arXivSub Start free trial

ICRA 2023 Papers — Page 9

IEEE International Conference on Robotics and Automation · 1341 papers

Monocular Visual-Inertial Odometry with Planar Regularities

Chuchu Chen, Guoquan Huang

Pose EstimationSimultaneous Localization and MappingImage

🎯 What it does: Designed a real-time monocular visual-inertial odometry system that utilizes planar features for complete constraints through a lightweight multi-state constraint Kalman filter (MSCKF)

MonoGraspNet: 6-DoF Grasping with a Single RGB Image

Guangyao Zhai, Benjamin Busam

Pose EstimationDepth EstimationRobotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Proposes a 6-DoF grasping framework called MonoGraspNet based on a single RGB image, which can simultaneously handle grasping of arbitrary objects and overcome challenges posed by optically difficult objects.

MonoPGC: Monocular 3D Object Detection with Pixel Geometry Contexts

Zizhang Wu, Jian Pu

Object DetectionDepth EstimationAutonomous DrivingTransformerImage

🎯 What it does: Propose MonoPGC, an end-to-end monocular 3D object detection framework that leverages rich pixel geometry context;

Morphological Characteristics That Enable Stable and Efficient Walking in Hexapod Robot Driven by Reflex-based Intra-limb Coordination

Wataru Sato, D. Owaki

Robotic Intelligence

🎯 What it does: Built an insect-inspired hexapod robot simulation model, employing reflex-based intra-foot coordination control to investigate the impact of body morphology on walking performance.

Motion Planning for a Climbing Robot with Stochastic Grasps

Stephanie Newdick, M. Pavone

OptimizationRobotic Intelligence

🎯 What it does: Developed a motion planning framework applicable to ReachBot, combining graph traversal algorithms to select discrete grasping sequences, and using a phased (body motion and end-effector motion) decoupled motion planner, employing sampling-based planning and sequential convex optimization methods for phase optimization; achieving trajectory planning with at least 90% success probability in a simulated 2D cave environment, and validating the generated body motion trajectories through a simplified prototype.

MPC with Sensor-Based Online Cost Adaptation

Avadesh Meduri, L. Righetti

OptimizationRobotic Intelligence

🎯 What it does: Introduce an MPC scheme based on neural networks that continuously updates the quadratic programming (QP) cost function, enabling robots to minimize general non-convex task loss through sensory inputs without solving non-convex problems, and directly adapt to environmental changes from sensor measurements.

MPOGames: Efficient Multimodal Partially Observable Dynamic Games

Oswin So, Evangelos A. Theodorou

OptimizationComputational EfficiencyReinforcement LearningMultimodality

🎯 What it does: Proposes the MPOGames method, which reformulates multi-modal partially observable dynamic game problems as POMDP, enabling efficient solutions for maximum entropy dynamic games, capturing interactions between different local Nash equilibria, and validating its significance and real-time performance in two-agent merging cases and a 1/10 scale vehicle platform.

MRI-powered Magnetic Miniature Capsule Robot with HIFU-controlled On-demand Drug Delivery

M. E. Tiryaki, M. Sitti

Drug DiscoveryMagnetic Resonance ImagingUltrasound

🎯 What it does: Designed and demonstrated a magnetic microrobotic capsule driven by MRI, controlled drug release using acoustic streaming generated by MRI-guided high-intensity focused ultrasound (HIFU), utilizing a polymer shell with bubble-blocked drug release pores, where HIFU pulses remove bubbles to achieve on-demand release;

Multi-Agent Active Search using Detection and Location Uncertainty

Arundhati Banerjee, J. Schneider

OptimizationReinforcement Learning

🎯 What it does: Proposed a joint reasoning method for handling object detection and location uncertainty, and built a decentralized multi-agent active search decision algorithm based on this method using Thompson sampling.

Multi-Agent Path Integral Control for Interaction-Aware Motion Planning in Urban Canals

Lucas Streichenberg, Javier Alonso-Mora

Autonomous DrivingOptimizationReinforcement Learning

🎯 What it does: Proposed a decentralized, communication-free interaction-aware motion planner for autonomous ships in urban canals, which uses MPPI to simultaneously compute collision-free vehicle trajectories and trajectory predictions of other agents, while ensuring rule compliance through a two-stage sample evaluation strategy and a compliance cost function.

Multi-Agent Spatial Predictive Control with Application to Drone Flocking

Andreas Brandstätter, R. Grosu

OptimizationRobotic Intelligence

🎯 What it does: Propose Spatial Predictive Control (SPC), a fully distributed controller that minimizes a distributed cost function by utilizing only the position observations of the robot itself and its neighbors, applicable to drone swarm flight.

Multi-Alpha Soft Actor-Critic: Overcoming Stochastic Biases in Maximum Entropy Reinforcement Learning

C. Igoe, J. Schneider

Autonomous DrivingRobotic IntelligenceReinforcement Learning

🎯 What it does: In robot control, this paper proposes Multi-Alpha Soft Actor-Critic (MAS) by treating the entropy coefficient α as a random variable, aiming to overcome the stochastic bias in SAC while maintaining learning stability and sample efficiency.

Multi-Contact Task and Motion Planning Guided by Video Demonstration

Kateryna Zorina, V. Petrík

Pose EstimationRobotic IntelligenceVideo

🎯 What it does: Propose a multi-contact task and motion planning method based on teaching videos, which utilizes grasp and release states extracted from videos to expand RRT, forming a multi-tree parallel search.

Multi-embodiment Legged Robot Control as a Sequence Modeling Problem

Chenyi Yu, Jun Wang

Robotic IntelligenceTransformer

🎯 What it does: Propose a learning-based control method that jointly models control and robot morphology using a plasticity-aware Transformer (EAT), enabling the transfer of control strategies across different morphologies.

Multi-Head Attention Machine Learning for Fault Classification in Mixed Autonomous and Human-Driven Vehicle Platoons

Theodore Wu, Deepa Kundur

ClassificationAnomaly DetectionAutonomous DrivingTransformerTime SeriesSequential

🎯 What it does: Propose a multi-head attention machine learning (MHA-ML) method for identifying five categories of faults and anomalies in mixed autonomous and human-driven vehicle fleets.

Multi-modal Hierarchical Transformer for Occupancy Flow Field Prediction in Autonomous Driving

Haochen Liu, Chen Lv

Autonomous DrivingTransformerMultimodality

🎯 What it does: Proposed a multi-modal hierarchical Transformer network for jointly predicting flow and occupancy information in the occupancy flow field.

Multi-modal Interactive Perception in Human Control of Complex Objects

Rashida Nayeem, D. Sternad

Robotic IntelligenceMultimodality

🎯 What it does: Investigate the role of visual and tactile information in human exploration interaction, using a robotic interface to simulate a nonlinear internal dynamics 'coffee cup' system, and compare performance under different perceptual conditions.

Multi-Modal Learning and Relaxation of Physical Conflict for an Exoskeleton Robot with Proprioceptive Perception

Xuan Zhang, Xiang Li

Anomaly DetectionSafty and PrivacyRobotic IntelligenceMultimodality

🎯 What it does: Propose a multi-modal ontology-aware learning scheme to detect physical conflicts between the human body and exoskeleton robots, and alleviate conflicts through continuous regulation of robot impedance to achieve safe and efficient interaction.

Multi-Object Navigation in real environments using hybrid policies

Assem Sadek, Christian Wolf

Autonomous DrivingSupervised Fine-TuningReinforcement LearningSimultaneous Localization and Mapping

🎯 What it does: This paper transfers the Multi-ON task from simulation environments to real-world environments and proposes a hybrid strategy: decomposing navigation into two types of skills, using classical SLAM and symbolic planners for route navigation, and leveraging deep neural networks combined with supervised learning and reinforcement learning for exploration, semantic mapping, and target retrieval.

Multi-Objective Ergodic Search for Dynamic Information Maps

Ananya Rao, H. Choset

OptimizationComputational Efficiency

🎯 What it does: Proposed the dynamic multi-objective isentropic search algorithm (D-MO-ES) for efficiently planning isentropic trajectories on varying objectives.

Multi-Robot 3D Gas Distribution Mapping: Coordination, Information Sharing and Environmental Knowledge

Chiara Ercolani, A. Martinoli

Robotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Investigate the collaborative effects of information-driven navigation strategies and multi-robot systems in gas distribution mapping tasks under time constraints, and propose a method to enhance mapping performance by injecting additional knowledge.

Multi-Robot Coordination and Cooperation with Task Precedence Relationships

W. Gosrich, Vijay R. Kumar

OptimizationRobotic IntelligenceGraph

🎯 What it does: Proposed a new multi-robot task planning and allocation model that covers task priority relationships, task coordination, and robot collaboration (through forming alliances to complete tasks that individual robots cannot accomplish); the model uses a task graph to describe tasks and their relationships, and defines a reward function to characterize the impact of collaboration scale and task performance.

Multi-Robot Mission Planning in Dynamic Semantic Environments

Samarth Kalluraya, Y. Kantaros

Robotic Intelligence

🎯 What it does: This paper proposes a multi-robot task planning method for uncertain and dynamically changing semantic environments, utilizing sensing robots to accomplish collaborative semantic tasks based on uncertain semantic target positions and labels;

Multi-segmented Adaptive Feet for Versatile Legged Locomotion in Natural Terrain

A. Chatterjee, Alexander Badri-Spröwitz

Robotic Intelligence

🎯 What it does: Develop a multi-segment adaptive foot mechanism, achieving and testing anti-slip and anti-sinking performance for the first time on bird-inspired multi-joint mechanical tendon-connected legs.

Multi-source Domain Adaptation for Unsupervised Road Defect Segmentation

Jongmin Yu, Shang Luo

SegmentationDomain AdaptationImage

🎯 What it does: Propose a multi-source domain adaptation method for unsupervised road surface defect segmentation

Multi-State Tightly-Coupled EKF-Based Radar-Inertial Odometry With Persistent Landmarks

J. Michalczyk, S. Weiss

Robotic IntelligenceSimultaneous Localization and MappingMultimodality

🎯 What it does: Propose a multi-state tightly coupled EKF radar-inertial odometry method, which utilizes the sparse noise signals of FMCW radar fused with IMU to estimate the complete 6-DOF pose and 3D velocity of a robot in unstructured environments.

Multi-swarm Genetic Gray Wolf Optimizer with Embedded Autoencoders for High-dimensional Expensive Problems

J. Bi, MengChu Zhou

OptimizationAuto EncoderBenchmark

🎯 What it does: Proposed an evolutionary optimization framework combining autoencoders to address high-dimensional, time-consuming evaluation problems;

Multi-Target Pursuit by a Decentralized Heterogeneous UAV Swarm using Deep Multi-Agent Reinforcement Learning

Maryam Kouzeghar, Roland Bouffanais

Robotic IntelligenceReinforcement Learning

🎯 What it does: Propose a role-based MADDPG algorithm for multi-target pursuit and exploration by multi-drone swarms in non-stationary, unknown environments.

Multi-to-Single Knowledge Distillation for Point Cloud Semantic Segmentation

Shoumeng Qiu, Jian Pu

SegmentationKnowledge DistillationPoint Cloud

🎯 What it does: Propose a multi-to-single knowledge distillation framework that focuses only on hard class instances, and employs multi-layer distillation (feature, logit, affinity) and instance-aware affinity algorithms to enhance the performance of 3D point cloud semantic segmentation for hard classes.

Multi-View Keypoints for Reliable 6D Object Pose Estimation

Alan Li, Angela P. Schoellig

Pose EstimationImagePoint Cloud

🎯 What it does: Propose a 6D object pose estimation method based on multi-view, which integrates heatmaps and keypoint estimates into a 3D space probability density map using an eye-in-hand camera transformation.

Multi-view object pose estimation from correspondence distributions and epipolar geometry

R. Haugaard, Thorbjørn Mosekjær Iversen

Pose EstimationImage

🎯 What it does: Proposes a multi-view object pose estimation method that generates an initial estimate by aggregating learned 2D-3D correspondence distributions from multiple views, with optional refinement;

Multiagent Reinforcement Learning for Autonomous Routing and Pickup Problem with Adaptation to Variable Demand

Daniel Garces, D. Bertsekas

Autonomous DrivingOptimizationReinforcement LearningTime Series

🎯 What it does: A multi-agent reinforcement learning framework is developed to generate adaptive autonomous vehicle dispatching and passenger pickup strategies for handling randomly appearing requests on urban maps, achieving coordination among vehicles, non-greedy decision-making, and adaptation to changes in demand distribution.

Multimodal Image Registration for GPS-denied UAV Navigation Based on Disentangled Representations

Huandong Li, Feiyan Wu

Pose EstimationAutonomous DrivingOptimizationOptical FlowMultimodality

🎯 What it does: A multimodal image registration method is proposed for near-infrared and visible light images, applied to UAV positioning in GPS-denied environments.

Multimodal Neural Radiance Field

Haidong Zhu, Cheng-Hao Kuo

GenerationData SynthesisNeural Radiance FieldImageMultimodalityPoint Cloud

🎯 What it does: Construct a multi-modal NeRF using multi-modal inputs such as RGB, infrared images, and point clouds to achieve more accurate scene reconstruction and address the geometric blur problem caused by using only RGB.

Multimodal Time Series Learning of Robots Based on Distributed and Integrated Modalities: Verification with a Simulator and Actual Robots

Hideyuki Ichiwara, T. Ogata

Robotic IntelligenceRecurrent Neural NetworkTime Series

🎯 What it does: Proposed an autonomous robot motion generation model based on distributed and integrated multimodal learning, verified on simulated and real robots.

Multiple Surgical Instruments Tracking-By-Prediction With Graph Hierarchy

Rui Guo, A. Jarc

Object TrackingGraph Neural NetworkBiomedical Data

🎯 What it does: Propose a new paradigm that integrates trajectory prediction into multi-instrument tracking to reduce data association errors caused by false detections

MVFusion: Multi-View 3D Object Detection with Semantic-aligned Radar and Camera Fusion

Zizhang Wu, Jian Pu

Object DetectionAutonomous DrivingTransformerMultimodality

🎯 What it does: Propose MVFusion, a multi-view radar-camera fusion 3D object detection method that can achieve semantic alignment of radar features and enhance cross-modal information interaction.

MVTrans: Multi-View Perception of Transparent Objects

Yi Ru Wang, Animesh Garg

SegmentationData SynthesisPose EstimationDepth EstimationConvolutional Neural NetworkImage

🎯 What it does: Proposed a multi-view based transparent object perception method called MVTrans, which can simultaneously perform depth estimation, segmentation, and pose estimation, and is trained using a new programmatically generated high-resolution synthetic dataset named Syn-TODD.

NanoFlowNet: Real-time Dense Optical Flow on a Nano Quadcopter

Rik J. Bouwmeester, G. D. Croon

Computational EfficiencyRobotic IntelligenceConvolutional Neural NetworkOptical FlowVideo

🎯 What it does: Propose NanoFlowNet, a lightweight convolutional neural network for real-time dense optical flow estimation on nano drones using edge computing hardware;

Natural Language Instruction Understanding for Robotic Manipulation: a Multisensory Perception Approach

Weihua Wang, Huaping Liu

Robotic IntelligenceVision-Language-Action ModelImageMultimodalityAudio

🎯 What it does: Propose a multisensory perception method that uses vision, touch, and hearing as three modalities to collaboratively achieve natural language instruction understanding for robotic operations and execute corresponding manipulation tasks.

Navigating Soft Robots through Wireless Heating

Yiwen Song, Swarun Kumar

Robotic Intelligence

🎯 What it does: Proposed a battery-free soft robot platform that utilizes wireless microwave heating to achieve crawling and tracking in confined spaces.

Navigation with polytopes and B-spline path planner

Ngoc Thinh Nguyen, F. Ernst

Autonomous DrivingOptimization

🎯 What it does: Proposes an algorithm for optimal path planning within a 2D non-convex polyhedral region, utilizing B-spline curves parameterized via Bézier representation, with linear constraints on control points to confine the entire curve within the polyhedron and ensure path feasibility.

NeRF-Loc: Visual Localization with Conditional Neural Radiance Field

Jianlin Liu, Chengjie Wang

Pose EstimationTransformerNeural Radiance Field

🎯 What it does: Propose a visual relocalization method based on directly matching implicit 3D descriptors with 2D images using a transformer.

nerf2nerf: Pairwise Registration of Neural Radiance Fields

Lily Goli, Andrea Tagliasacchi

Pose EstimationOptimizationNeural Radiance FieldPoint Cloud

🎯 What it does: Proposed a registration method called nerf2nerf based on NeRF, introducing a surface field to achieve illumination-invariant point cloud registration, and further extending the local registration of classical ICP to NeRF scenes;

NeRF2Real: Sim2real Transfer of Vision-guided Bipedal Motion Skills using Neural Radiance Fields

Arunkumar Byravan, N. Heess

Domain AdaptationRobotic IntelligenceReinforcement LearningNeural Radiance FieldVideo

🎯 What it does: Utilizing short videos obtained from ordinary smartphones, the method employs NeRF to learn the contact geometry of static scenes and new view synthesis functions; dynamic objects (e.g., robot bodies, spheres) are overlaid onto NeRF-rendered static scenes. Contact dynamics are computed in a physics simulator based on the rendering engine, enabling the learning of vision-based whole-body navigation and sphere-pushing strategies within this simulation. These strategies are ultimately successfully transferred to a real 20-degree-of-freedom humanoid robot.

NeRFing it: Offline Object Segmentation Through Implicit Modeling

Kenneth Blomqvist, R. Siegwart

SegmentationNeural Radiance FieldVideo

🎯 What it does: Infer scene geometry using NeRF, combine with a user-provided single 3D bounding box, and compute high-quality object segmentation maps from RGB-D video sequences.

Neural Contact Fields: Tracking Extrinsic Contact with Tactile Sensing

C. Higuera, Mustafa Mukadam

Robotic IntelligenceNeural Radiance FieldMultimodality

🎯 What it does: Proposed the Neural Contact Fields method for tracking external contact between objects and the environment

Neural Grasp Distance Fields for Robot Manipulation

Thomas Weng, Mustafa Mukadam

OptimizationRobotic Intelligence

🎯 What it does: Frame grasp learning as a neural field, taking a 6D end-effector pose as input and outputting the distance to a continuous valid grasp manifold.

Neural Implicit Surface Reconstruction using Imaging Sonar

Mohamad Qadri, Ioannis Gkioulekas

GenerationNeural Radiance FieldImageMeshUltrasound

🎯 What it does: Achieving dense 3D reconstruction using imaging sonar

Neural Optimal Control using Learned System Dynamics

Selim Engin, Volkan Isler

OptimizationReinforcement Learning

🎯 What it does: Studies how to generate control laws for systems with unknown dynamics by representing the controller and value function with neural networks, and training them using a modified Hamilton-Jacobi-Bellman (HJB) equation; first learns state transitions from offline data, then integrates them into the HJB equation for forward simulation control.

Neural-Kalman GNSS/INS Navigation for Precision Agriculture

Yayun Du, M. Srivastava

Simultaneous Localization and MappingVideoAgriculture Related

🎯 What it does: Proposed a lightweight neural-Kalman filter, a user-friendly video processing toolbox, and a publicly available precision agriculture robot neural-inertial navigation dataset.

Neuro-Adaptive Dynamic Control with Edge-Computing for Collaborative Digital Twin of an Industrial Robotic Manipulator

S. Das, S. Baidya

OptimizationRobotic IntelligenceWorld Model

🎯 What it does: Established a digital twin architecture based on a model-free neural adaptive controller and edge computing for industrial robot operations, with experimental validation of grasp-place task performance under dynamic obstacles.

New Bracket Polynomials Associated with the General Gough-Stewart Parallel Robot Singularities

F. Thomas

Robotic Intelligence

🎯 What it does: Derived a new bracket polynomial for describing singularities of the Gough-Stewart platform, interpreting all brackets as reciprocal products between lines using linear algebra methods.

NIFT: Neural Interaction Field and Template for Object Manipulation

Zeyu Huang, Ruizhen Hu

Robotic Intelligence

🎯 What it does: Introduces NIFT, an interactive representation for describing and implementing object manipulation to facilitate imitation learning

NOCaL: Calibration-Free Semi-Supervised Learning of Odometry and Camera Intrinsics

Ryan-Rhys Griffiths, D. Dansereau

Pose EstimationAutonomous Driving

🎯 What it does: Proposes NOCaL, a semi-supervised learning architecture capable of performing pose estimation, camera intrinsic parameter prediction, and scene appearance modeling for previously unseen cameras without calibration;

Noise and Environmental Justice in Drone Fleet Delivery Paths: A Simulation-Based Audit and Algorithm for Fairer Impact Distribution

Zewei Zhou, Martim Brandão

OptimizationTabular

🎯 What it does: Study the impact of drone delivery fleets on urban noise pollution and environmental justice, using simulation to analyze noise distribution and propose an algorithm to balance noise.

Non-cooperative Stochastic Target Encirclement by Anti-synchronization Control via Range-only Measurement

Fen Liu, Lihua Xie

OptimizationRobotic Intelligence

🎯 What it does: Studied the positioning estimation of non-cooperative random targets in extreme environments where GPS is unavailable, weight is limited, and there is no ground guidance, and designed a unique distributed anti-synchronization controller (DASC) enabling two drones to rapidly track and surround the target;

Non-Minimal Solvers for Relative Pose Estimation with a Known Relative Rotation Angle

Deshun Hu

Pose Estimation

🎯 What it does: Proposes two solvers for relative pose estimation with non-minimal correspondence numbers under known relative rotation angles.

Noncontact Particle Manipulation on Water Surface with Ultrasonic Phased Array System and Microscopic Vision

Yexin Zhang, Song Liu

Robotic IntelligenceImageUltrasoundPhysics Related

🎯 What it does: Dynamic non-contact manipulation of microparticles on water surfaces using an ultrasonic phased array system and microscopic vision, including automatic capture, closed-loop positioning, and real-time motion planning.

Nonlinear Model Predictive Control of a 3D Hopping Robot: Leveraging Lie Group Integrators for Dynamically Stable Behaviors

Noel Csomay-Shanklin, A. Ames

OptimizationRobotic IntelligenceStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: In this paper, the authors implement dynamic stable bouncing, trajectory tracking, and flipping of a novel three-dimensional bouncing robot through a multi-rate hierarchical structure that combines hybrid nonlinear model predictive control (MPC) with a low-level feedback controller;

NOPA: Neurally-guided Online Probabilistic Assistance for Building Socially Intelligent Home Assistants

Xavier Puig, A. Torralba

Robotic Intelligence

🎯 What it does: Proposed a neural network-based online goal inference and assistance method (NOPA) for building social intelligent robots in home environments.

Novel Spring Mechanism Enables Iterative Energy Accumulation under Force and Deformation Constraints

Cole A. Dempsey, D. Braun

Physics Related

🎯 What it does: By using a novel float spring mechanism, negative work is accumulated over multiple motion cycles, enabling energy storage in the assist spring.

NVRadarNet: Real-Time Radar Obstacle and Free Space Detection for Autonomous Driving

A. Popov, Nikolai Smolyanskiy

Autonomous DrivingTime Series

🎯 What it does: Developed a real-time dynamic obstacle and free space detection deep neural network called NVRadarNet based on RADAR.

Object Reconfiguration with Simulation-Derived Feasible Actions

Yiyuan Lee, L. Kavraki

Robotic Intelligence

🎯 What it does: In the 3D object reassembly task, the authors embed a physics simulator into a motion planner to implicitly discover and specify valid actions for each state, avoiding manual encoding of action semantics.

Object-based SLAM utilizing unambiguous pose parameters considering general symmetry types

Tae-Kyeong Lee, H. Kim

Object TrackingOptimizationSimultaneous Localization and Mapping

🎯 What it does: By classifying symmetric objects into three categories based on their symmetry features, extracting unambiguous parameters, and using them for joint optimization of camera and object pose to enhance SLAM robustness.

Obscuring Objectives with Pareto-Optimal Privacy-Aware Trajectories in Multi-Robot Coverage

Brennan Brodt, Alyssa Pierson

OptimizationSafty and Privacy

🎯 What it does: Propose a method using genetic algorithms to generate Pareto-optimal privacy-aware multi-robot coverage paths to conceal the team's objectives;

Observability-Aware Active Extrinsic Calibration of Multiple Sensors

S. Xu, Sen Wang

Optimization

🎯 What it does: Propose an entropy-based active extrinsic calibration algorithm that utilizes observability analysis and information entropy to enhance the accuracy and reliability of multi-sensor extrinsic calibration.

Obstacle avoidance using Raycasting and Riemannian Motion Policies at kHz rates for MAVs

Michael Pantic, Lionel Ott

Robotic IntelligencePoint Cloud

🎯 What it does: Propose a method for real-time obstacle avoidance on voxelized maps using GPU ray casting and thousands of parallel Riemannian Motion Policies (RMP), demonstrating successful avoidance of static and dynamic obstacles on a real MAV.

Obstacle Identification and Ellipsoidal Decomposition for Fast Motion Planning in Unknown Dynamic Environments

Mehmetcan Kaymaz, N. K. Ure

Autonomous Driving

🎯 What it does: Obstacles are identified using ellipses, and their linear and angular velocities are estimated.

Obstacle-Aware Topological Planning over Polyhedral Representation for Quadrotors

Junjie Gao, Yu Yao

Autonomous DrivingMeshGraph

🎯 What it does: A framework for obstacle perception and topological planning for quadrotor drones is proposed. First, the occupied voxels are approximated as polyhedra through a polyhedral mapping algorithm, and a dedicated data structure is designed to enable real-time information extraction. Subsequently, a local topological graph is constructed to guide path search, with segmented graph vertex searches achieving rapid convergence. After path search, smoothing and penalty optimization are applied to generate a refined trajectory that maintains obstacle clearance. Finally, the framework is validated in simulation and real-world flight.

Occlusion Reasoning for Skeleton Extraction of Self-Occluded Tree Canopies

Chung Hee Kim, G. Kantor

SegmentationConvolutional Neural NetworkImageAgriculture Related

🎯 What it does: Proposes a method to extract the self-occluded tree crown skeleton by estimating unobserved tree structures.

Occlusion-Aware Crowd Navigation Using People as Sensors

Ye-Ji Mun, K. Driggs-Campbell

Autonomous DrivingReinforcement LearningAuto Encoder

🎯 What it does: Propose integrating social inference technology into the planning pipeline to achieve obstacle-free navigation for occluded crowds

Off-policy Imitation Learning from Visual Inputs

Zhihao Cheng, D. Tao

Robotic IntelligenceImage

🎯 What it does: Propose a framework called OPIfVI for offline policy imitation learning from visual inputs to enhance data efficiency and performance.

On Domain-Specific Pre- Training for Effective Semantic Perception in Agricultural Robotics

Gianmarco Roggiolani, Jens Behley

SegmentationConvolutional Neural NetworkImageAgriculture Related

🎯 What it does: Using self-supervised pre-training and domain-specific augmentation strategies to reduce the labeling effort in semantic perception tasks for agricultural robots

On Human Grasping and Manipulation in Kitchens: Automated Annotation, Insights, and Metrics for Effective Data Collection

Nathan Elangovan, Minas Liarokapis

Data-Centric LearningRobotic IntelligenceVideoBenchmark

🎯 What it does: Collect and analyze key attributes of human grasping and manipulation strategies in kitchen tasks, constructing a 7-hour high-definition video dataset, annotating over 10,000 activities, and using machine learning to automatically extract grasp types, hand, and object information. Subsequently, clustering algorithms are employed to identify patterns, and improved data collection strategies are proposed based on individual differences.

On Improving Boundary Quality of Instance Segmentation in Cluttered and Chaotic Scenarios

Biqi Yang, P. Heng

SegmentationImage

🎯 What it does: Proposes an IoU-based boundary-aware Mask head (IBM head) and a two-stage coarse-to-fine instance segmentation framework to improve boundary quality in cluttered scenes for instance segmentation

On Legible and Predictable Robot Navigation in Multi-Agent Environments

J. Bastarache, Stephen L. Smith

Explainability and InterpretabilityRobotic Intelligence

🎯 What it does: Proposes a robot navigation framework that can reason about its readability and predictability in dynamic interactions.

On Locally Optimal Redundancy Resolution using the Basis of the Null Space

Eugenio Monari, R. Vertechy

OptimizationRobotic Intelligence

🎯 What it does: Proposed two methods for computing null space velocity commands for redundant robots, and tested them in welding applications.

On Shortest Arc-To-Arc Dubins Path

S. Manyam, D. Casbeer

Autonomous DrivingOptimization

🎯 What it does: For a given set of tracks, the paper studies and solves the Arc-to-Arc Dubins Path (ATAD) problem, proposing an optimal solution to the problem and an algorithm to compute a tight lower bound for the Orbiting Dubins Traveling Salesman Problem (ODTSP).

On Tendon Driven Continuum Robots with Compressible Backbones

Manu Srivastava, I. Walker

Computational EfficiencyRobotic Intelligence

🎯 What it does: Studied the effect of axial backbone compression on traction-driven continuum robots, and proposed a mechanics compensation model that does not require tension sensing or knowledge of material properties, while also providing an analytical expression for the minimum prestretch required to achieve a specified bending, and implemented real-time control on low-cost hardware.

On the Impact of Interruptions During Multi-Robot Supervision Tasks

Abhinav Dahiya, Stephen L. Smith

Robotic Intelligence

🎯 What it does: A user study with 39 participants was conducted, monitoring remote mobile robots and being interrupted during the process by intrinsic (robot failure correction) or extrinsic (message task) interruptions, evaluating the impact on task performance and subjective workload.

On the Learned Balance Manifold of Underactuated Balance Robots

Feng Han, Jinmin Yi

Robotic Intelligence

🎯 What it does: Proposes a learning-based approach that uses Gaussian processes to estimate the balance equivalent model (BEM) and zero dynamics of underactuated balancing robots, and experimentally validates it on a rotational pendulum and bipedal robot.

On the programming effort required to generate Behavior Trees and Finite State Machines for robotic applications

Matteo Iovino, Christian Smith

Robotic Intelligence

🎯 What it does: Demonstrated through practical demonstrations that the modularity of Behavior Trees (BT) significantly reduces the programming workload for robot tasks compared to Finite State Machines (FSM)

On the Use of Torque Measurement in Centroidal State Estimation

Shahram Khorshidi, M. Khadiv

Robotic Intelligence

🎯 What it does: Propose to utilize joint torque measurements by projecting full-body dynamics into the zero space of contact constraints, directly linking torque with center of mass (CoM) state dynamics, and employing an extended Kalman filter (EKF) to integrate torque information into CoM state estimation;

On the Utility of Buffers in Pick-n-Swap Based Lattice Rearrangement

Kai Gao, Jingjin Yu

Optimization

🎯 What it does: Studied the utility of using multiple buffers in the Pick‑n‑Swap grid arrangement problem, and proposed a structural and algorithmic scheme to enhance solution optimality and optimize motion distance using buffers.

On-Demand Multi-Agent Basket Picking for Shopping Stores

Mattias Tiger, F. Heintz

OptimizationBenchmark

🎯 What it does: This paper formally models the real-time multi-agent basket picking problem in online shopping stores, proposes a new method, and demonstrates the ability to solve the problem in online real-time environments.

Onboard Controller Design for Nano UAV Swarm in Operator-Guided Collective Behaviors

Tugay Alperen Karagüzel, E. Ferrante

Robotic Intelligence

🎯 What it does: Designed and implemented local controllers for six Crazyflie micro-drones to enable them to perform various collective behaviors under operator guidance, including formation flight, gradient tracking, target localization, formation, and dispersion search.

One Training for Multiple Deployments: Polar-based Adaptive BEV Perception for Autonomous Driving

Huitong Yang, Yuexin Ma

Autonomous Driving

🎯 What it does: Propose an adaptive BEV perception method based on polar coordinates, enabling a single training to adapt to various deployments.

One-Shot Reachability Analysis of Neural Network Dynamical Systems

Shaoru Chen, Mahyar Fazlyab

🎯 What it does: This paper proposes a one-time reachability analysis framework for neural network dynamics systems, and proves that this method can significantly reduce error accumulation in recursive analysis;

One-shot Visual Imitation via Attributed Waypoints and Demonstration Augmentation

Matthew Chang, Saurabh Gupta

Robotic IntelligenceImage

🎯 What it does: Analyze existing techniques and design a new solution for single-example visual imitation, adopting a modular approach to separate task reasoning from task execution, and utilizing data augmentation and generation techniques to alleviate overfitting.

Online augmentation of learned grasp sequence policies for more adaptable and data-efficient in-hand manipulation

E. Gordon, Rana Soltani-Zarrin

Robotic IntelligenceReinforcement Learning

🎯 What it does: In hand tool manipulation, this paper combines discrete dynamic programming with reinforcement learning, achieving tree search and policy replay through simplified environment approximation to improve the RL learning grasping sequence policy.

Online Consistent Video Depth with Gaussian Mixture Representation

Chao Liu, Jan Kautz

Depth EstimationOptical FlowVideo

🎯 What it does: An optical flow-guided single-image depth estimation method that online estimates consistent and accurate depth for static scene video sequences.

Online Coverage Path Planning Scheme for a Size-Variable Robot

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

Robotic Intelligence

🎯 What it does: Proposed an online coverage path planning scheme applicable to robots that can change their size.

Online Hand-Eye Calibration with Decoupling by 3D Textureless Object Tracking

Li Jin, Xueying Qin

Pose EstimationRobotic Intelligence

🎯 What it does: Proposes an online hand-eye calibration method based on natural three-dimensional objects, capable of achieving automatic calibration on textureless or weakly textured objects; the method first uses a Pose Refinement Network (PR-Net) to improve the accuracy of three-dimensional object tracking, then constructs three-dimensional convergence point constraints using multi-view information to correct object positions, and finally optimizes the hand-eye pose through loop closure constraints to avoid falling into local minima.

Online Learning and Suppression of Vibration in Collaborative Robots with Power Tools

Gokhan Solak, Arash Ajoudani

Robotic Intelligence

🎯 What it does: Learn and compensate for vibrations from electric tools, generate a feedforward Cartesian force using the BMFLC algorithm to achieve vibration suppression in collaborative robots within industrial environments.

Online Non-linear Centroidal MPC for Humanoid Robots Payload Carrying with Contact-Stable Force Parametrization

Mohamed Elobaid, D. Pucci

OptimizationRobotic Intelligence

🎯 What it does: Propose a method combining online nonlinear centroid model predictive control (MPC) with contact stable force parameterization, enabling humanoid robots carrying loads to follow a predefined gait under continuous disturbances.

Online Safety Property Collection and Refinement for Safe Deep Reinforcement Learning in Mapless Navigation

Luca Marzari, A. Farinelli

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposes the CROP framework, which collects and refines online safety properties through cost signals during training to quantify the number of violations in DRL strategies.

Online Social Robot Navigation in Indoor, Large and Crowded Environments

Steven Silva, J. D. Hernández

Robotic Intelligence

🎯 What it does: Proposes an online social robot navigation framework capable of navigating and demonstrating appropriate social behaviors in indoor large-area and densely populated environments.

Online Tool Selection with Learned Grasp Prediction Models

Khashayar Rohanimanesh, Aviv Tamar

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Proposes a tool selection and planning method based on a learned grasp prediction model to maximize throughput in robotic box grasping tasks; achieves tool switching and grasping sequence planning based solely on currently visible objects through Markov Decision Processes and Model Predictive Control.

Online Update of Safety Assurances Using Confidence-Based Predictions

Kensuke Nakamura, Somil Bansal

Safty and Privacy

🎯 What it does: Proposes a method based on Hamilton-Jacobi reachability for online safety updates in human-robot interaction scenarios, utilizing confidence parameters as virtual states to compute parameterized forward reach tubes (FRT), and performing online queries to update robot planning when confidence parameters change.