ICRA 2023 Papers — Page 8
IEEE International Conference on Robotics and Automation · 1341 papers
Learning Continuous Control Policies for Information-Theoretic Active Perception
Pengzhi Yang, Nikolay A. Atanasov
Robotic IntelligenceReinforcement Learning
🎯 What it does: Proposes a continuous control strategy learning method for exploration and active feature point localization, utilizing mobile robots to detect feature points within a limited perceptual range and learning control policies by maximizing the mutual information between feature point states and sensor observations.
Learning Decoupled Multi-touch Force Estimation, Localization and Stretch for Soft Capacitive E-skin
Abu Bakar Dawood, K. Althoefer
Biomedical Data
🎯 What it does: Studied the decoupled estimation of multi-touch force, position, and overall stretching in soft capacitive e-skin, using machine learning models to predict collected sensor data and analyzed the effects of simultaneous multi-point interactions.
Learning Deposition Policies for Fused Multi-Material 3D Printing
K. Liao, Vahid Babaei
OptimizationReinforcement Learning
🎯 What it does: Learning and optimizing deposition strategies for multi-material 3D printing
Learning Depth Completion of Transparent Objects using Augmented Unpaired Data
Floris Erich, Y. Domae
Depth EstimationGenerative Adversarial Network
🎯 What it does: Proposes a technique for depth completion of transparent objects using enhanced data collected directly from complex real environments.
Learning Dexterous Manipulation from Exemplar Object Trajectories and Pre-Grasps
S. Dasari, Vikash Kumar
Robotic IntelligenceReinforcement LearningMultimodalityBenchmark
🎯 What it does: Proposes the Pre-Grasp informed Dexterous Manipulation (PGDM) framework, which achieves diverse dexterous manipulation behaviors using a single pre-grasp posture, and automatically generates the TCDM benchmark dataset to verify its effectiveness.
Learning Exploration Strategies to Solve Real-World Marble Runs
Alisa Allaire, C. Atkeson
Robotic IntelligenceMixture of ExpertsPhysics Related
🎯 What it does: Designed and evaluated a multi-random sub-policy structure based on the Mixture of Experts (MoE) strategy to address the physical problems of Marble Runs in the real world with local instability or discrete dynamics, enabling rapid adaptation to new environments.
Learning Feasibility of Factored Nonlinear Programs in Robotic Manipulation Planning
Joaquim Ortiz de Haro, Marc Toussaint
OptimizationRobotic IntelligenceGraph Neural NetworkGraph
🎯 What it does: Propose a graph neural network architecture to predict which variables and constraints are infeasible together in decomposed nonlinear programs, evaluated in robotic manipulation planning.
Learning Food Picking without Food: Fracture Anticipation by Breaking Reusable Fragile Objects
Rinto Yagawa, Hideo Saito
Robotic Intelligence
🎯 What it does: Train a robot to perform food picking without consuming real food by utilizing fracture experience with reusable fragile objects (e.g., wood blocks, ping pong balls, jellies) and domain generalization techniques, and test its performance in real robot experiments.
Learning from Physical Human Feedback: An Object-Centric One-Shot Adaptation Method
Alvin C M Shek, Changliu Liu
Reinforcement Learning from Human FeedbackReinforcement LearningPhysics Related
🎯 What it does: Propose an object preference-based one-time adaptation method (OPA), which pre-trains a baseline policy and performs online updates after a single human intervention to align with human feedback.
Learning Generalizable Pivoting Skills
Xiang Zhang, Diego Romeres
Robotic IntelligenceSupervised Fine-TuningReinforcement LearningImage
🎯 What it does: Propose a three-step framework for learning robust and generalizable robot placement skills: ① Learn placement strategies on a single object through reinforcement learning; ② Use supervised learning to obtain an object feature space, encoding kinematic properties of arbitrary objects; ③ Learn data-driven projection based on object features to adapt state and action spaces, transferring a single strategy to multi-object scenarios.
Learning Height for Top-Down Grasps with the DIGIT Sensor
Thais Bernardi, S. Ivaldi
Robotic IntelligenceImage
🎯 What it does: Learn and employ a regression model to predict grasp height based on images, aiming to improve the success rate of grasping unknown objects.
Learning Low-Frequency Motion Control for Robust and Dynamic Robot Locomotion
Siddhant Gangapurwala, I. Havoutis
Robotic IntelligenceReinforcement Learning
🎯 What it does: Robust and dynamic walking was achieved on the real ANYmal C quadruped robot using a learned low-frequency controller (minimum 8 Hz), enabling high heading speeds of 1.5 m/s, navigating uneven terrain, and resisting sudden external disturbances.
Learning Modular Robot Visual-motor Locomotion Policies
Julian Whitman, H. Choset
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: Developed a generic visual-motor control strategy applicable across different module designs and environments, training modular policies within a deep reinforcement learning framework to enable multi-module robots to walk in more challenging environments using vision.
Learning Neuro-symbolic Programs for Language Guided Robot Manipulation
Namasivayam Kalithasan, Rohan Paul
Robotic IntelligenceVision-Language-Action ModelImageText
🎯 What it does: Train a model to generate executable robot operation programs from natural language instructions and input scenes; use neuro-symbolic programs to achieve language-guided robot manipulation.
Learning on the Job: Self-Rewarding Offline-to-Online Finetuning for Industrial Insertion of Novel Connectors from Vision
Ashvin Nair, S. Levine
Representation LearningRobotic IntelligenceReinforcement LearningImage
🎯 What it does: Proposes a self-rewarding offline-to-online fine-tuning method, where a learned reward function helps robots adapt to unseen connectors in vision-based industrial insertion tasks.
Learning Perception-Aware Agile Flight in Cluttered Environments
Yunlong Song, D. Scaramuzza
Computational EfficiencyKnowledge DistillationRobotic IntelligenceReinforcement LearningImage
🎯 What it does: Proposed a learning system that first trains a perceptual teacher policy using reinforcement learning, then distills its knowledge into a student policy based solely on vision, achieving flight completion in crowded environments with minimal time; the system realizes tight coupling between perception and control.
Learning Perceptual Hallucination for Multi-Robot Navigation in Narrow Hallways
J. Park, P. Stone
GenerationRobotic Intelligence
🎯 What it does: Proposes a method called Perceptual Hallucination for Hallway Passing (PHHP), enabling two robots to pass through a narrow corridor without collision, stopping, or waiting.
Learning Personalised Human Sit-to-Stand Motion Strategies via Inverse Musculoskeletal Optimal Control
Daniel F. N. Gordon, S. Vijayakumar
OptimizationRobotic IntelligenceBiomedical Data
🎯 What it does: Learning individualized sit-to-stand movement strategies through inverse musculoskeletal optimal control
Learning Pre-Grasp Manipulation of Flat Objects in Cluttered Environments using Sliding Primitives
Jiaxi Wu, Yinlin Li
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper studies pre-grasping operations for grasping thin planar objects in cluttered environments, proposing to model the task as a parameterized action Markov decision process and introducing sliding primitives for control based on deep reinforcement learning;
Learning Responsibility Allocations for Safe Human-Robot Interaction with Applications to Autonomous Driving
Ryan K. Cosner, M. Pavone
Autonomous Driving
🎯 What it does: Propose a responsibility-aware control barrier function (RA-CBF) and develop a data-based responsibility allocation learning method, integrating safety-critical control with learning techniques to achieve scenario-dependent responsibility allocation and synthesize safe and efficient driving behaviors, followed by testing and validation on real-world driving data.
Learning Reward Functions for Robotic Manipulation by Observing Humans
Minttu Alakuijala, C. Schmid
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningContrastive LearningVideo
🎯 What it does: Learning a general reward function using unlabeled human video data to guide the learning of robot manipulation tasks.
Learning Risk-Aware Costmaps via Inverse Reinforcement Learning for Off-Road Navigation
S. Triest, S. Scherer
Autonomous DrivingRobotic IntelligenceReinforcement Learning
🎯 What it does: Trained a deep cost function using inverse reinforcement learning to achieve an uncertainty-aware aerial cost map for aerial navigation;
Learning Robotic Cutting from Demonstration: Non-Holonomic DMPs using the Udwadia-Kalaba Method
Arturas Straivzys, S. Ramamoorthy
OptimizationRobotic Intelligence
🎯 What it does: Introduce a coupling term into the DMP framework to satisfy predefined non-homogeneous constraints, enabling the learning of tasks such as robotic tool cutting.
Learning Sim-to-Real Dense Object Descriptors for Robotic Manipulation
Hoang-Giang Cao, I-Chen Wu
Domain AdaptationRobotic IntelligenceImage
🎯 What it does: Proposes Sim-to-Real Dense Object Nets (SRDONs) and a cross-domain image object-to-object matching method to achieve dense object descriptors in visual robots from simulation to real-world scenarios.
Learning Stabilization Control from Observations by Learning Lyapunov-like Proxy Models
Milan Ganai, Sicun Gao
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a learning method for an intermediate agent model based on Lyapunov stability theory, employing a two-step training process: first, learning a Lyapunov-like terrain agent model from expert state sequences; subsequently, using this agent model as a reward function to guide the learner's policy, achieving stable learning in continuous control problems.
Learning Stable Dynamics via Iterative Quadratic Programming
Paul Gesel, M. Begum
OptimizationRobotic Intelligence
🎯 What it does: Proposes a learning stable dynamics (LSD-IQP) controller based on iterative quadratic programming, which learns trajectories from demonstrations and learns energy functions and automatic dynamic systems (ADS) through semi-infinite quadratic programming, applying constraints on ADS to ensure convergence to a single target position.
Learning State Conditioned Linear Mappings for Low-Dimensional Control of Robotic Manipulators
Michael Przystupa, Martin Jagersand
Robotic IntelligenceAuto Encoder
🎯 What it does: Proposed and verified a state-conditioned linear mapping to achieve linear control of high-dimensional robot actuators using a low-dimensional action space.
Learning Tethered Perching for Aerial Robots
Fabian Hauf, M. Kovač
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose an algorithm for automatically generating trajectories for a drone to hover on branches, utilizing a tethered hovering mechanism and a pendulum-like structure to achieve energy optimization, controlled 180° flips, and safe inverted landing.
Learning to Estimate 3-D States of Deformable Linear Objects from Single-Frame Occluded Point Clouds
Kangchen Lv, Xiang Li
Pose EstimationPoint Cloud
🎯 What it does: Robustly estimating the 3D state of deformable linear objects from single-frame occluded point clouds
Learning to Explore Informative Trajectories and Samples for Embodied Perception
Ya Jing, Tao Kong
Robotic IntelligenceReinforcement Learning
🎯 What it does: By constructing a 3D semantic distribution map and self-supervisedly training exploration strategies using semantic distribution inconsistency and uncertainty rewards, followed by selecting hard samples from the trajectory based on semantic distribution uncertainty to enhance perception models.
Learning to Forecast Aleatoric and Epistemic Uncertainties over Long Horizon Trajectories
Aastha Acharya, Nisar Ahmed
Safty and PrivacyRobotic IntelligenceWorld Model
🎯 What it does: Use a learned world model to predict the complete trajectory of autonomous agents over long time spans and estimate their uncertainty.
Learning to Influence Vehicles' Routing in Mixed-Autonomy Networks by Dynamically Controlling the Headway of Autonomous Cars
Xiaoyu Ma, Negar Mehr
Autonomous DrivingReinforcement Learning
🎯 What it does: In mixed automated networks, a method is proposed to influence vehicle path selection by dynamically controlling the inter-vehicle distance of autonomous vehicles, thereby reducing congestion, and training reinforcement learning strategies to achieve this goal.
Learning to Predict Action Feasibility for Task and Motion Planning in 3D Environments
Smail Ait Bouhsain, T. Siméon
Robotic IntelligenceSupervised Fine-Tuning
🎯 What it does: Propose a learning method for 3D environments that uses deep neural networks to predict the probability of action feasibility and integrates it into task and motion planning (TAMP) algorithms to reduce the number of geometric planning calls.
Learning to View: Decision Transformers for Active Object Detection
Wenhao Ding, Arnie Sen
Object DetectionRobotic IntelligenceTransformerSupervised Fine-TuningReinforcement LearningImage
🎯 What it does: This paper proposes an active object detection method that combines Decision Transformer with online fine-tuning, using reinforcement learning to control robots to collect perspective images that maximize detection quality.
Learning to Walk by Steering: Perceptive Quadrupedal Locomotion in Dynamic Environments
Mingyo Seo, Yuke Zhu
Reinforcement Learning
🎯 What it does: Proposed a hierarchical learning framework named PRELUDE to address the perception-based locomotion problem of quadruped robots in dynamic environments, dividing the task into high-level decision-making for generating navigation instructions and low-level gait generation to execute instructions.
Learning Tool Morphology for Contact-Rich Manipulation Tasks with Differentiable Simulation
Mengxi Li, Jeannette Bohg
OptimizationRobotic IntelligencePhysics Related
🎯 What it does: Propose an end-to-end framework that utilizes differentiable physics simulation to automatically learn tool morphology for contact-rich manipulation tasks;
Learning Video-Conditioned Policies for Unseen Manipulation Tasks
Elliot Chane-Sane, I. Laptev
Robotic IntelligenceVision-Language-Action ModelVideoSequential
🎯 What it does: This study proposes Video-conditioned Policy learning (ViP), which achieves robot control over unseen tasks by mapping human demonstration videos from natural environments to robot manipulation skills.
Learning Visual Locomotion with Cross-Modal Supervision
Antonio Loquercio, J. Malik
Robotic IntelligenceImageMultimodality
🎯 What it does: Learned a visual locomotion policy using only a monocular RGB camera and proprioception.
Learning Visual-Audio Representations for Voice-Controlled Robots
Peixin Chang, K. Driggs-Campbell
Representation LearningRobotic IntelligenceReinforcement LearningImageMultimodalityAudio
🎯 What it does: Propose a task-oriented speech-controlled robot pipeline that leverages visual-audio representation learning, first learning a visual-audio representation (VAR) that associates images with speech commands, and subsequently using rewards generated by VAR through reinforcement learning to execute speech commands.
Learning-Based Dimensionality Reduction for Computing Compact and Effective Local Feature Descriptors
Hao Dong, C. Stachniss
RetrievalCompressionImage
🎯 What it does: Utilize a lightweight multilayer perceptron (MLP) to low-dimensionalize local feature descriptors, enhancing descriptor quality while reducing storage and computational costs, and conduct a comprehensive analysis across unsupervised, semi-supervised, and supervised settings; evaluate on tasks including visual localization, patch verification, image matching, and retrieval.
Learning-based Initialization of Trajectory Optimization for Path-following Problems of Redundant Manipulators
Minsung Yoon, Sung-Eui Yoon
OptimizationRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposes a learning-based initial trajectory generation method for trajectory optimization of redundant manipulators; the method generates high-quality initial trajectories within a short time budget using example-guided reinforcement learning and introduces zero-space projection imitation rewards to consider zero-space constraints, learning feasible motions from expert demonstrations;
Learning-based Relational Object Matching Across Views
Cathrin Elich, Joerg Stueckler
RetrievalGraph Neural NetworkImage
🎯 What it does: Propose a learning-based method for cross-view object detection matching in RGB images by combining local keypoints with novel object-level features.
Learning-based Uncertainty-aware Navigation in 3D Off-Road Terrains
Ho-Joo Lee, Cheolhyeon Kwon
Autonomous DrivingOptimizationRepresentation Learning
🎯 What it does: A safe, efficient, and agile 3D off-road terrain vehicle navigation algorithm was developed. The algorithm learns terrain uncertainty from driving data, encodes the learned uncertainty distribution into passage cost for path evaluation, and ultimately achieves safe and flexible vehicle mobility by optimizing uncertainty-aware passage cost.
Legs as Manipulator: Pushing Quadrupedal Agility Beyond Locomotion
Xuxin Cheng, Deepak Pathak
Robotic IntelligenceReinforcement Learning
🎯 What it does: Train quadruped robots to perform tasks such as wall climbing, button pressing, and object interaction using their front limbs on the basis of locomotion, and combine locomotion and manipulation skills into long-term task planning through behavior trees;
LEMURS: Learning Distributed Multi-Robot Interactions
Eduardo Sebastián, C. Sagüés
Robotic IntelligenceTransformerOrdinary Differential Equation
🎯 What it does: Proposed the LEMURS algorithm, which learns scalable multi-robot control strategies through collaborative task demonstration learning.
LES: Locally Exploitative Sampling for Robot Path Planning
S. Joshi, P. Tsiotras
OptimizationRobotic Intelligence
🎯 What it does: Propose an optimization-based sampling method that generates new sample points by improving the cost-to-start values within the neighborhood, thereby incorporating exploitation tendencies into robot path planning.
Leveraging Scene Embeddings for Gradient-Based Motion Planning in Latent Space
Jun Yamada, I. Posner
OptimizationRepresentation LearningRobotic Intelligence
🎯 What it does: Model motion planning as an optimization problem in a structured latent space, leveraging scene embeddings and robot generative models to drive the optimization, while introducing efficient collision checking for regularization.
LGCNet: Feature Enhancement and Consistency Learning Based on Local and Global Coherence Network for Correspondence Selection
Tzu-Han Wu, Kuan-Wen Chen
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposes using local and global handcrafted coherent features to enhance initial features in correspondence selection, followed by searching for neighbors in both coordinate space and feature space, and introducing a new neighbor representation and fusion architecture to improve the discriminative ability of correspondences and remove obvious outliers.
Lidar Augment: Searching for Scalable 3D LiDAR Data Augmentations
Zhaoqi Leng, Mingxing Tan
Object DetectionHyperparameter SearchData-Centric LearningConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposed a scalable data augmentation strategy called LidarAugment for 3D LiDAR point cloud object detection, significantly simplifying hyperparameter configuration and improving model performance.
LiDAR-based Indoor Localization with Optimal Particle Filters using Surface Normal Constraints
Heruka Andradi, P. Plöger
OptimizationSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposed a LiDAR-based indoor localization system that utilizes an optimal particle filter along with high-precision and low-precision observation models to address particle degradation and map discrepancy issues.
LiDAR-SGM: Semi-Global Matching on LiDAR Point Clouds and Their Cost-Based Fusion into Stereo Matching
Bianca Forkel, Hans-Joachim Wuensche
Depth EstimationAutonomous DrivingImageMultimodalityPoint Cloud
🎯 What it does: Proposes an efficient method to integrate stereo matching and LiDAR at the Semi-Global Matching (SGM) cost level
Light-Weight Pointcloud Representation with Sparse Gaussian Process
Mahmoud Ali, Lantao Liu
CompressionPoint Cloud
🎯 What it does: Propose a framework that compresses high-fidelity point cloud sensor observations into a compact form using sparse Gaussian processes to achieve efficient communication and storage.
Lighthouses and Global Graph Stabilization: Active SLAM for Low-compute, Narrow-FoV Robots
Mohit Deshpande, James C. Zamiska
Robotic IntelligenceSimultaneous Localization and Mapping
🎯 What it does: Proposes a SLAM-aware exploration strategy (SAE) based on the lighthouse concept, by using high-information panoramic views in the local neighborhood to maintain map stability, and ultimately achieving stabilization of the global pose graph
Lightweight Monocular Depth Estimation via Token-Sharing Transformer
Dong-Jae Lee, Junmo Kim
Depth EstimationTransformerImage
🎯 What it does: Proposes a Token-Sharing Transformer (TST) architecture for lightweight monocular depth estimation, specifically optimized for embedded devices.
Limit Cycle Generation with Pneumatically Driven Physical Reservoir Computing
Hiroaki Shinkawa, Kenji Kawashima
Robotic IntelligenceRecurrent Neural NetworkTime SeriesPhysics Related
🎯 What it does: Studied the use of a pneumatic pipe system as a physical recurrent computation resource, and by feeding back estimated information into the system, it generates stable limit cycle pressure variations similar to walking;
Linear Auto-calibration of Pan-Tilt-Zoom Cameras With Rotation Center Offset
Yu Liu, Hui Zhang
Image
🎯 What it does: Linear auto-calibration of PTZ cameras by utilizing the offset between the camera center and the rotation center to recover all intrinsic parameters.
Linear Delta Arrays for Compliant Dexterous Distributed Manipulation
Sarvesh Patil, F. Z. Temel
Robotic IntelligenceReinforcement Learning
🎯 What it does: Designed and implemented a distributed compliant manipulator array consisting of 64 linear-driven Delta robots, demonstrating its capabilities in various distributed manipulation tasks such as planar, vertical, and grasping.
Loc-NeRF: Monte Carlo Localization using Neural Radiance Fields
Dominic Maggio, L. Carlone
Robotic IntelligenceNeural Radiance FieldImage
🎯 What it does: Proposes Loc-NeRF, a real-time visual robot localization method that integrates Monte Carlo localization with neural radiance fields (NeRF), utilizing a pre-trained NeRF as the map and achieving real-time localization using only RGB cameras.
Local Layer Splitting: An Additive Manufacturing Method to Define the Mechanical Properties of Soft Pneumatic Actuators During Fabrication
Brice Parilusyan, Marcos Serrano
Robotic Intelligence
🎯 What it does: Proposed and implemented a local layer slicing (LLS) method, defining the mechanical properties of silicone components by locally adjusting layer thickness during the printing process, evaluated the effect of layer thickness on stiffness through tensile tests, developed a custom slicer to generate G-code with localized layer thickness variations, fabricated and tested soft pneumatic actuators (SPA) using the LLS process, and compared their bending and mechanical performance with state-of-the-art SPA.
Local Neural Descriptor Fields: Locally Conditioned Object Representations for Manipulation
Ethan Chun, L. Kaelbling
Robotic Intelligence
🎯 What it does: Developed a method utilizing a local geometry neural descriptor field (L-NDF) to generalize object manipulation skills acquired from limited demonstrations to new objects in unseen shape categories.
Local_INN: Implicit Map Representation and Localization with Invertible Neural Networks
Zirui Zang, Rahul Mangharam
Robotic IntelligenceFlow-based ModelSimultaneous Localization and Mapping
🎯 What it does: Propose the Local_INN framework, which utilizes invertible neural networks to achieve implicit map representation and inverse localization, and outputs robot poses with covariance through latent space sampling to realize uncertainty estimation.
Locate before Segment: Topology-guided Retinal Layer Segmentation in Optical Coherence Tomography Images
Ye Lu, Max Q.-H. Meng
SegmentationBiomedical Data
🎯 What it does: Proposes the Locate-to-Segment (L2S) framework, which first predicts the location of each layer region by using the Structured Boundary Regression Network (SBRNet) to estimate the surface positions of layers, and then uses the localization results as additional input to guide the pixel-level segmentation network, achieving retinal layer segmentation with correct topology and smooth boundaries.
LODE: Locally Conditioned Eikonal Implicit Scene Completion from Sparse LiDAR
Pengfei Li, Ya-Qin Zhang
Autonomous DrivingPoint Cloud
🎯 What it does: Proposed a sparse LiDAR scene completion method based on a local conditional Eikonal implicit representation
Loitering and Trajectory Tracking of Suspended Payloads in Cable-Driven Balloons Using UGVs
Julius Wanner, M. Gharib
OptimizationRobotic IntelligencePhysics Related
🎯 What it does: Designed and implemented a collaborative control robot platform that uses multiple unmanned ground vehicles (UGVs) to jointly drive ropes for controlling balloons and suspended loads, and validated the load trajectory tracking performance through simulations and experimental tests using ground prototypes and dynamic models.
Loosely-coupled localization fusion system based on track-to-track fusion with bias alignment
Soyeong Kim, K. Jo
Autonomous Driving
🎯 What it does: Proposed a loosely coupled localization fusion system based on trajectory-to-trajectory (T2T) fusion with bias alignment, first estimating the slow drift biases between different localization systems and aligning them, then using the aligned estimates for fusion;
Lossless SIMD Compression of LiDAR Range and Attribute Scan Sequences
J. Ford, Jordan Ford
CompressionAutonomous DrivingPoint Cloud
🎯 What it does: Developed a fast lossless compression algorithm for LiDAR range and attribute scan sequences
Low-level controller in response to changes in quadrotor dynamics
JaeKyung Cho, Seong-Woo Kim
Robotic IntelligenceRecurrent Neural NetworkReinforcement Learning
🎯 What it does: Train a low-level controller to respond instantly to changes in quadrotor dynamics without requiring prior knowledge or parameter tuning.
M-EMBER: Tackling Long-Horizon Mobile Manipulation via Factorized Domain Transfer
Bohan Wu, Li Fei-Fei
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose the M-EMBER method, which decomposes long-term mobility operations into a series of primitive visual skills, achieving kitchen cleaning tasks on a real robot after reinforcement learning in simulation.
Magnetic Ball Chain Robots for Endoluminal Interventions
G. Pittiglio, P. Dupont
Robotic Intelligence
🎯 What it does: Designed and verified a super-redundant robot composed of a chain of permanently magnetized spheres enclosed in a cylindrical polymer sheath, capable of controllable bending with extremely small curvature radius through external magnetic field manipulation, suitable for endoscopic surgical navigation.
MagTac: Magnetic Six-Axis Force/Torque Fingertip Tactile Sensor for Robotic Hand Applications
Sungwoo Park, Donghyun Hwang
Robotic IntelligencePhysics Related
🎯 What it does: Developed a six-axis force/torque tactile sensor based on the Hall effect and integrated it into the mechanical fingertip
Mapping Waves with an Uncrewed Surface Vessel via Gaussian Process Regression
T. Sears, J. Marshall
Robotic IntelligenceTime SeriesPhysics Related
🎯 What it does: Collect low-cost inertial measurement unit (IMU) data using an unmanned surface vessel (USV) along the coast of Lake Ontario in Canada, and construct a spatiotemporal wave model through Gaussian process regression (GPR) during the offline phase.
Mask3D: Mask Transformer for 3D Semantic Instance Segmentation
Jonas Schult, Bastian Leibe
SegmentationTransformerPoint Cloud
🎯 What it does: Propose Mask3D, a Transformer-based 3D semantic instance segmentation method that directly predicts instance masks in point clouds using instance queries and multi-scale attention.
Mechanical Intelligence for Prehensile In-Hand Manipulation of Spatial Trajectories
Qiujie Lu, Nicolás Rojas
Robotic Intelligence
🎯 What it does: Proposed a helical grasping intrinsic motion technology for aerial manipulation based on mechanical intelligence, generating complex spatial trajectories under open-loop control using a minimal number of actuators and low-level non-configuration modes to achieve helical grasping of deformable objects.
Memory-based Exploration-value Evaluation Model for Visual Navigation
Yongquan Feng, Wenjing Yang
Autonomous DrivingReinforcement Learning
🎯 What it does: Proposes a hierarchical visual navigation scheme called Memory-based Exploration-value Evaluation Model (MEEM), which addresses the sparse reward problem using a hierarchical strategy, stores agent's historical information through episodic memory, and calculates exploration value for action planning at each position in the observable area via an exploration value evaluation model.
Meta-Learning-Based Optimal Control for Soft Robotic Manipulators to Interact with Unknown Environments
Zhiqiang Tang, Cecilia Laschi
OptimizationRobotic IntelligenceMeta Learning
🎯 What it does: A meta-learning based optimal control framework is proposed, which utilizes goal-oriented active search to collect environment-specific data. It employs meta-learning to train data-driven probabilistic model priors and updates them online, ultimately achieving simultaneous target position and contact force control for soft robots in unknown environments through model-based optimal control strategies.
Meta-Reinforcement Learning via Language Instructions
Zhenshan Bing, Alois Knoll
Robotic IntelligenceMeta LearningReinforcement LearningVision-Language-Action ModelTextBenchmark
🎯 What it does: Developed and evaluated a meta-reinforcement learning algorithm that leverages language instructions for robot multi-task manipulation.
METEOR: A Dense, Heterogeneous, and Unstructured Traffic Dataset with Rare Behaviors
Rohan Chandra, Dinesh Manocha
Object DetectionObject TrackingAutonomous DrivingVideoMultimodalityBenchmark
🎯 What it does: Released and constructed a new traffic dataset named Meteor, collecting over 1000 one-minute videos, more than 2 million frames with annotated bounding boxes and GPS trajectories, covering 16 different traffic agents, totaling over 13 million bounding boxes; this dataset focuses on rare multi-agent driving behaviors, including traffic violations, atypical interactions, and diverse scenarios, and assigns multi-dimensional contextual labels such as weather, time, road conditions, and traffic density for each video.
Mimicking Real Forces on a Drone Through a Haptic Suit to Enable Cost-Effective Validation
Carl Hildebrandt, Sebastian G. Elbaum
Robotic Intelligence
🎯 What it does: Propose a low-cost, easy-to-deploy, compact framework that uses a force feedback suit (equipped with directional propellers) to simulate external forces acting on a drone, along with a synthesizer and controller that convert target forces into suit directional propeller commands; experimentally evaluate performance under various force fields.
Minimally Constrained Multi-Robot Coordination with Line-of-Sight Connectivity Maintenance
Yupeng Yang, Wenhao Luo
OptimizationRobotic Intelligence
🎯 What it does: Studied how a multi-robot team can execute multiple behaviors while maintaining global and subgroup line-of-sight connectivity with minimal interference to existing behaviors.
Minimizing Human Assistance: Augmenting a Single Demonstration for Deep Reinforcement Learning
Abraham George, A. Farimani
Data-Centric LearningReinforcement Learning
🎯 What it does: By leveraging a single example collected via VR, augmenting it to generate a large number of human-like demonstrations, and combining them with DDPG+HER to train RL agents, thus significantly accelerating training and solving tasks that cannot be completed using DDPG+HER alone.
Mixed Observable RRT: Multi-Agent Mission-Planning in Partially Observable Environments
Kasper Johansson, A. Ames
Optimization
🎯 What it does: Propose a hybrid observable RRT method for multi-agent task planning in partially observable environments, combining hidden Markov models for state estimation and dynamic programming for path selection, ultimately integrating high-level planning with model predictive control for experimental verification.
MMIC-I: A Robotic Platform for Assembly Integration and Internal Locomotion through Mechanical Meta-Material Structures
Olivia Formoso, Kenneth C. Cheung
Robotic Intelligence
🎯 What it does: Designed and implemented the Mobile Mechanical Metamaterial Internal Integrator (MMIC-I) to achieve alignment and fixation of modular structural units in space.
MMRDN: Consistent Representation for Multi-View Manipulation Relationship Detection in Object-Stacked Scenes
Han Wang, Nanning Zheng
RecognitionImagePoint Cloud
🎯 What it does: Proposed a multi-view fusion framework called MMRDN for operational relationship detection in multi-view object stacking scenarios.
Mobility Analysis of Screw-Based Locomotion and Propulsion in Various Media
Jason Lim, Michael C. Yip
Robotic IntelligencePhysics Related
🎯 What it does: Conducted systematic experiments and performance analysis on helical movement in various media.
Model Based Position Control of Soft Hydraulic Actuators
M. Runciman, G. Mylonas
Physics Related
🎯 What it does: Built a port-Hamiltonian dynamic model for soft hydraulic actuators resistant to disturbances, and designed an energy shaping control algorithm considering fluid pressure dynamics; simultaneously introduced a nonlinear observer to compensate for unknown external forces.
Model Predictive Optimized Path Integral Strategies
Dylan M. Asmar, M. Kochenderfer
OptimizationReinforcement Learning
🎯 What it does: Rewrite MPPI as a single joint distribution across control sequences and introduce adaptive importance sampling to improve sampling efficiency
Model- and Acceleration-based Pursuit Controller for High-Performance Autonomous Racing
Jonathan Becker, Michele Magno
Autonomous DrivingOptimization
🎯 What it does: Designed and verified a model-based and acceleration-based tracking controller (MAP) for high-speed autonomous racing trajectory tracking.
Model-Agnostic Multi-Agent Perception Framework
Weizhe (Wesley) Chen, Jiaqi Ma
Object DetectionAutonomous Driving
🎯 What it does: Propose a model-agnostic multi-agent perception framework that reduces the negative impacts caused by model differences through a confidence calibrator and bounding box aggregation algorithm.
Model-Based Pose Estimation of Steerable Catheters under Bi-Plane Image Feedback
J. Lawson, N. Simaan
Pose EstimationBiomedical Data
🎯 What it does: Estimate the bending plane of a small guidewire using dual-plane image feedback and compare two error minimization strategies.
Model-Mediated Teleoperation for Remote Haptic Texture Sharing: Initial Study of Online Texture Modeling and Rendering
Mudassir Ibrahim Awan, Seokhee Jeon
Robotic IntelligenceTime Series
🎯 What it does: Proposed and implemented the first model-mediated teleoperation (MMT) framework for sharing surface tactile textures, which constructs and updates a local texture simulation model using high-frequency acceleration signals collected from the follower side, and achieves latency-free, stable, and precise texture feedback through local simulation on the host side.
Modeling and Inertial Parameter Estimation of Cart-like Nonholonomic Systems Using a Mobile Manipulator
Sergio Aguilera, Seth Hutchinson
OptimizationRobotic IntelligencePhysics Related
🎯 What it does: This paper derives the dynamics model of a shopping cart with non-homogeneous constraints using the restricted Euler-Lagrange formula, and designs a Kalman filter based on this model to estimate the inertial parameters of the truck.
Modeling of a Robotic Transcatheter Delivery System
Namrata U. Nayar, J. Desai
Robotic Intelligence
🎯 What it does: Fabricated a 4-degree-of-freedom robotic conduit, derived the relationship between tendon displacement and joint angles, and established a tendon tension model and tendon elongation (TE) model for controlling joint movements.
Modular and Parallelizable Multibody Physics Simulation via Subsystem-Based ADMM
Jeongmin Lee, Dongjun Lee
OptimizationPhysics Related
🎯 What it does: Proposed a multi-body physics simulation framework based on subsystem structure and ADMM
Modular Multi-axis Elastic Actuator with Torque Sensing Capable p-CFH for Highly Impact Resistive Robot Leg
Youngrae Kim, Dong-Woo Yun
Robotic Intelligence
🎯 What it does: Proposed a multi-axis elastic actuator (MAEA) with multi-axis compliance and the ability to measure torque without additional encoders, and analyzed the 6-axis stiffness of paired cross flexible hinges (p-CFH) from small deformation to large deformation; verified the accuracy of the stiffness analysis through finite element analysis (FEA) and experiments; achieved torque measurement and feedback torque control based on this analysis; finally integrated the MAEA into a robot leg and demonstrated its multi-axis impact resistance performance through landing experiments at different angles.
MOFT: Monocular odometry based on deep depth and careful feature selection and tracking
Karlo Koledic, Ivan Petrović
Pose EstimationDepth EstimationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkVideo
🎯 What it does: Propose a monocular odometry framework based on depth prediction and precise feature selection and tracking, capable of achieving lightweight frame-to-frame estimation and generating scale-consistent trajectories;
Moment-Based Kalman Filter: Nonlinear Kalman Filtering with Exact Moment Propagation
Y. Shimizu, Shinpei Kato
Optimization
🎯 What it does: Developed a novel nonlinear filter called Moment-based Kalman Filter (MKF), which utilizes the exact matrix propagation method to achieve precise matrix calculations of the probability distribution of random variables.
Mono-STAR: Mono-Camera Scene-Level Tracking and Reconstruction
Haonan Chang, Abdeslam Boularias
Object TrackingDepth EstimationOptimizationOptical FlowVideo
🎯 What it does: Real-time monocular 3D reconstruction system supporting semantic fusion, fast motion tracking, non-rigid object deformation, and topological changes.
Monocular Reactive Collision Avoidance for MAV Teleoperation with Deep Reinforcement Learning
Raffaele Brilli, G. Costante
Robotic IntelligenceConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A new teleoperation setup is proposed, where the operator only provides desired velocity and direction signals. Subsequently, an end-to-end deep reinforcement learning model computes control commands based on monocular RGB images and the robot's current position, achieving desired trajectory tracking and obstacle avoidance.
Monocular Simultaneous Localization and Mapping using Ground Textures
Kyle M. Hart, D. Martinez
Robotic IntelligenceSimultaneous Localization and MappingImage
🎯 What it does: Proposes a complete monocular ground texture-based SLAM system that does not require pre-existing maps. It detects image keypoints, projects them onto the ground plane, uses visual Bag-of-Words and threshold parameters to identify overlapping images and revisited areas, and employs a robust M-estimator to estimate transformation between robot poses, generating a map for navigation.
Monocular Visual-Inertial Depth Estimation
Diana Wofk, V. Koltun
Depth EstimationSimultaneous Localization and MappingMultimodality
🎯 What it does: Proposed a visual-inertial depth estimation pipeline that integrates monocular depth estimation with visual inertial odometry to generate dense depth estimates with metric scale.