arXivSub Start free trial

ICRA 2023 Papers — Page 4

IEEE International Conference on Robotics and Automation · 1341 papers

Curriculum-Based Imitation of Versatile Skills

M. Li, G. Neumann

Mixture of ExpertsMultimodality

🎯 What it does: Propose a multi-modal imitation learning method based on curriculum learning, which uses data point weighting and entropy rewards to specialize the model on representable sub-data, and covers all data through a Mixture of Experts (MoE).

Curvature-Aware Model Predictive Contouring Control

Lorenzo Lyons, L. Ferranti

Autonomous DrivingOptimization

🎯 What it does: Proposes a curvature-aware model predictive contour control (CA-MPCC) scheme for path planning and tracking in mobile robots

CurveFormer: 3D Lane Detection by Curve Propagation with Curve Queries and Attention

Yifeng Bai, Erkang Cheng

Autonomous DrivingTransformer

🎯 What it does: Propose CurveFormer, a single-stage Transformer method that directly computes 3D lane parameters, avoiding perspective transformation.

D-Align: Dual Query Co-attention Network for 3D Object Detection Based on Multi-frame Point Cloud Sequence

Junhyung Lee, J. Choi

Object DetectionAutonomous DrivingTransformerPoint CloudBenchmark

🎯 What it does: Propose a 3D object detection method called D-Align based on multi-frame point cloud sequences, which generates powerful bird's-eye-view (BEV) features and completes detection tasks by aligning and aggregating features from target frames and support frames.

D2CoPlan: A Differentiable Decentralized Planner for Multi-Robot Coverage

V. Sharma, Pratap Tokekar

OptimizationRobotic Intelligence

🎯 What it does: Proposes a learning-based differentiable distributed coverage planner (D2CoPLAN) that achieves efficient runtime and scalability in the number of agents for multi-robot coverage planning;

D2NT: A High-Performing Depth-to-Normal Translator

Yi Feng, Rui Fan

Image TranslationDepth EstimationConvolutional Neural NetworkImage

🎯 What it does: Proposed an ultra-fast depth-to-normal converter (D2NT) that directly converts depth images into surface normal maps; simultaneously introduced an adaptive breakpoint-aware gradient (DAG) filter to improve depth gradient estimation; and developed a surface normal refinement module that can be integrated into any existing depth-to-normal estimator, significantly enhancing estimation accuracy.

DAMS-LIO: A Degeneration-Aware and Modular Sensor-Fusion LiDAR-inertial Odometry

Fuzhang Han, Yanmei Jiao

Autonomous DrivingOptimizationSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Propose a lightweight iEKF-based LiDAR-inertial odometry system with a degradation-aware modular sensor fusion pipeline, which updates using LiDAR points and the relative pose of another odometry as measurements only when degradation is detected.

Data-Association-Free Landmark-based SLAM

Yihao Zhang, Kasra Khosoussi

OptimizationSimultaneous Localization and Mapping

🎯 What it does: Studies landmark-based SLAM under unknown data association, inferring robot trajectory, unknown number of landmark positions, and measurement-landmark associations, and proposes an inner-outer optimization framework along with a discrete-continuous optimization approach for association.

Data-Driven Estimation of Forces Along the Backbone of Concentric Tube Continuum Robots

Heiko Donat, Jochen J. Steil

Data-Centric LearningRobotic Intelligence

🎯 What it does: Propose a data-driven method for estimating contact forces on the body of a conjugate tube continuous robot.

Data-driven Loop Closure Detection in Bathymetric Point Clouds for Underwater SLAM

Jiarui Tan, J. Folkesson

Robotic IntelligenceSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed and trained a neural network architecture for loop closure detection and rough alignment of underwater bathymetric point clouds.

Data-driven optimal control under safety constraints using sparse Koopman approximation

Hongzhe Yu, Yongxin Chen

OptimizationSafty and Privacy

🎯 What it does: Solved the doubly optimal reachable safety control problem using sparse Koopman operator approximation.

Data-Driven Risk-sensitive Model Predictive Control for Safe Navigation in Multi-Robot Systems

Atharva Navsalkar, A. Hota

OptimizationRobotic Intelligence

🎯 What it does: A data-driven distributionally robust model predictive control method is proposed, achieving safe navigation in multi-robot systems through finite-horizon optimization with CVaR-based collision avoidance constraints.

Data-Driven Spectral Submanifold Reduction for Nonlinear Optimal Control of High-Dimensional Robots

J. I. Alora, M. Pavone

OptimizationRobotic IntelligencePhysics Related

🎯 What it does: Propose a data-driven method based on SSMR for extracting control-oriented low-dimensional models from data, applicable to modeling and control of high-dimensional nonlinear robotic systems.

Data-Driven Stochastic Motion Evaluation and Optimization with Image by Spatially-Aligned Temporal Encoding

Takeru Oba, N. Ukita

Autonomous DrivingOptimizationImageSequential

🎯 What it does: A probabilistic motion prediction method for long-term motion is proposed, aiming to predict the motion required to complete a task from the initial state of a given image; spatially aligned temporal encoding integrates images and motion data into the image feature domain, and an energy-based model (EBM) is used to evaluate task reachability; simultaneously, a self-supervised deep motion optimizer (DMO) is proposed to work with EBM to address hyperparameter tuning and local optimal problems in gradient optimization.

Data-Efficient Characterization of the Global Dynamics of Robot Controllers with Confidence Guarantees

Ewerton R. Vieira, Kostas E. Bekris

Robotic IntelligenceGaussian Splatting

🎯 What it does: Integrate Gaussian process surrogate models with topological methods to describe the global dynamics of robot controllers, including black-box controllers, using minimal data.

Data-Efficient Learning of Natural Language to Linear Temporal Logic Translators for Robot Task Specification

Jiayi Pan, D. Berenson

Data-Centric LearningRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningText

🎯 What it does: Propose a learning-based method to translate natural language commands into linear temporal logic (LTL) specifications using minimal manually annotated data.

Data-efficient Non-parametric Modelling and Control of an Extensible Soft Manipulator

M. Kasaei, Mohsen Khadem

Data-Centric LearningRobotic Intelligence

🎯 What it does: Developed a data-efficient, non-parametric modeling and control method for soft robotic hands, achieving a continuous model using only 25 real demonstration points and implementing trajectory tracking control based on this model.

DC-MOT: Motion Deblurring and Compensation for Multi-Object Tracking in UAV Videos

Song Cheng, Xueming Xiao

RestorationObject TrackingVideo

🎯 What it does: Proposes a multi-object tracking framework for UAV videos that addresses two types of motion errors: motion blur and camera motion coupling; the framework includes a hybrid deblurring module and a motion compensation module, which can enhance tracking performance while maintaining spatiotemporal consistency.

DDK: A Deep Koopman Approach for Longitudinal and Lateral Control of Autonomous Ground Vehicles

Yongqian Xiao

Autonomous DrivingOptimization

🎯 What it does: A data-driven approach using the Koopman operator is employed to model vehicle dynamics, and a linear model predictive control (L-MPC) is designed based on this linear time-invariant model to achieve coupled tracking of longitudinal and lateral trajectories.

DDS3D: Dense Pseudo-Labels with Dynamic Threshold for Semi-Supervised 3D Object Detection

Jingyu Li, Dingkang Liang

Object DetectionPoint Cloud

🎯 What it does: Propose a semi-supervised 3D object detector named DDS3D.

Dealing with Sparse Rewards in Continuous Control Robotics via Heavy-Tailed Policy Optimization

Souradip Chakraborty, Dinesh Manocha

OptimizationRobotic IntelligenceReinforcement LearningBenchmark

🎯 What it does: Proposed a heavy-tailed random policy gradient (HT-PSG) algorithm to address the sparse reward problem in continuous control robotic tasks.

Decentralised Active Perception in Continuous Action Spaces for the Coordinated Escort Problem

Rhett Hull, R. Fitch

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Investigated the problem of enabling support robots to assist primary robots in accomplishing tasks within decentralized teams, and proposed a method to assess the impact of support robots on primary robot tasks.

Decentralized Deadlock-free Trajectory Planning for Quadrotor Swarm in Obstacle-rich Environments

Jungwon Park, H. J. Kim

Autonomous DrivingOptimization

🎯 What it does: Proposed a distributed multi-UAV trajectory planning algorithm that generates safe, deadlock-free trajectories in obstacle-rich environments under limited communication ranges.

Decentralized Multi-agent Exploration with Limited Inter-agent Communications

Hans J. He, Benjamin Biggs

OptimizationReinforcement Learning

🎯 What it does: Propose a decentralized environment learning method that maximizes multi-agent joint information gain under constrained communication conditions.

Decision diagrams as plans: Answering observation-grounded queries

Dylan A. Shell, J. O’Kane

Robotic Intelligence

🎯 What it does: Utilize simplified ordered binary decision diagrams (BDDs) to represent the world knowledge and queries required by robots when answering structured queries about the environment, and propose a new product operation as well as extend the classic dynamic variable reordering technique for planning, to directly utilize BDDs for reasoning and planning.

Decoupling Skill Learning from Robotic Control for Generalizable Object Manipulation

Kai Lu, A. Markham

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: Propose separating skill learning from whole-body control, determining the skill dynamics of a virtual manipulator through reinforcement learning, and achieving optimal whole-body control for high-dimensional joint movements via quadratic programming (QP).

DEdgeNet: Extrinsic Calibration of Camera and LiDAR with Depth-discontinuous Edges

Yiyang Hu, Hui Zhang

Autonomous DrivingConvolutional Neural NetworkImagePoint Cloud

🎯 What it does: Proposed a real-time network that achieves extrinsic calibration between RGB cameras and LiDAR by extracting depth discontinuity edges from a single image.

Deep Interactive Full Transformer Framework for Point Cloud Registration

Guangyan Chen, Yufeng Yue

Pose EstimationTransformerPoint Cloud

🎯 What it does: Proposed a point cloud registration network called Deep Interactive Full Transformer (DIFT)

Deep Learning on Home Drone: Searching for the Optimal Architecture

Alaa Maalouf, Dan Feldman

Object DetectionSegmentationCompressionNeural Architecture SearchConvolutional Neural Network

🎯 What it does: Built a system for real-time semantic segmentation on the Raspberry Pi Zero v2, mounted on a DJI Tello toy drone, achieving fully autonomous real-time object detection and classification without external computers or human intervention.

Deep Masked Graph Matching for Correspondence Identification in Collaborative Perception

Peng Gao, Haotian Zhang

Autonomous DrivingGraph Neural NetworkMultimodality

🎯 What it does: Achieving object correspondence identification in multi-robot collaborative perception through deep occlusion graph matching.

Deep metric learning for visual servoing: when pose and image meet in latent space

Samuel Felton, É. Marchand

Robotic IntelligenceContrastive LearningMultimodality

🎯 What it does: Propose a visual servoing method that controls robot motion in the latent space, leveraging metric learning to associate camera pose representations with corresponding image embeddings.

Deep Neural Network Architecture Search for Accurate Visual Pose Estimation aboard Nano-UAVs

Elia Cereda, D. Palossi

Pose EstimationRobotic IntelligenceNeural Architecture SearchConvolutional Neural Network

🎯 What it does: Utilize a novel neural architecture search (NAS) technique to automatically identify multiple Pareto-optimal convolutional neural networks (CNNs) for visual pose estimation on small unmanned aerial vehicles (UAVs), and implement closed-loop deployment on the 27-g Crazyflie nano-UAV.

Deep Occupancy-Predictive Representations for Autonomous Driving

Eivind Meyer, M. Althoff

Autonomous DrivingRepresentation LearningGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: Propose learning task-related features to encode probabilistic occupancy maps into pre-trained environment state representations, and leverage map-based traffic graph construction to build an agent-centric encoder generalizable to any road network and traffic conditions, significantly enhancing downstream performance of reinforcement learning agents in urban traffic environments.

Deep Reinforcement Learning based Personalized Locomotion Planning for Lower-Limb Exoskeletons

Javad Khodaei-Mehr, M. Tavakoli

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed an intelligent central pattern generator (iCPG) method based on reinforcement learning for planning personalized gait trajectories for lower-limb exoskeletons

Deep Reinforcement Learning Based Tracking Control of an Autonomous Surface Vessel in Natural Waters

Wei Wang, D. Rus

Autonomous DrivingReinforcement Learning

🎯 What it does: Train and validate an ASV trajectory tracking controller based on deep reinforcement learning, and compare it with advanced nonlinear model predictive controllers

Deep Reinforcement Learning for Autonomous Driving using High-Level Heterogeneous Graph Representations

Maximilian Schier, B. Rosenhahn

Autonomous DrivingRepresentation LearningGraph Neural NetworkReinforcement LearningGraph

🎯 What it does: Propose an entity-relation framework that intuitively models road networks and traffic participants using heterogeneous graphs, converts them into homogeneous graphs for deep reinforcement learning training, and verifies its superiority over predefined handcrafted features in intersection scenarios.

Deep Underwater Monocular Depth Estimation with Single-Beam Echosounder

Haowen Liu, Alberto Quattrini Li

Data SynthesisDepth EstimationImage

🎯 What it does: Propose a self-supervised monocular depth estimation method for underwater environments using low-cost single-beam echo sounders (SBES) and generate a synthetic dataset.

Deep Unsupervised Visual Odometry Via Bundle Adjusted Pose Graph Optimization

G. Lu

Pose EstimationDepth EstimationAutonomous DrivingOptimizationSimultaneous Localization and MappingVideo

🎯 What it does: Introduce pose graphs and bundle adjustment optimization during the training process of an unsupervised visual odometry network, iteratively updating motion and depth estimation while enforcing unsupervised photometric and geometric constraints.

DeepRING: Learning Roto-translation Invariant Representation for LiDAR based Place Recognition

Shan Lu, Yue Wang

RetrievalContrastive LearningPoint Cloud

🎯 What it does: Propose the DeepRING method, which learns rotation-translation invariant representations by extracting sinogram features from LiDAR scans and aggregating magnitude spectra, achieving similar representations for the same location under different viewpoints, and treating location recognition as a one-to-one learning problem.

DeepSeeColor: Realtime Adaptive Color Correction for Autonomous Underwater Vehicles via Deep Learning Methods

S. Jamieson, Yogesh A. Girdhar

RestorationAutonomous DrivingComputational Efficiency

🎯 What it does: Proposed the DeepSeeColor algorithm for real-time adaptive underwater color correction.

DefGraspNets: Grasp Planning on 3D Fields with Graph Neural Nets

Isabella Huang, D. Fox

OptimizationComputational EfficiencyRobotic IntelligenceGraph Neural NetworkGraph

🎯 What it does: Proposed and trained a predictive graph neural network, DefGraspNets, for predicting 3D deformation and stress fields to accelerate grasping planning of deformable objects.

Demonstration-Bootstrapped Autonomous Practicing via Multi-Task Reinforcement Learning

Abhishek Gupta, Karol Hausman

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed a self-training system that utilizes multi-task reinforcement learning combined with prior data guidance to achieve continuous autonomous practice and reduce the number of resets.

Demonstration-guided Optimal Control for Long-term Non-prehensile Planar Manipulation

Teng Xue, S. Calinon

OptimizationRobotic IntelligenceReinforcement Learning from Human Feedback

🎯 What it does: Propose a hierarchical optimization framework based on demonstrations for offline task and motion planning, extending the pusher-slider dynamics model and utilizing human demonstrations for warm-start solving.

Demonstration-Guided Reinforcement Learning with Efficient Exploration for Task Automation of Surgical Robot

Tao Huang, Qingxu Dou

Robotic IntelligenceReinforcement LearningBiomedical Data

🎯 What it does: Proposed a reinforcement learning algorithm called DEX based on expert demonstrations to improve exploration efficiency in the automation of surgical robot tasks.

Dense Depth Completion Based on Multi-Scale Confidence and Self-Attention Mechanism for Intestinal Endoscopy

Ruyu Liu, Weiguo Sheng

Depth EstimationTransformerMultimodalityBiomedical Data

🎯 What it does: Leveraging sparse depth information, a deep learning method is proposed to complete the depth in colonoscopy, along with the design of a multi-scale confidence prediction network and a self-attention structure perception module.

DenseTact 2.0: Optical Tactile Sensor for Shape and Force Reconstruction

Won Kyung Do, Monroe Kennedy

RestorationDomain AdaptationRobotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: This paper proposes DenseTact 2.0, an optical tactile sensor that visualizes the deformed surface of a soft fingertip and achieves calibrated shape reconstruction and 6-axis torque estimation through a neural network.

Density Planner: Minimizing Collision Risk in Motion Planning with Dynamic Obstacles using Density-based Reachability

Laura Lutzow, Chuchu Fan

Autonomous DrivingOptimizationRobotic Intelligence

🎯 What it does: Proposes a density-based motion planning method that uses neural networks and the Liouville equation to learn density evolution under uncertain initial states, and minimizes collision risk through gradient optimization.

Density-aware NeRF Ensembles: Quantifying Predictive Uncertainty in Neural Radiance Fields

Niko Sünderhauf, Dimity Miller

Explainability and InterpretabilityNeural Radiance FieldBenchmark

🎯 What it does: Proposes a density-aware NeRF ensemble method that quantifies predictive uncertainty by considering the termination probability along rays.

Depth Estimation for Oral Cavity by Shape from Shading with Endoscope

Xi Wu, G. Zheng

Depth EstimationImage

🎯 What it does: Recover depth information of the oral environment using a low-cost endoscope and employing Shape from Shading (SFS) technique

Depth Is All You Need for Monocular 3D Detection

Dennis Park, Adrien Gaidon

Object DetectionDepth EstimationDomain AdaptationAutonomous DrivingVideoPoint Cloud

🎯 What it does: Propose to fine-tune depth estimation using LiDAR or RGB videos in an unsupervised manner to enhance monocular 3D detection performance.

Descriptor Distillation for Efficient Multi-Robot SLAM

Xiyue Guo, Guofeng Zhang

Computational EfficiencyKnowledge DistillationRobotic IntelligenceSimultaneous Localization and Mapping

🎯 What it does: Generating low-latency and discriminative compressed feature descriptors in multi-robot SLAM to achieve accurate localization and reduce communication bandwidth

Design and characterization of a low mechanical loss, high-resolution wearable strain gauge

Addison Liu, R. Wood

🎯 What it does: Designed and characterized a low mechanical loss, high-resolution wearable strain sensor, improved the SCARS mechanism, and adopted a new process using silicone materials and a liquid elastomer adhesive layer.

Design and Control of a Micro Overactuated Aerial Robot with an Origami Delta Manipulator

E. Cuniato, R. Siegwart

Robotic Intelligence

🎯 What it does: Designed and controlled a small, lightweight micro aerial robot, integrating an 8-DOF platform and a 3-DOF foldable triangular arm that can be independently controlled.

Design and Control of a Tunable-Stiffness Coiled-Spring Actuator

Shivangi Misra, C. Sung

Robotic IntelligencePhysics Related

🎯 What it does: Proposed a lightweight, compact adjustable stiffness coil spring actuator capable of achieving up to 20 times stiffness variation

Design and Development of a Hydrogel-based Soft Sensor for Multi-Axis Force Control

Yichen Cai, T. G. Thuruthel

Robotic Intelligence

🎯 What it does: Designed and developed a multi-axis soft sensor based on gelatin ionic hydrogel

Design and Development of a Novel Force-Sensing Robotic System for the Transseptal Puncture in Left Atrial Catheter Ablation

A. Zeidan, K. Rhode

Robotic Intelligence

🎯 What it does: Proposed and developed a novel force-sensing robotic system for transseptal puncture (TSP) in left atrial catheter ablation, integrating optical sensors to measure tissue contact force and puncture force.

Design and Evaluation of an Augmented Reality Head-Mounted Display User Interface for Controlling Legged Manipulators

Rodrigo Chacón Quesada, Y. Demiris

Robotic Intelligence

🎯 What it does: Designed and evaluated an augmented reality head-mounted display (AR HMD) user interface for controlling a leg manipulator, and compared it with existing control methods.

Design and Mechanics of Cable-Driven Rolling Diaphragm Transmission for High-Transparency Robotic Motion

Hoi Man Lam, Michael C. Yip

Robotic Intelligence

🎯 What it does: Designed and studied a coaxial opposed rolling membrane drive system with cable-driven mechanisms, proposing mechanical characteristics to achieve force balance, decoupling of transmission pressure and bearing load, and cable tension maintenance, while providing automated procedures for drive configuration, phase, and operation, and experimentally verifying the prototype's performance.

Design and Validation of a Multi-Arm Relocatable Manipulator for Space Applications

E. Hoffman, N. Tsagarakis

OptimizationRobotic Intelligence

🎯 What it does: Computational design and verification of the three-arm space robot Multi-Arm Relocatable Manipulator (MARM), proposing a comprehensive design and verification process, comparing different arm designs to ensure their ability to perform complex space maneuvers and operational tasks.

Design of a Multimodal Fingertip Sensor for Dynamic Manipulation

Andrew SaLoutos, Sangbae Kim

Robotic IntelligenceMultimodalityTime Series

🎯 What it does: Designed and implemented a spherical contact sensor based on a barometer and time-of-flight (TOF) sensors, and experimentally validated it in dynamic manipulation tasks

Design of a Variable Stiffness Spring with Human-Selectable Stiffness

Chase W. Mathews, D. Braun

🎯 What it does: Designed and verified a variable stiffness spring that allows users to adjust stiffness manually, and demonstrated its assistive effect in hip joint movement through leg swing experiments.

Design of an Energy-Aware Cartesian Impedance Controller for Collaborative Disassembly

Sebastian Hjorth, Arash Ajoudani

Safty and PrivacyRobotic Intelligence

🎯 What it does: Proposed and implemented an energy-aware Cartesian impedance controller for human-robot collaborative disassembly tasks.

Design Optimization and Data-driven Shallow Learning for Dynamic Modeling of a Smart Segmented Electroadhesive Clutch

N. Feizi, Rajnikant V. Patel

OptimizationData-Centric LearningRecurrent Neural Network

🎯 What it does: A segmented electrode design is proposed to modulate the electric field on the medium's surface using DC signals while maintaining low power consumption, thereby preventing electroadhesive decay, and design optimization is conducted based on a novel analytical model; meanwhile, a hybrid shallow learning method is developed by combining long short-term memory networks (LSTM) with the analytical model for dynamic modeling, and the performance of the semi-active clutch and data-driven hybrid model is validated through experiments.

Detecting spatio-temporal Relations by Combining a Semantic Map with a Stream Processing Engine

Lennart Niecksch, T. Wiemann

Simultaneous Localization and Mapping

🎯 What it does: Integrate semantic maps with a stream processing framework to perform real-time analysis of objects' spatiotemporal relationships, and implement a prototype system for tracking daily events in an office.

Development and Evaluation of a Robotic Vessel Positioning System for Semi-Automatic Microvascular Anastomosis

Jesse Haworth, Axel Krieger

Robotic IntelligenceBiomedical Data

🎯 What it does: Developed and evaluated a robotic vascular positioning system integrating a suturing robot and tissue positioning system, capable of performing semi-automated microvascular anastomosis on synthetic vessels

Development and Experimental Verification of a 3D Dynamic Absolute Nodal Coordinate Formulation Model of Flexible Prostate Biopsy/Brachytherapy Needles

Athanasios Martsopoulos, Antonia Tzemanaki

OptimizationComputational EfficiencyRobotic IntelligenceBiomedical Data

🎯 What it does: Developed and experimentally validated a 3D dynamic absolute node coordinate expression model for a flexible needle applicable to prostate biopsy/radiation therapy

Development of Hydraulically-driven Soft Hand for Handling Heavy Vegetables and its Experimental Evaluation

Osamu Azami, Ko Yamamoto

Robotic IntelligenceAgriculture Related

🎯 What it does: Developed a hydraulically driven flexible robotic hand for handling heavy vegetables in a vegetable factory, with experimental validation conducted.

DexGraspNet: A Large-Scale Robotic Dexterous Grasp Dataset for General Objects Based on Simulation

Ruicheng Wang (Peking University), He Wang (Peking University)

Data SynthesisRobotic IntelligenceBenchmark

🎯 What it does: Generated a large-scale robotic multi-finger grasp dataset called DexGraspNet, containing 1.32 million grasp samples, covering 5,355 objects across 133 categories.

Dexterous Imitation Made Easy: A Learning-Based Framework for Efficient Dexterous Manipulation

Sridhar Pandian Arunachalam, Lerrel Pinto

Robotic IntelligenceImage

🎯 What it does: Proposes the Dexterous Imitation Made Easy (DIME) framework, which uses a single RGB camera to capture teleoperation demonstrations by human operators, followed by training multi-finger grasping control strategies through advanced imitation learning methods, achieving complex in-hand manipulation tasks such as flipping and rotating on real robots;

Dexterous Manipulation from Images: Autonomous Real-World RL via Substep Guidance

Kelvin Xu, S. Levine

Robotic IntelligenceReinforcement LearningImage

🎯 What it does: Developed a vision-based multi-fingered robotic hand system for programming-free manipulation, allowing users to define final tasks and intermediate subtasks through image examples. The system uses reinforcement learning to autonomously perform multi-stage object manipulation tasks.

DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality

Ankur Handa, Gavriel State

Pose EstimationDomain AdaptationRobotic IntelligenceReinforcement Learning

🎯 What it does: Trained policies capable of executing robust dexterous manipulation in real-world environments, along with a reliable pose estimator for real-time acquisition of manipulated object states. The policy, trained under multiple conditions using Allegro Hand and Isaac Gym GPU simulation, was migrated to real-world environments;

Dextrous Tactile In-Hand Manipulation Using a Modular Reinforcement Learning Architecture

Johannes Pitz, Berthold Bauml

Domain AdaptationRobotic IntelligenceReinforcement Learning

🎯 What it does: Feasibility of reorienting a cube to any of the 24 possible target poses was achieved using the torque-controlled DLR-Hand II hand without external sensors.

DFR-FastMOT: Detection Failure Resistant Tracker for Fast Multi-Object Tracking Based on Sensor Fusion

Mohamed Nagy, S. Javed

Object TrackingMultimodalityPoint Cloud

🎯 What it does: Propose a lightweight multi-object tracking method DFR-FastMOT based on camera and LiDAR sensor fusion, utilizing algebraic target association and fusion to achieve long-term memory for handling occlusions.

DifFAR: Differentiable Frequency-based Disentanglement for Aerial Video Action Recognition

D. Kothandaraman, Dinesh Manocha

RecognitionVideo

🎯 What it does: Propose the DifFAR algorithm for action recognition in UAV videos

Differentiable Collision Detection for a Set of Convex Primitives

K. Tracy, Zachary Manchester

Optimization

🎯 What it does: Proposes DCOL, a fully differentiable collision detection framework for a set of composite convex primitives, solving convex optimization via minimal uniform scaling to determine collisions.

Differentiable Collision Detection: a Randomized Smoothing Approach

Louis Montaut, Justin Carpentier

Computational Efficiency

🎯 What it does: Proposed a generic and efficient method to compute the derivatives of collision detection between any two convex shapes.

Differentiable Dynamics Simulation Using Invariant Contact Mapping and Damped Contact Force

Minji Lee, Dongjun Lee

OptimizationRobotic IntelligenceOrdinary Differential Equation

🎯 What it does: Proposes a differentiable contact dynamics framework that includes invariant contact sets and uses the signed distance function (SDF) for coordinate differentiation, while introducing a damped contact model to eliminate jitter and additional local minima caused by smoothing.

Differentiable Parsing and Visual Grounding of Natural Language Instructions for Object Placement

Zirui Zhao, David Hsu

Graph Neural NetworkVision-Language-Action ModelText

🎯 What it does: Propose a method called PARAGON that uses differentiable parsing to convert natural language instructions into object-centric graph representations, and employs particle-based graph neural networks for object placement under uncertainty.

Differential Dynamic Programming based Hybrid Manipulation Strategy for Dynamic Grasping

Cheng Zhou, Yu Zheng

OptimizationRobotic IntelligenceReinforcement Learning

🎯 What it does: This paper proposes a complete dynamic grasping process, from approaching the target object, collision, rolling, to finally achieving grasping through mastery or fingertip grasp, enabling smooth grasping by the robot's end-effector; a unified model is derived through spatial vectorized link dynamics, and the entire process is formulated as a free-endpoint constrained multi-phase optimal control problem; subsequently, differential dynamic programming (DDP) is extended to solve this free-endpoint OCP, with constrained quadratic programming (QP) introduced in the backward pass and solved using the primal-dual augmented Lagrangian (PDAL) method; experiments and simulations verify the effectiveness of the proposed method.

Dimensional Optimization and Anti-Disturbance Analysis of an Upgraded Feed Mechanism in FAST

Xiaoyan Wang, Weiwei Shang

OptimizationPhysics Related

🎯 What it does: Designed and optimized an upgraded feed mechanism for lightweight cable structures to enhance the observation angle and anti-disturbance capability of the FAST telescope.

Direct and inverse modeling of soft robots by learning a condensed FEM model

Etienne Ménager, Christian Duriez

Knowledge DistillationRepresentation LearningRobotic Intelligence

🎯 What it does: This study proposes a learning-based FEM model compression method, using nonlinear compliance data to construct compact forward and inverse kinematic models for soft robots, and verifies its learnability and efficiency on a two-finger gripper.

Direct Angular Rate Estimation Without Event Motion-Compensation At High Angular Rates

M. Ng, S. Foong

Pose EstimationSequential

🎯 What it does: Propose a Fourier transform-based angular velocity estimator that can directly estimate high angular velocity from event camera images without motion compensation.

Direct LiDAR-Inertial Odometry: Lightweight LIO with Continuous-Time Motion Correction

Kenny Chen, B. Lopez

Autonomous DrivingComputational EfficiencySimultaneous Localization and MappingPoint Cloud

🎯 What it does: Developed DLIO, a lightweight LiDAR-inertial odometry algorithm that achieves precise localization using continuous-time motion correction and point-wise back-warping.

DisCo: A Multiagent 3D Coordinate System for Lattice Based Modular Self-Reconfigurable Robots

Benoît Piranda, J. Bourgeois

Robotic IntelligenceAgentic AI

🎯 What it does: Established a three-dimensional grid coordinate system based on a multi-agent system, utilizing the DisCo algorithm to achieve module localization and orientation inference for modular self-reconfiguring robots.

Discovering Multiple Algorithm Configurations

L. Keselman, M. Hebert

OptimizationHyperparameter SearchRobotic Intelligence

🎯 What it does: Expand algorithm configurations to automatically discover multiple patterns in tuning datasets, and propose three pattern discovery methods: posterior method, multi-stage method, and online algorithm based on multi-armed bandit.

Discrete-time model based control of soft manipulator with FBG sensing

Enrico Franco, F. Baena

Robotic Intelligence

🎯 What it does: Studied discrete-time model-based control for planar soft continuum manipulators using fiber Bragg grating (FBG) sensing, and designed a discrete energy regulation control algorithm considering digital control lag, employing a nonlinear observer to estimate uncertain bending stiffness and compensate for constant matched disturbances.

Discriminative 3D Shape Modeling for Few-Shot Instance Segmentation

A. Cherian, Alan Sullivan

SegmentationMeta LearningImagePoint Cloud

🎯 What it does: By computing surface geodesics between 3D points in depth images and using a neural network to determine whether geodesics belong to the same instance, few-shot instance segmentation is achieved.

Distributed barrier function-enabled human-in-the-loop control for multi-robot systems

Victor Nan Fernandez-Ayala, Dimos V. Dimarogonas

OptimizationRobotic Intelligence

🎯 What it does: Propose a distributed control scheme that utilizes control barrier functions to simultaneously handle multiple constraints in multi-robot systems and integrates human-robot interaction control;

Distributed Data-Driven Predictive Control for Multi-Agent Collaborative Legged Locomotion

Randall T. Fawcett, K. Hamed

OptimizationRobotic Intelligence

🎯 What it does: Proposed a distributed data-driven predictive control planner that enables quadruped robots to achieve robust gait in complex environments through collaborative multi-robot systems.

Distributed Initialization for Visual-Inertial-Ranging Odometry with Position-Unknown UWB Network

Shenhan Jia, Yue Wang

OptimizationRobotic IntelligenceSimultaneous Localization and MappingMultimodality

🎯 What it does: Proposed a distributed initialization method for position-unknown UWB networks, achieving continuous estimation of visual-inertial-ranging (VIR) odometry;

Distributed Model Predictive Formation Control with Gait Synchronization for Multiple Quadruped Robots

Shaohang Xu, C. Ho

OptimizationRobotic Intelligence

🎯 What it does: Proposed a complete distributed model predictive control (MPC) framework for multiple quadruped robots to achieve synchronized gaits, obstacle avoidance, and navigation in obstacle-filled environments;

Distributed Potential iLQR: Scalable Game-Theoretic Trajectory Planning for Multi-Agent Interactions

Zach Williams, Negar Mehr

OptimizationRobotic Intelligence

🎯 What it does: Developed a scalable local trajectory optimization algorithm enabling robots to interact with other robots.

Distributional Instance Segmentation: Modeling Uncertainty and High Confidence Predictions with Latent-MaskRCNN

Yuxuan Liu, Xi Chen

SegmentationRobotic IntelligenceImage

🎯 What it does: Explore distributed instance segmentation models using latent codes, and propose a confidence mask method for robotic grasping, significantly reducing critical errors.

Distributionally Robust Optimization with Unscented Transform for Learning-Based Motion Control in Dynamic Environments

A. Hakobyan, Insoon Yang

Autonomous DrivingOptimization

🎯 What it does: Enhancing the robustness of learning-based motion controllers using distributionally robust optimization and the unscented transform

Distributionally Robust RRT with Risk Allocation

Kajsa Ekenberg, Björn Olofsson

OptimizationRobotic Intelligence

🎯 What it does: Integrate distributed robust risk allocation into a sample-based motion planning algorithm, achieving non-uniform risk allocation by decomposing the joint risk constraints over the entire planning horizon into individual risk constraints; define an exact risk allocation process through deterministic tightening based on individual risk constraints; embed this technique into sample-based motion planning, generating trajectories that are both conservative and increasingly feasible, thereby enhancing search efficiency in the state space.

Disturbance Observer Based Contact Detection for Motorized Hydraulic Actuators

Chunpeng Wang, J. P. Whitney

Robotic Intelligence

🎯 What it does: Demonstrated contact detection for remotely operated hydraulic-driven robotic grippers without endpoint tactile sensors or joint position sensors, using a disturbance observer, achieving a maximum endpoint impedance bandwidth of 40 dB and a contact detection threshold of 0.2N

Ditto in the House: Building Articulation Models of Indoor Scenes through Interactive Perception

Cheng-Chun Hsu, Yuke Zhu

Robotic IntelligenceImage

🎯 What it does: By leveraging interactive perception in indoor environments, the paper constructs a joint model of manipulable objects. It first identifies articulatable objects using sensitivity prediction, then interacts with them to generate motion, and subsequently infers their joint attributes by analyzing visual observations before and after the interaction.

Diver Interest via Pointing: Human-Directed Object Inspection for AUVs

Chelsey Edge, Junaed Sattar

Object DetectionPose EstimationRobotic IntelligenceConvolutional Neural NetworkImage

🎯 What it does: Proposes the DIP algorithm, which utilizes the pointing gestures of divers to enable an AUV to perceive and locate regions of interest through a single camera.

DLOFTBs – Fast Tracking of Deformable Linear Objects with B-splines

Piotr Kicki, K. Walas

Object TrackingImage

🎯 What it does: Proposes a fast tracking algorithm for deformable linear objects (DLO) based on masked images. First, the mask is skeletonized, then skeleton segments are sequentially traversed and arranged into a path, followed by fitting a B-spline curve to represent the DLO shape.

DMMGAN: Diverse Multi Motion Prediction of 3D Human Joints using Attention-Based Generative Adversarial Network

Payam Nikdel, Mo Chen

GenerationPose EstimationTransformerGenerative Adversarial NetworkPoint Cloud

🎯 What it does: Propose a Transformer-based generative model for predicting diverse future 3D human actions, including absolute 3D trajectories.