Tourani, Holger Voos

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Generate a hierarchical representation of the environment using a monocular camera and fiducial markers, and improve camera pose estimation.

MaskBEV: Joint Object Detection and Footprint Completion for Bird's-Eye View 3D Point Clouds

William Guimont-Martin, P. Giguère

Object DetectionSegmentationAutonomous DrivingPoint Cloud

🎯 What it does: Propose MaskBEV, a bird's-eye view (BEV) based mask detection network that predicts instance masks for object footprints, supporting single-shot detection and footprint completion.

Masked Imitation Learning: Discovering Environment-Invariant Modalities in Multimodal Demonstrations

Yilun Hao, Dorsa Sadigh

Representation LearningRobotic IntelligenceMultimodality

🎯 What it does: Propose the Masked Imitation Learning (MIL) method, which employs a maskable policy network and two-layer optimization to automatically select useful modalities from multi-modal demonstrations, addressing the problem of over-specified states.

Material-Agnostic Shaping of Granular Materials with Optimal Transport

Nikhilesh Alatur, Lionel Ott

OptimizationRobotic IntelligenceImage

🎯 What it does: Generate robot-agnostic motion priors using computational Optimal Transport (OT), combine the next-best-sweep planner and material-agnostic sweep model to directly plan sweeping on heightmap representations, avoiding particle-level processing, thereby achieving morphing of source piles into target distributions; validate the effectiveness of this method in complex shaping tasks such as aggregation, separation, and writing letters through extensive simulations and hardware experiments.

Mathematical Modelling and Experimental Validation of an Articulated Vacuum Gripper

Matteo Maggi, G. Mantriota

Robotic Intelligence

🎯 What it does: Designed and verified a low-power vacuum gripper named Polypus, integrating low-power vacuum with asymmetric grasping, allowing adjustable phalanges based on application requirements, with a cost below 100 euros;

mCLARI: A Shape-Morphing Insect-Scale Robot Capable of Omnidirectional Terrain-Adaptive Locomotion in Laterally Confined Spaces

Heiko Kabutz, Kaushik Jayaram

Robotic Intelligence

🎯 What it does: Developed a 20mm, 0.97g quadrupedal micro-robot named mCLARI with passive morphing capability, capable of achieving omnidirectional terrain adaptation walking in laterally confined spaces, reaching a maximum speed of approximately 3 body lengths per second in experiments, with a compression ratio of up to 1.5 times;

MDT3D: Multi-Dataset Training for LiDAR 3D Object Detection Generalization

Louis Soum-Fontez, Franccois Goulette

Object DetectionPoint Cloud

🎯 What it does: Proposed and implemented a multi-dataset training (MDT3D) method to enhance the robustness of 3D object detection models in new environments with different sensor configurations.

Measuring Human-Robot Team Benefits Under Time Pressure in a Virtual Reality Testbed

Katarina Popovic, Todd D. Murphey

Robotic Intelligence

🎯 What it does: Develop an open-source VR testing platform and assess the effectiveness of four human-robot team interaction paradigms compared to a baseline without robot assistance.

Measuring People's Boredom and Indifference to the Robot's Explanation in a Museum Scenario

Rei Nagaya, Takayuki Kanda

ClassificationVideo

🎯 What it does: Collect behavioral videos of participants in a laboratory environment, annotate instances of boredom or indifference, and analyze behavioral patterns using decision trees and random forests.

MELP: Model Embedded Linear Policies for Robust Bipedal Hopping

Raghav Soni, Shishir N Y Kolathaya

Robotic Intelligence

🎯 What it does: Proposed the MELP method that embeds the Spring Loaded Inverted Pendulum (SLIP) model into a linear policy, achieving continuous bipedal jumping in simulation and hardware.

MEM: Multi-Modal Elevation Mapping for Robotics and Learning

Gian Erni, Marco Hutter

Computational EfficiencyRobotic IntelligenceSimultaneous Localization and MappingImageMultimodalityPoint Cloud

🎯 What it does: By fusing multi-modal information (point cloud and image), the 2.5D robot-centric elevation mapping framework is extended to achieve unified processing of multi-source data and realize real-time multi-modal layers;