ICRA 2024 Papers — Page 9
IEEE International Conference on Robotics and Automation · 1760 papers
Interactive Continual Learning Architecture for Long-Term Personalization of Home Service Robots
Ali Ayub, Kerstin Dautenhahn
Robotic IntelligenceMeta LearningReinforcement Learning from Human Feedback
🎯 What it does: An interactive continual learning architecture was developed, enabling home service robots to continuously learn and reason about environmental semantic knowledge through human-robot interaction, with a two-month system evaluation conducted in a laboratory environment using a mobile manipulator robot.
Interactive Navigation in Environments with Traversable Obstacles Using Large Language and Vision-Language Models
Zhen Zhang, K. W. S. Au
TransformerLarge Language ModelVision Language ModelVision-Language-Action ModelTextPoint Cloud
🎯 What it does: Designed and verified an interactive navigation framework based on a large language model (GPT-3.5) and a vision-language model (Grounding DINO), achieving an end-to-end process from text instructions to feasible paths.
Interactive Planning Using Large Language Models for Partially Observable Robotic Tasks
Lingfeng Sun, Diego Romeres
Robotic IntelligenceLarge Language ModelSupervised Fine-TuningChain-of-Thought
🎯 What it does: This study proposes an interactive planning method using a large language model (LLM) for partially observable robot tasks, where the robot collects missing environmental information and uses the LLM to infer states, guiding the robot to perform required actions.
InterCoop: Spatio-Temporal Interaction Aware Cooperative Perception for Networked Vehicles
Wentao Wang, Guang Tan
Autonomous DrivingGraph Neural NetworkTime Series
🎯 What it does: Proposes a new collaborative perception method that integrates road topology and the trajectory history of neighboring vehicles, learns interaction scores for each vehicle, and selectively fuses perception data based on these scores.
InterRep: A Visual Interaction Representation for Robotic Grasping
Yu Cui, Jiming Chen
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose the InterRep interactive representation method and use deep reinforcement learning to learn grasping strategies.
Intraoperatively Iterative Hough Transform Based In-plane Hybrid Control of Arterial Robotic Ultrasound for Magnetic Catheterization
Zhengyang Li, Qingsong Xu
Robotic IntelligenceBiomedical DataUltrasound
🎯 What it does: Implemented planar hybrid control and ultrasound tracking based on the iterative Hough transform during internal surgery, completing catheter guidance experiments with a magnetic catheter on an in vitro soft tissue phantom.
Intrinsic Language-Guided Exploration for Complex Long-Horizon Robotic Manipulation Tasks
Eleftherios Triantafyllidis, Zhibin Li
Robotic IntelligenceTransformerLarge Language ModelReinforcement Learning
🎯 What it does: The IGE-LLMs framework is proposed for complex long-horizon robotic manipulation tasks by using large language models (LLMs) as internal rewards to guide reinforcement learning (RL) exploration.
Introducing CEA-IMSOLD: an Industrial Multi-Scale Object Localization Dataset
Boris Meden, Steve Bourgeois
Object DetectionImageBenchmark
🎯 What it does: Proposed the industrial multi-scale object localization dataset CEA-IMSOLD, providing images at different observation distances and giving baseline results.
Introducing the Carpal-Claw: a Mechanism to Enhance High-Obstacle Negotiation for Quadruped Robots
V. Barasuol, C. Semini
Robotic Intelligence
🎯 What it does: Designed and verified the Carpal-Claw mechanism to enhance the ability of quadroped robots to traverse high obstacles
Investigation on the multi-solution problem of the kinetostatics of cable-driven continuum manipulators
Yicheng Dai, Han Yuan
OptimizationRobotic IntelligencePhysics Related
🎯 What it does: Studied the dynamic multi-solution problem of cable-driven continuum manipulators, proposed a dynamic model, and conducted simulations using interval analysis and floating-point optimization algorithms under the same driving force and external load.
IPC: Incremental Probabilistic Consensus-based Consistent Set Maximization for SLAM Backends
Emilio Olivastri, Alberto Pretto
Anomaly DetectionOptimizationSimultaneous Localization and MappingBenchmark
🎯 What it does: Proposed and implemented an incremental probabilistic consensus maximum consensus set method called IPC, used to filter the most consistent measurements and suppress outliers in pose graph optimization.
IQL-TD-MPC: Implicit Q-Learning for Hierarchical Model Predictive Control
Rohan Chitnis, Olivier Delalleau
Robotic IntelligenceReinforcement LearningSequentialBenchmark
🎯 What it does: Proposed an offline model-based reinforcement learning algorithm called IQL-TD-MPC, which serves as a manager in hierarchical control systems and can be combined with any offline RL worker, generating 'intent embeddings' as subgoals through planning;
iRoCo: Intuitive Robot Control From Anywhere Using a Smartwatch
F. Weigend, H. B. Amor
Robotic Intelligence
🎯 What it does: Proposed the iRoCo framework, which utilizes smartwatches and smartphones to achieve ubiquitous intuitive robot control, and demonstrated and evaluated it in teleoperation and drone control tasks.
Is it a Bug? Understanding Physical Unit Mismatches in Robot Software
Paulo Canelas, C. Timperley
Robotic Intelligence
🎯 What it does: Analyze the phenomenon of physical unit mismatches in robot software, study how developers intentionally introduce these mismatches, and propose error types and a classification system.
ISR-LLM: Iterative Self-Refined Large Language Model for Long-Horizon Sequential Task Planning
Zhehua Zhou, Lei Ma
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AITextSequential
🎯 What it does: Introduce the ISR-LLM framework, achieving long-term sequence task planning through preprocessing translation, LLM planning, and iterative self-improvement.
Iterative Learning Control for Deformable Open-Frame Cable-Driven Parallel Robots
Wuichung Cheng, Darwin Lau
Robotic Intelligence
🎯 What it does: Proposes an iterative learning control (ILC) scheme for deformable open-frame cable-driven parallel robots (D-CDPRs).
Iterative PnP and its application in 3D-2D vascular image registration for robot navigation
Jingwei Song, Maani Ghaffari
Pose EstimationAutonomous DrivingComputational EfficiencyImageBiomedical Data
🎯 What it does: Proposed a real-time, outlier-robust, and non-rigid vascular shape alignment 3D-2D image registration algorithm
JacobiGPU: GPU-Accelerated Numerical Differentiation for Loop Closure in Visual SLAM
Dhruva Kumar, Steven Y. Ko
Computational EfficiencySimultaneous Localization and MappingVideo
🎯 What it does: Introduce JacobiGPU, which uses GPU acceleration to speed up the Jacobian approximation based on the finite difference method in visual-inertial SLAM systems, thereby improving loop closure efficiency.
Jacquard V2: Refining Datasets using the Human In the Loop Data Correction Method
Qiuhao Li, Shenghai Yuan
Object DetectionData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: Correct and enhance the Jacquard Grasp Dataset using human-computer interaction methods, generating the Jacquard V2 Grasp Dataset.
Jade: A Differentiable Physics Engine for Articulated Rigid Bodies with Intersection-Free Frictional Contact
Gang Yang, Lin Shao
OptimizationPhysics Related
🎯 What it does: Developed a differentiable physics engine Jade specialized for articulated rigid bodies, supporting non-penetrating collision simulation and stable multi-friction contact LCP solving.
JaywalkerVR: A VR System for Collecting Safety-Critical Pedestrian-Vehicle Interactions
Kenta Mukoya, K. Kitani
Data SynthesisAutonomous DrivingVideo
🎯 What it does: Developed JaywalkerVR, a virtual reality human-computer interaction simulator, for collecting safety-critical pedestrian-vehicle interaction data, and constructed the CARLA-VR dataset based on this simulator.
Jerk-limited Traversal of One-dimensional Paths and its Application to Multi-dimensional Path Tracking
Jonas C. Kiemel, T. Kröger
OptimizationTabular
🎯 What it does: Propose an iterative rapid traversal method for multi-dimensional paths considering jerk constraints.
Johnsen-Rahbek Capstan Clutch: A High Torque Electrostatic Clutch
Timothy E. Amish, J. Lipton
Physics Related
🎯 What it does: Combined the Johnsen-Rahbek effect with the exponential tension scaling capstan effect to design and manufacture a high-torque, low-power consumption electrostatic friction brake.
Joint Response and Background Learning for UAV Visual Tracking
Biao Wang, Yang Liu
Object TrackingConvolutional Neural NetworkVideo
🎯 What it does: Propose a response consistency and reversibility learning module, a background-aware module, and a fast coarse-to-fine scale search strategy to construct two CF trackers, RBLT and DeepRBLT.
Joint-Loss Enhanced Self-Supervised Learning for Refinement-Coupled Object 6D Pose Estimation
Fengjun Mu, Hong Cheng
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: Propose a self-supervised 6D pose estimation method based on a pixel-level weighted dense fusion architecture, which directly learns from unannotated RGB-D data using an iterative annotation parser, and introduce a pose refinement approach with a differentiable renderer based on joint loss.
JSTR: Joint Spatio-Temporal Reasoning for Event-based Moving Object Detection
Hanyu Zhou, Luxin Yan
Object DetectionPoint Cloud
🎯 What it does: Proposes a joint spatiotemporal reasoning method for event-based moving object detection.
K-BMPC: Derivative-based Koopman Bilinear Model Predictive Control For Tractor-trailer Trajectory Tracking With Unknown Parameters
Zehao Wang, Jingchuan Wang
Autonomous DrivingOptimization
🎯 What it does: Propose a derivative-based Koopman bilinear model predictive control (K-BMPC) for tractor-trailer trajectory tracking with unknown parameters.
KDD-LOAM: Jointly Learned Keypoint Detector and Descriptors Assisted LiDAR Odometry and Mapping
Renlang Huang, Liang Li
Pose EstimationAutonomous DrivingComputational EfficiencyConvolutional Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposed a tightly coupled keypoint detector and descriptor network (TCKDD), and built a keypoint detector and descriptor assisted LiDAR odometry and mapping framework (KDD-LOAM), achieving real-time odometry and efficient scan-map registration using sparse keypoints.
Keypoints-guided Lightweight Network for Single-view 3D Human Reconstruction
Yuhang Chen, Chenxing Wang
Pose EstimationConvolutional Neural NetworkAuto EncoderImageMesh
🎯 What it does: Developed a lightweight network based on key points, using an encoder-decoder framework to achieve single-view 3D human reconstruction.
Kinematic Modeling and Control of a Soft Robotic Arm with Non-constant Curvature Deformation
Zhanchi Wang, N. Freris
Robotic Intelligence
🎯 What it does: Propose a 3D kinematic model for soft robotic arms and design a corresponding inverse kinematic controller for positioning and trajectory tracking.
Kinematic Optimization of a Robotic Arm for Automation Tasks with Human Demonstration
Inbar Meir, A. Sintov
OptimizationRobotic Intelligence
🎯 What it does: Based on human demonstrations, optimize the kinematics of the robot arm to complete specific automated tasks, generating optimal robot designs and paths.
Kinematic Synergy Primitives for Human-Like Grasp Motion Generation
J. Starke, Tamim Asfour
Robotic IntelligenceSequential
🎯 What it does: Proposed an adaptive kinematic grasp primitive in the synergistic space, which can generate new human-like grasp motions by directly controlling high-level grasp parameters.
Kinematic-aware Prompting for Generalizable Articulated Object Manipulation with LLMs
Wenke Xia, Di Hu
Robotic IntelligenceLarge Language ModelPrompt EngineeringChain-of-Thought
🎯 What it does: Proposed a kinematic knowledge prompting-based LLM framework for general robotic manipulation of articulated objects.
Kinesthetic-based In-Hand Object Recognition with an Underactuated Robotic Hand
Julius Arolovitch, A. Sintov
RecognitionRobotic IntelligenceTime Series
🎯 What it does: Utilizing a sensorless actuator-driven arm, kinesthetic tactile recognition is performed during in-hand manipulation to identify trained objects by perceiving actuator position and torque;
Kitchen Artist: Precise Control of Liquid Dispensing for Gourmet Plating
Hung-Jui Huang, Wenzhen Yuan
OptimizationRobotic Intelligence
🎯 What it does: Developed a sauce drip irrigation robot capable of precisely controlling the thickness of the dripped liquid on the surface;
Knowledge acquisition plans: Generation, combination, and execution
Dylan A. Shell, J. O’Kane
OptimizationRobotic IntelligenceWorld ModelGraph
🎯 What it does: Propose a method that enables robots to answer questions by exploring the environment and integrating existing world knowledge, achieving decoupling between knowledge management and learning through graph structure representation; further improve plan execution quality by leveraging reasoning, information overlap, and multi-robot collaboration.
L-DYNO: Framework to Learn Consistent Visual Features Using Robot’s Motion
Kartikeya Singh, Karthik Dantu
Pose EstimationRepresentation LearningRobotic IntelligenceImageTime Series
🎯 What it does: Proposed the L-DYNO framework, which uses external motion signals (such as inertial sensing) to guide visual feature learning, maintaining consistency between features and the robot's relative pose changes;
L-VIWO: Visual-Inertial-Wheel Odometry based on Lane Lines
Bin Zhao, Yulong Li
Autonomous DrivingOptimizationSimultaneous Localization and MappingImageMultimodality
🎯 What it does: Propose a visual-inertial-wheel speed odometer method (L-VIWO) based on lane lines, which eliminates odometry drift through multi-frame lane line tracking, optimization of lane line sampling points, construction of a local lane line map, vehicle pose correction, and a graph optimization model.
LaCE-LHMP: Airflow Modelling-Inspired Long-Term Human Motion Prediction By Enhancing Laminar Characteristics in Human Flow
Yufei Zhu, A. Lilienthal
Autonomous DrivingExplainability and InterpretabilityRobotic IntelligenceFlow-based ModelTime Series
🎯 What it does: Proposes a long-term human motion prediction method based on laminar flow enhancement (LaCE-LHMP) for safe operation of robots and vehicles in crowded environments.
LAGOON: Language-Guided Motion Control
Shusheng Xu, Yi Wu
Domain AdaptationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Propose a multi-stage reinforcement learning method called LAGOON, which first uses a pre-trained model to map natural language instructions to human actions, then trains control strategies in a simulated environment to imitate these actions, and finally deploys the learned strategies on a quadruped robot through domain randomization, enabling the robot to perform diverse behaviors based on natural language instructions in the real world.
Language and Sketching: An LLM-driven Interactive Multimodal Multitask Robot Navigation Framework
Weiqin Zu, Jun Wang
Robotic IntelligenceLarge Language ModelReinforcement LearningAgentic AIMultimodality
🎯 What it does: Designed and implemented an interactive multi-modal multi-task robot navigation framework called LIM2N based on a large language model (LLM), utilizing language and hand-drawn inputs as navigation constraints and control goals, and achieving multi-task processing through reinforcement learning agents.
Language to Map: Topological map generation from natural language path instructions
Hideki Deguchi, Shun Taguchi
GenerationLarge Language ModelPrompt EngineeringTextGraph
🎯 What it does: Propose a method for generating topological maps based on natural language path descriptions and automatically generating new paths
Language-Conditioned Affordance-Pose Detection in 3D Point Clouds
Toan Nguyen, Anh Nguyen
Object DetectionPose EstimationDiffusion modelPoint Cloud
🎯 What it does: Propose a language-based joint learning method for 3D point cloud manipulability and pose, capable of detecting manipulable regions and generating corresponding 6-degree-of-freedom (6-DoF) poses for arbitrary open-vocabulary labels.
Language-Conditioned Robotic Manipulation with Fast and Slow Thinking
Minjie Zhu, Jian Tang
Robotic IntelligenceSupervised Fine-TuningVision Language ModelTextSequential
🎯 What it does: Propose a language-based robotic manipulation framework RFST, which includes an instruction discriminator and a slow thinking system that aligns a fine-tuned vision-language model with a policy network.
Language-EXtended Indoor SLAM (LEXIS): A Versatile System for Real-time Visual Scene Understanding
Christina Kassab, M. Fallon
ClassificationSegmentationPose EstimationTransformerLarge Language ModelVision Language ModelSimultaneous Localization and MappingImageText
🎯 What it does: Developed a real-time indoor SLAM system LEXIS based on LLM, achieving the unification of scene understanding and pose estimation.
Language-guided Active Sensing of Confined, Cluttered Environments via Object Rearrangement Planning
Weihan Chen, A. H. Qureshi
Robotic IntelligenceVision-Language-Action ModelImageText
🎯 What it does: Propose a language-guided active perception method that manipulates objects in narrow, cluttered environments to provide dense perception for specified regions.
LB-R2R-Calib: Accurate and Robust Extrinsic Calibration of Multiple Long Baseline 4D Imaging Radars for V2X
Jun Zhang, Danwei Wang
Autonomous DrivingPoint Cloud
🎯 What it does: Proposed an extrinsic parameter calibration method called LB-R2R-Calib for multi-path long-baseline 4D radar, and verified its feasibility through experiments.
LCCRAFT: LiDAR and Camera Calibration Using Recurrent All-Pairs Field Transforms Without Precise Initial Guess
Yunju Lee, Kuan-Wen Chen
Autonomous DrivingRecurrent Neural NetworkOptical FlowImagePoint Cloud
🎯 What it does: Proposed an online LiDAR and camera calibration network called LCCRAFT, which utilizes RAFT's 4D correlation volume and correlation search to associate RGB images with projected depth maps, and achieves robust error compensation without requiring precise initial guesses through a weight-sharing recursive update mechanism.
Leaf-Inspired FSR Array and Insole-Type Sensor Module for Mobile Three-Dimensional Ground Reaction Force Estimation
Taeyeon Kim, Kyoungchul Kong
Time SeriesBiomedical Data
🎯 What it does: Proposed a shoe insole sensor module based on leaf-inspired force-sensitive resistor (FSR) arrays for accurately estimating three-dimensional ground reaction force (GRF) during various human movements.
LeagTag: An Elongated High-Accuracy Fiducial Marker for Tight Spaces *
Hideyuki Tanaka, Kunihiro Ogata
Pose Estimation
🎯 What it does: Developed a world-first slender fingerprint marker for achieving high-precision 6-degree-of-freedom (6-DOF) measurement in confined spaces.
LeapRun: A Dynamic Soft Robot with Running and Jumping Capabilities
J. Lu, M. Zhang
Robotic Intelligence
🎯 What it does: Proposed and manufactured a dynamic soft robot named LeapRun, capable of running, continuous jumping, and crossing complex rough surfaces while achieving untethered operation.
Learn to Navigate in Dynamic Environments with Normalized LiDAR Scans
Wei Zhu, M. Hayashibe
Autonomous DrivingRecurrent Neural NetworkPoint CloudTime Series
🎯 What it does: Utilizing serialized normalized LiDAR scans (LNDNL) and LSTM models to learn direct mapping from raw sensor observations to robot actions for navigation in dynamic environments.
Learning a Stable Dynamic System with a Lyapunov Energy Function for Demonstratives Using Neural Networks
Yu Zhang, Long Cheng
Optimization
🎯 What it does: Proposed a dynamic system algorithm based on neural networks that can extract key insights from demonstration data and learn candidate Lyapunov energy functions consistent with the demonstrations to achieve precise learning and global stability.
Learning active manipulation to target shapes with model-free, long-horizon deep reinforcement learning
Matias Sivertsvik, E. Misimi
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: Trained a robotic arm in a simulated environment using model-free, long-horizon deep reinforcement learning (PPO), with segmented images as observational inputs, actively deforming elastic-plastic objects into arbitrary target shapes and achieving zero-shot real-world transfer.
Learning Adaptive Safety for Multi-Agent Systems
Luigi Berducci, R. Grosu
OptimizationSafty and PrivacyReinforcement Learning
🎯 What it does: Proposed and implemented the ASRL framework, which uses reinforcement learning to automatically optimize control barrier functions (CBF) and policy coefficients in multi-agent systems, thereby enhancing safety and long-term performance.
Learning Agile Bipedal Motions on a Quadrupedal Robot
Yunfei Li, Yi Wu
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: Implement human-like bipedal motion on a lightweight quadruped robot, developing a hierarchical framework with a low-level motion conditional control strategy and a high-level motion generator that can track base and front limb movements while maintaining rear foot balance.
Learning Agile Locomotion and Adaptive Behaviors via RL-augmented MPC
Lydia Y. Chen, Quan Nguyen
Robotic IntelligenceReinforcement Learning
🎯 What it does: By combining reinforcement learning with model predictive control, the study investigates and implements adaptive balance and foot swing reflex in quadruped robots operating in vision-blind environments, thereby enhancing the robot's robustness and agility in complex terrains.
Learning Barrier-Certified Polynomial Dynamical Systems for Obstacle Avoidance with Robots
Martin Schonger, Sami Haddadin
OptimizationRobotic Intelligence
🎯 What it does: Propose incorporating barrier certificates into the optimization problem when learning stable dynamical systems (DS), achieving stable and certificate-constrained DS to enhance the robustness of robots in static obstacle avoidance tasks; use polynomial representations for DS and solve the optimization problem via sum-of-squares (SOS) techniques; this method can handle obstacle shapes that are difficult to process in traditional DS learning frameworks.
Learning Complex Motion Plans using Neural ODEs with Safety and Stability Guarantees
Farhad Nawaz, Nadia Figueroa
OptimizationSafty and PrivacyRobotic IntelligenceTextOrdinary Differential Equation
🎯 What it does: Propose a method based on dynamical systems (DS) and neural ordinary differential equations (NODE) to learn complex, periodic motion plans from tactile demonstrations, achieving safety and stability through online quadratic programming correction.
Learning Contact for Haptic Feedback: Switching X-lateral Teleoperators
Nural Yilmaz, U. Tumerdem
Safty and PrivacyRobotic IntelligenceSupervised Fine-Tuning
🎯 What it does: Proposed the X-lateral hybrid unilateral-bilateral teleoperation framework and designed a learning-based contact detection and switching mechanism
Learning Continuous Control with Geometric Regularity from Robot Intrinsic Symmetry
Shengchao Yan, Wolfram Burgard
Robotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkTransformerReinforcement Learning
🎯 What it does: Proposed and implemented a geometric regularization network architecture that leverages the inherent reflection and rotational symmetry of robot structures for single-agent control learning, and realized this framework in both online and offline learning methods.
Learning Covariances for Estimation with Constrained Bilevel Optimization
Mohamad Qadri, Michael Kaess
Robotic Intelligence
🎯 What it does: Propose a gradient-based method that learns the error covariance matrix for robot state estimation on a factor graph through constrained bi-level optimization to achieve well-conditioned covariance estimation;
Learning Crowd Behaviors in Navigation with Attention-based Spatial-Temporal Graphs
Yanying Zhou, J. Garcke
Robotic IntelligenceRecurrent Neural NetworkGraph Neural NetworkGraph
🎯 What it does: Proposed a deep graph learning architecture based on attention mechanisms, utilizing space-temporal graphs to enhance robot navigation performance in dynamic crowd environments.
Learning Decentralized Flocking Controllers with Spatio-Temporal Graph Neural Network
Siji Chen, Chang-Tien Lu
Robotic IntelligenceGraph Neural NetworkGraph
🎯 What it does: Implement decentralized group control using spatiotemporal graph neural networks (STGNN), training and verifying its performance in tasks such as coordinated flight, leader following, and obstacle avoidance.
Learning Distributional Demonstration Spaces for Task-Specific Cross-Pose Estimation
Jenny Wang, David Held
Pose EstimationMultimodality
🎯 What it does: Proposed a method based on distributed demonstration space that can learn precise relative placement tasks using only 10-20 unlabeled multimodal demonstrations.
Learning Diverse Skills for Local Navigation under Multi-constraint Optimality
Jin Cheng, G. Martius
Autonomous DrivingOptimization
🎯 What it does: Proposes a method for learning diverse local navigation skills within a multi-constraint optimality framework.
Learning Dual-arm Object Rearrangement for Cartesian Robots
Shishun Zhang, Kai Xu
Robotic IntelligenceTransformerReinforcement Learning
🎯 What it does: Proposes an online task allocation method based on reinforcement learning (RL) and an attention network for the dual-arm object rearrangement problem, aiming to reduce the total completion time through reasonable object-arm correspondence.
Learning Emergent Gaits with Decentralized Phase Oscillators: on the role of Observations, Rewards, and Feedback
Jenny Zhang, Sangbae Kim
Robotic IntelligenceReinforcement Learning
🎯 What it does: Proposed a minimized phase oscillator model for learning quadruped robot locomotion.
Learning Extrinsic Dexterity with Parameterized Manipulation Primitives
Shih-Min Yang, Todor Stoyanov
Robotic IntelligenceReinforcement Learning
🎯 What it does: Learned a hierarchical strategy of parameterized primitive operations to move target objects into a graspable pose by leveraging the environment.
Learning Fabric Manipulation in the Real World with Human Videos
Robert Lee, Peter Corke
Robotic IntelligenceVideo
🎯 What it does: Learn fabric manipulation demonstrations directly collected from human videos, train a pick-and-place strategy using a minimal amount of demonstration data, and subsequently implement fabric smoothing and folding tasks on a real robot.
Learning for Deformable Linear Object Insertion Leveraging Flexibility Estimation from Visual Cues
Mingen Li, Changhyun Choi
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: Propose a two-stage DLO insertion method, first estimating flexibility through vision, then training an insertion strategy based on the flexibility estimation using reinforcement learning.
Learning for Dynamic Subteaming and Voluntary Waiting in Heterogeneous Multi-Robot Collaborative Scheduling
W. J. Jose, Hao Zhang
Robotic IntelligenceGraph Neural NetworkTransformerReinforcement Learning
🎯 What it does: Propose a graph learning-based heterogeneous robot collaborative scheduling method, formulated as a bipartite matching problem to maximize the reward matrix, thereby achieving dynamic subgroup formation and voluntary waiting.
Learning Force Control for Legged Manipulation
Tifanny Portela, Pulkit Agrawal
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a reinforcement learning task specification that explicitly matches the desired contact force level, and combine force control with the coordination of the robot's body and manipulator, proposing an end-to-end legged manipulation control strategy;
Learning Generalizable Patrolling Strategies through Domain Randomization of Attacker Behaviors
Carlos Diaz Alvarenga, Stefano Carpin
Reinforcement Learning
🎯 What it does: This paper proposes a data-driven reinforcement learning method to address the adaptation and defense against unknown attacker strategies in graph patrolling;
Learning Heterogeneous Multi-Agent Allocations for Ergodic Search
Ananya Rao, H. Choset
OptimizationRobotic IntelligenceReinforcement Learning
🎯 What it does: Built a reinforcement learning framework for heterogeneous multi-agent search and allocation based on frequency domain spectral coefficients, using spectral decomposition to guide robots with different sensors and motion models to achieve information coverage.
Learning Highly Dynamic Behaviors for Quadrupedal Robots
Chong Zhang, Lei Han
Robotic IntelligenceReinforcement LearningSequential
🎯 What it does: Propose a learning-based method enabling quadruped robots to learn highly dynamic behaviors from animal motion data, deploy the learned controller onto the robot to achieve behaviors such as sprinting, jumping, and sharp turns; human-robot interaction via a marked stick can activate multiple behaviors including walking, running, sitting, and jumping.
Learning Interaction Constraints for Robot Manipulation via Set Correspondences
Junyu Nan, Brian Okorn
Pose EstimationRobotic IntelligencePoint Cloud
🎯 What it does: Propose the SCAlign cross-pose estimation method, which uses the Set Correspondence Network (SCN) to assign set labels to points in point clouds, achieving set-level correspondence, and computes SE(3) transformations via an alignment module to obtain cross-poses;
Learning Language-Conditioned Deformable Object Manipulation with Graph Dynamics
Kaichun Mo, Xueqian Wang
Robotic IntelligenceGraph Neural NetworkTransformerVision-Language-Action ModelTextMultimodality
🎯 What it does: Proposes a language instruction-based multi-task manipulation framework for deformable objects, using a unified Transformer architecture to process multimodal data and modeling object nonlinear dynamics through a visible connectivity graph.
Learning manipulation of steep granular slopes for fast Mini Rover turning
Deniz Kerimoglu, Daniel I. Goldman
Robotic IntelligencePhysics Related
🎯 What it does: Using simulation and physical experiments to explore the rapid turning mechanism of small robots on low-viscosity powder slopes.
Learning Model Predictive Control with Error Dynamics Regression for Autonomous Racing
Haoru Xue, Francesco Borrelli
Autonomous Driving
🎯 What it does: Proposed a learning model predictive control (LMPC) strategy for high-speed extreme driving, capable of iteratively exploring and learning unknown dynamics.
Learning Motion Reconstruction from Demonstration via Multi-Modal Soft Tactile Sensing
Cheng Pan, Josie Hughes
Robotic IntelligenceRecurrent Neural NetworkMultimodality
🎯 What it does: Proposes a framework that utilizes a wearable fingertip multimodal tactile sensor to learn and reconstruct robot motion from human demonstrations; by collecting tactile data from different motion primitives (striking, rotating, translating) and using an LSTM model to replicate robot motion solely based on tactile demonstrations.
Learning Multi-Scale Context Mask-RCNN Network for Slant Angled Aerial Imagery in Instance Segmentation in a Sim2Real setup
Qiranul Saadiyean, Suresh Sundaram
SegmentationData SynthesisDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Proposed a Sim2Real-based multi-scale context Mask-RCNN (MSC-RCNN) network for instance segmentation of tilted-angle aerial images, enhancing detection performance by generating synthetic datasets and fusing two feature pyramid backbones.
Learning Quadrupedal Locomotion with Impaired Joints Using Random Joint Masking
Mincheol Kim, Jung-Yup Kim
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes a deep reinforcement learning framework aimed at enabling quadruped robots to walk even when joints are damaged.
Learning Realistic and Reasonable Grasps for Anthropomorphic Hand in Cluttered Scenes
Haonan Duan, Peng Wang
Robotic IntelligencePoint Cloud
🎯 What it does: Proposes a framework for learning realistic and reasonable grasping for humanoid hands in cluttered scenes.
Learning Risk-Aware Quadrupedal Locomotion using Distributional Reinforcement Learning
Lukas Schneider, Marco Hutter
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a risk-aware quadruped robot gait training method based on distributed reinforcement learning, integrating risk measures into PPO to form DPPO;
Learning Self-Confidence from Semantic Action Embeddings for Improved Trust in Human-Robot Interaction
Cedric Goubard, Y. Demiris
Representation LearningRobotic Intelligence
🎯 What it does: Propose the SCONE strategy, which utilizes semantic action embeddings to learn robot confidence from experience, thereby enhancing trust in human-robot interaction.
Learning Temporal Cues by Predicting Objects Move for Multi-camera 3D Object Detection
Seokha Moon, Jinkyu Kim
Object DetectionAutonomous DrivingVideoPoint CloudBenchmark
🎯 What it does: Proposed a prediction-based dual-branch network called DAP to leverage historical observations for learning object motion temporal cues, thereby improving multi-camera 3D object detection.
Learning Terminal State of the Trajectory Planner: Application for Collision Scenarios of Autonomous Vehicles
Joonhee Lim, Dongsuk Kum
Autonomous DrivingOptimizationSafty and PrivacyExplainability and InterpretabilityReinforcement Learning
🎯 What it does: Propose a CAMs trajectory planning method combining deep reinforcement learning (DRL) with quintic polynomial (QP) trajectory planning, using DRL to predict terminal states and trajectory confidence, and then generating QP trajectories based on these predictions to produce smoother and more interpretable paths.
Learning to Catch Reactive Objects with a Behavior Predictor
Kai Lu, Andrew Markham
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a method combining an explicitly learned target state predictor with reinforcement learning to capture mobile targets exhibiting reactive behavior.
Learning to Design 3D Printable Adaptations on Everyday Objects for Robot Manipulation
Michelle Guo, C. K. Liu
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes a framework for automatically designing 3D-printable adapters to improve the performance of robots when interacting with everyday objects, enhancing 'robot ergonomics';
Learning to estimate incipient slip with tactile sensing to gently grasp objects
Dirk Boonstra, Michaël Wiertlewski
Robotic IntelligenceConvolutional Neural NetworkImage
🎯 What it does: Estimate the safety margin using tactile sensing and machine learning methods to predict slippage during grasping in advance, thereby achieving gentle and robust grasping.
Learning to Explore Indoor Environments using Autonomous Micro Aerial Vehicles
Yuezhan Tao, Vijay Kumar
Robotic IntelligenceReinforcement Learning
🎯 What it does: Using deep learning to predict unknown indoor maps and deep reinforcement learning for efficient exploration of autonomous micro aerial vehicles.
Learning to Grasp in Clutter with Interactive Visual Failure Prediction
Michael Murray, Maya Cakmak
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: An interactive visual failure prediction (IVFP) method is proposed, which uses an interactive detector to perform visual evaluation and verify the success of grasping without fully executing the grasping action;
Learning to Play Foosball: System and Baselines
Janosch Moos, Debora Clever
Robotic IntelligenceReinforcement LearningBenchmark
🎯 What it does: Built an automated table football machine and a corresponding simulation environment, demonstrating diverse robot learning tasks on the platform and conducting baseline experiments for the first time on this platform.
Learning to walk in confined spaces using 3D representation
Takahiro Miki, Marco Hutter
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a quadruped robot control method based on reinforcement learning and 3D volume representation, achieving stable gaits in narrow and rugged environments.
Learning Vision-Based Bipedal Locomotion for Challenging Terrain
Helei Duan, Alan Fern
Domain AdaptationRobotic IntelligenceReinforcement LearningImageTabular
🎯 What it does: Train the controller using height maps in the robot's local coordinate system in simulation, and train the height map predictor with depth maps and robot state data, enabling vision-based bipedal robots to walk on challenging terrain at commanded speeds and directions.
Learning Vision-based Pursuit-Evasion Robot Policies
Andrea V. Bajcsy, J. Malik
Robotic IntelligenceSupervised Fine-TuningImage
🎯 What it does: Convert the pursuit-evasion robot strategy learning problem into a supervised learning problem by generating supervisory signals for partially observable strategies using fully observable robot strategies, and conducting real-world deployment in the wild on a quadruped robot equipped with RGB-D cameras;
Learning When to Ask for Help: Efficient Interactive Navigation via Implicit Uncertainty Estimation
Ifueko Igbinedion, S. Karaman
Computational EfficiencyRobotic IntelligenceImage
🎯 What it does: Trained a lightweight interaction strategy that enables the robot to decide whether to act autonomously or request expert assistance when uncertainty is inferred.
Learning Which Side to Scan: Multi-View Informed Active Perception with Side Scan Sonar for Autonomous Underwater Vehicles
A. Sethuraman, James McMahon
ClassificationRobotic IntelligenceGraph Neural NetworkImage
🎯 What it does: Proposes an active perception framework for multi-view adaptive survey and reacquisition based on side-scan sonar images, utilizing graph theory to represent tasks and employing graph neural networks for view classification and selection of the next best view.
Learning with Chemical versus Electrical Synapses Does it Make a Difference?
M'onika Farsang, R. Grosu
Autonomous Driving
🎯 What it does: Compare the performance of chemical synapses and electrical synapses within the same neural network architecture, conducting experiments on sparse and fully connected networks.