arXivSub Start free trial

ICRA 2025 Papers — Page 5

IEEE International Conference on Robotics and Automation · 1604 papers

DOGE: An Extrinsic Orientation and Gyroscope Bias Estimation for Visual-Inertial Odometry Initialization

Zewen Xu, Yihong Wu

Pose EstimationOptimizationSimultaneous Localization and MappingMultimodality

🎯 What it does: Proposes a new visual-inertial odometry (VIO) initialization method that can jointly estimate the external orientation and gyroscope bias under pure rotational motion, and achieves precise initialization before sufficient translational parallax is obtained through weighted and failure detection strategies combined with maximum a posteriori (MAP) optimization.

Domain Adaptation-Based Crossmodal Knowledge Distillation for 3D Semantic Segmentation

Jialiang Kang, Dingsheng Luo

SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkMultimodalityPoint Cloud

🎯 What it does: Propose two cross-modal knowledge distillation methods (UDAKD and FSKD), leveraging synchronized camera and LiDAR data to achieve 3D point cloud semantic segmentation without 3D annotations.

Domain Randomization for Object Detection in Manufacturing Applications Using Synthetic Data: A Comprehensive Study

Xiaomeng Zhu, Atsuto Maki

Object DetectionData SynthesisDomain AdaptationConvolutional Neural NetworkImage

🎯 What it does: Studied domain randomization methods for object detection in manufacturing, constructed a complete data generation pipeline, and proposed the SIP15-OD dataset.

DoorBot: Closed-Loop Task Planning and Manipulation for Door Opening in the Wild with Haptic Feedback

Zhi Wang, Wenzhen Yuan

Robotic Intelligence

🎯 What it does: Developed a closed-loop hierarchical control framework based on tactile feedback, enabling robots to explore and open doors they have not encountered before.

Doppler Former: Velocity Supervision of Raw Radar Data

Shuo Zhao, Zhaoyi Jiang

Autonomous DrivingConvolutional Neural NetworkPoint Cloud

🎯 What it does: Propose the Doppler Former (DPF) module to efficiently extract velocity information from raw radar data, and propose a Fully Complex Convolutional Network (FCCN) backbone network that is more suitable for raw data; integrate DPF into FCCN to enhance the performance of downstream radar perception tasks.

DOPT: D-Learning with Off-Policy Target toward Sample Efficiency and Fast Convergence Control

Zhaolong Shen, Quan Quan

Reinforcement Learning

🎯 What it does: Proposes an offline policy variant of D-learning that iteratively improves the performance of neural network controllers by leveraging current and historical data within the Lyapunov theory framework.

DP-Habitat: Bridging the Gap Between Simulation and Reality for Visual Navigation in Dynamic Pedestrian Environments

Liang Qin, Houqiang Li

Autonomous DrivingBenchmark

🎯 What it does: Developed the DP-Habitat dynamic pedestrian simulator on the Habitat platform, and proposed the Adaptive Object Navigation with Dynamic Mapping (AON-DM) baseline method specifically designed for dynamic pedestrian environments.

Dragonfly Drone: A Novel Tilt-Rotor Aerial Platform with Body-Morphing Capability

Syed Waqar Hameed, Mir Feroskhan

Robotic Intelligence

🎯 What it does: Propose a novel tilt-rotor, shape-variable UAV called Dragonfly, which can alter its body shape and orientation while maintaining position tracking, supporting six degrees of freedom motion.

DreamDrive: Generative 4D Scene Modeling from Street View Images

Jiageng Mao, Yue Wang

GenerationData SynthesisAutonomous DrivingDiffusion modelGaussian SplattingImageVideo

🎯 What it does: Proposes DreamDrive, a 4D spatiotemporal scene generation method that combines the advantages of generation and reconstruction. It uses a video diffusion model to generate visual reference sequences, elevates them to 4D through a mixed Gaussian representation, and then renders 3D-consistent driving videos under a given driving trajectory using Gaussian splatting.

DreamFLEX: Learning Fault-Aware Quadrupedal Locomotion Controller for Anomaly Situation in Rough Terrains

Seunghyun Lee, Hyun Myung

Robotic Intelligence

🎯 What it does: Proposes DreamFLEX, a robust fault-tolerant gait controller that enables robots to walk in complex environments even when joints fail.

Drive with the Flow

Enrico Mannocci, Stefano Mattoccia

Autonomous DrivingOptical FlowBenchmark

🎯 What it does: Introduce an optical flow auxiliary task into the perception module of an end-to-end autonomous driving model to explicitly model short-term memory, and train the FlowFuser model using a new benchmark dataset generated by the CARLA simulator to verify its ability to avoid collisions in congested traffic.

DRIVE: Dependable Robust Interpretable Visionary Ensemble Framework in Autonomous Driving

Songning Lai, Yutao Yue

Autonomous DrivingExplainability and Interpretability

🎯 What it does: Designed and verified the DRIVE framework to enhance the interpretability reliability and stability of end-to-end unsupervised autonomous driving models, addressing the issue of unstable interpretation in DCG models.

DroneDiffusion: Robust Quadrotor Dynamics Learning with Diffusion Models

Avirup Das, Wei Pan

Robotic IntelligenceDiffusion model

🎯 What it does: Proposed the DroneDiffusion framework, which uses a conditional diffusion model to learn quadrotor dynamics and combines the learned dynamics with an adaptive controller to achieve trajectory tracking; verified its robustness through simulations and real flight experiments.

DROP: Dexterous Reorientation via Online Planning

Albert H. Li, Aaron D. Ames

Pose EstimationRobotic IntelligenceImage

🎯 What it does: Propose an online planning method using a sample-based predictive controller and a vision-based pose estimator to achieve contact-rich control for in-hand cube reorientation tasks.

DTRT: Enhancing Human Intent Estimation and Role Allocation for Physical Human-Robot Collaboration

Haotian Liu, Zhengtao Zhang

Robotic IntelligenceTransformerAuto EncoderSequential

🎯 What it does: Propose a Dual Transformer-based Robot Trajectron (DTRT) that rapidly captures intent changes by utilizing human-guided motion and mechanics data, enabling multi-step trajectory prediction and dynamic robot behavior adjustment to enhance the safety and efficiency of physical human-robot collaboration.

Dual-AEB: Synergizing Rule-Based and Multimodal Large Language Models for Effective Emergency Braking

Wei Zhang, Yilun Chen

Autonomous DrivingTransformerLarge Language ModelMultimodality

🎯 What it does: Proposed the Dual-AEB system, which integrates advanced multimodal large language models (MLLM) with traditional rule-based fast AEB to enhance adaptability in open scenarios.

Dual-BEV Nav: Dual-Layer BEV-Based Heuristic Path Planning for Robotic Navigation in Unstructured Outdoor Environments

Jianfeng Zhang, Xiong You

OptimizationRobotic Intelligence

🎯 What it does: Proposed the Dual-BEV Nav method, which employs local and global Bird's Eye View (BEV) representations for two-layer heuristic path planning, enabling long-distance navigation for robots in unstructured outdoor environments.

Dual-Conditioned Temporal Diffusion Modeling for Driving Scene Generation

Xiangyu Bai, Sarah Ostadabbas

GenerationData SynthesisAutonomous DrivingDiffusion modelVideoMultimodality

🎯 What it does: Proposes the Dual-Conditioned Temporal Diffusion Model (DcTDM) for generating realistic long-term driving videos, and simultaneously constructs the DriveSceneDDM driving video dataset containing text descriptions, dense depth maps, and Canny edge data.

Dualdiff: Dual-Branch Diffusion Model for Autonomous Driving with Semantic Fusion

Haoteng Li, Longjun Liu

Autonomous DrivingDiffusion model

🎯 What it does: Propose the DualDiff dual-branch conditional diffusion model for multi-view driving scene generation, introducing Occupancy Ray Sampling, Semantic Fusion Attention, and foreground-aware mask loss.

Duolingo: Dynamics Utilization for Online Translation of Actions

Karthikeya Vemuri, Abhishek Gupta

Robotic IntelligenceWorld Model

🎯 What it does: Proposed an efficient algorithm for modeling and compensating system dynamics changes in damaged or hysteresis-prone non-rigid systems, achieving continuous learning and transfer through a simple calibration process that learns non-linear action translation models.

Dur360BEV: A Real-World 360-Degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving

E. Wenke, T. Breckon

Autonomous DrivingImageMultimodalityPoint CloudBenchmark

🎯 What it does: Constructed a dataset for autonomous driving named Dur360BEV based on a single 360-degree panoramic camera, 128-channel high-resolution 3D LiDAR, and RTK-GNSS/INS, and proposed a corresponding BEV generation benchmark architecture.

DVLO4D: Deep Visual-Lidar Odometry with Sparse Spatial-Temporal Fusion

Mengmeng Liu, Hao Cheng

Pose EstimationAutonomous DrivingSimultaneous Localization and MappingImageMultimodalityPoint Cloud

🎯 What it does: Propose a deep visual-lidar odometry framework called DVLO4D to enhance localization accuracy and robustness.

DVM-SLAM: Decentralized Visual Monocular Simultaneous Localization and Mapping for Multi-Agent Systems

Joshua Bird, Amanda Prorok

Robotic IntelligenceSimultaneous Localization and MappingImage

🎯 What it does: This paper proposes and implements the first open-source distributed monocular visual cooperative localization and mapping system (DVM-SLAM), which is verified on a physical robot equipped with a custom collision avoidance framework.

DVN-SLAM: Dynamic Visual Neural Slam Based on Local-Global Encoding

Wenhua Wu, Hesheng Wang

Simultaneous Localization and Mapping

🎯 What it does: Proposed a dynamic visual SLAM system DVN-SLAM based on local-global fused neural implicit representation.

DVS-Aware Visual Perception for Pose Estimation of Mobile Robots with Neuromorphic Implementation

Hanzhong Zhong, Xiang Li

Pose EstimationSpiking Neural Network

🎯 What it does: A DVS visual perception method for mobile robot pose estimation was developed, incorporating novel markers, a recognition algorithm based on Spiking Convolutional Neural Networks (SCNN), and an integrated implementation of a neuromorphic computing accelerator.

Dy3DGS-SLAM: Monocular 3D Gaussian Splatting SLAM for Dynamic Environments

Mingrui Li, Ahmad Osman

Pose EstimationDepth EstimationGaussian SplattingOptical FlowImage

🎯 What it does: Propose Dy3DGS-SLAM, a monocular RGB input-based 3D Gaussian Splatting SLAM method for localization and reconstruction in dynamic environments.

Dynamem: Online Dynamic Spatio-Semantic Memory for Open World Mobile Manipulation

Peiqi Liu, Lerrel Pinto

Robotic IntelligenceLarge Language ModelVision Language ModelPoint Cloud

🎯 What it does: This paper proposes DynaMem, an online dynamic spatial semantic memory framework for open-world mobile manipulation;

Dynamic Bipedal MPC with Foot-Level Obstacle Avoidance and Adjustable Step Timing

Tianze Wang, Christian Hubicki

OptimizationRobotic Intelligence

🎯 What it does: Proposed a real-time MPC framework to achieve collision-free motion planning for dynamic bipedal robots in irregular environments, capable of simultaneously avoiding body and foot collisions.

Dynamic Compact Consensus Tracking for Aerial Robots

Xiaolou Sun, Yongming Huang

Object TrackingRobotic IntelligenceTransformer

🎯 What it does: Propose a dynamic compact consensus tracker DC2T, adopting Compact Token Encoder and Dynamic Consensus Attention to achieve real-time tracking

Dynamic End Effector Trajectory Tracking for Small-Scale Underwater Vehicle-Manipulator Systems (UVMS): Modeling, Control, and Experimental Validation

Niklas Trekel, R. Seifried

Robotic Intelligence

🎯 What it does: Implement dynamic end-effector trajectory tracking on a small underwater vehicle-manipulator system (UVMS), and complete modeling, control, and experimental validation using a task-priority control approach.

Dynamic Gap: Safe Gap-based Navigation in Dynamic Environments

Max Asselmeier, P. Vela

Robotic IntelligenceBenchmark

🎯 What it does: Extend gap-based local planner in unknown dynamic environments to generate a provably collision-safe hierarchical navigation system.

Dynamic Mode Decomposition with Sonomyography and Electromyography for Predictive Modeling of Lower Limb Exoskeleton Walking

Krysten Lambeth, Nitin Sharma

Computational EfficiencyRobotic IntelligenceBiomedical DataUltrasound

🎯 What it does: A linearized model based on Koopman was constructed, utilizing muscle activity indicators measured by EMG and sonomyography (ultrasound imaging) to predict joint angles in lower-limb exoskeleton gait assistance, thereby reducing the computational burden of multi-joint nonlinear dynamics.

Dynamic Multi-Objective Ergodic Path Planning Using Decomposition Methods

Abigail Breitfeld, David Wettergreen

Optimization

🎯 What it does: Proposed a dynamic multi-objective trajectory planning method utilizing the Boundary Intersection Decomposition Technique, achieving adaptive planning in response to changes in multiple objectives.

Dynamic Non-Prehensile Object Transport via Model-Predictive Reinforcement Learning

Neel Jawale, M. Bhardwaj

Robotic IntelligenceReinforcement Learning

🎯 What it does: This study enables a robot manipulator to perform dynamic non-grasping object transportation (i.e., robot server task) by learning from limited real demonstrations, combining batch reinforcement learning with model predictive control (MPC).

Dynamic Object Goal Pushing with Mobile Manipulators Through Model-Free Constrained Reinforcement Learning

Ioannis Dadiotis, Marco Hutter

Robotic IntelligenceReinforcement Learning

🎯 What it does: Developed a learning-based mobile manipulator controller that moves unknown objects to target positions and orientations through a series of pushing actions.

Dynamic Perception-Enhanced Motion Planning and Control for UAVs Flights in Challenging Dynamic Environments

Luyao Liu, Hong Zhang

Autonomous DrivingOptimization

🎯 What it does: Propose a complete system to enable safe autonomous flight of drones in unknown crowded environments, including 3D dynamic Euclidean signed distance field (ESDF) mapping, trajectory planning, and control.

Dynamic Tube MPC: Learning Tube Dynamics with Massively Parallel Simulation for Robust Safety in Practice

William D. Compton, Aaron D. Ames

OptimizationRobotic IntelligenceWorld Model

🎯 What it does: Propose a dynamic pipe MPC method that utilizes large-scale parallel simulation to learn dynamic pipe representations, and then optimizes trajectory planning based on dynamic pipe constraints for the 3D jumping robot ARCHER to achieve safe navigation.

Dynamically Feasible Path Planning in Cluttered Environments via Reachable BéZier Polytopes

Noel Csomay-Shanklin, Aaron D. Ames

OptimizationRobotic Intelligence

🎯 What it does: Explore the use of reachable Bézier polytopes as an efficient tool for rapidly generating paths that satisfy kinematic and dynamic constraints, achieving real-time performance through GPU acceleration; propose a hierarchical control architecture and demonstrate its application in complex 3D jumping environments.

Dynamics Modeling Using Visual Terrain Features for High-Speed Autonomous Off-Road Driving

Jason Gibson, Patrick Spieler

Autonomous DrivingWorld ModelImage

🎯 What it does: A hybrid dynamic model was constructed, which utilizes visual input to predict dynamic changes caused by terrain, and an end-to-end trained projection distance-agnostic feature encoder was proposed to compress visual features, generating real-time lightweight maps.

DynORecon: Dynamic Object Reconstruction for Navigation

Yiduo Wang, Viorela Ila

Autonomous DrivingOptimizationSimultaneous Localization and MappingImageVideo

🎯 What it does: Generate a voxel map of moving entities and estimate free space using dynamic SLAM information to support navigation.

E2B: A Single Modality Point-Based Tracker with Event Cameras

Hongwei Ren, Bo-Xun Cheng

Object TrackingPoint Cloud

🎯 What it does: Propose the E2B monomodal point cloud tracker, which directly processes raw outputs from event cameras and utilizes coordinate guidance to map event cloud features to 2D bounding boxes.

E2Map: Experience-and-Emotion Map for Self-Reflective Robot Navigation with Language Models

Chan Kim, Seong-Woo Kim

Robotic IntelligenceTransformerLarge Language ModelSimultaneous Localization and Mapping

🎯 What it does: Proposed an experience and emotion map called E2Map to enhance performance in navigation tasks by enabling robots to update their own experiences in one go when executing language instructions.

EAR-SLAM: Environment-Aware Robust Localization System for Terrestrial-Aerial Bimodal Vehicles

Wenjun He, Yanjun Cao

Anomaly DetectionAutonomous DrivingSimultaneous Localization and MappingImageMultimodalityPoint Cloud

🎯 What it does: Proposed an environment-aware robust localization system for ground-air dual-mode vehicles, achieving precise positioning by fusing multi-sensor data.

Effective Heterogeneous Point Cloud-Based Place Recognition and Relative Localization for Ground and Aerial Vehicles

Rui Mao, Hui Cheng

Autonomous DrivingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Proposed a novel pipeline based on BEV density maps and an enhanced data structure for place recognition and relative localization in collaborative systems between ground and aerial robots, along with an efficient height alignment algorithm.

Effective Self-Righting Strategies for Elongate Multi-Legged Robots

Erik Teder, Daniel I. Goldman

Robotic IntelligenceVideo

🎯 What it does: Studied the self-righting strategy of multi-legged robots using a combined approach of biology and robotics experiments, recording the self-righting behavior of centipede-like animals, assuming it is described by two wave motions in the horizontal and vertical planes, verifying it on a static multi-legged robot, and evaluating the impact of waveform parameters on self-righting effectiveness.

Effective Tuning Strategies for Generalist Robot Manipulation Policies

Wenbo Zhang, Lingqiao Liu

Robotic IntelligenceSupervised Fine-Tuning

🎯 What it does: Conducted an in-depth empirical study to investigate the impact of key factors in the fine-tuning process of general-purpose robotic manipulation policies (GMPs), including action space, policy heads, supervision signals, and tunable parameters, with 2500 trials evaluated for each configuration.

Efficient 3D Perception on Multi-Sweep Point Cloud with Gumbel Spatial Pruning

Jianhao Li, Hengshuang Zhao

Autonomous DrivingComputational EfficiencyPoint Cloud

🎯 What it does: Studied the 3D point cloud perception problem from accumulated multi-frame LiDAR scans, and proposed the Gumbel Spatial Pruning (GSP) layer to dynamically prune redundant points to improve perception accuracy.

Efficient 7-DoF Grasp for Target-Driven Object in Dense Cluttered Scenes

Tianjiao Lei, Tao Jiang

Pose EstimationRobotic IntelligencePoint Cloud

🎯 What it does: Propose a data and model-agnostic, efficient 7-DoF grasping method that generates grasping configurations for arbitrary target objects using single-view point cloud data, and rapidly adjusts the grasping configurations through object detection and multi-region point cloud distribution perception, enabling real-time precise grasping by robots in dense cluttered environments; meanwhile, design a grasping framework to reduce the time consumed during grasping and improve the grasping efficiency for specified target objects.

Efficient and Diverse Generative Robot Designs Using Evolution and Intrinsic Motivation

L. L. Goff, Simón C. Smith

Robotic Intelligence

🎯 What it does: Propose a generative design method for robots that combines morphological evolution with intrinsic motivation, using a homeokinetic controller to replace the costly learning phase, thereby improving diversity and speed.

Efficient Collision Detection Framework for Enhancing Collision-Free Robot Motion

Xiankun Zhu, Bin Liang

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: Proposed an efficient collision detection framework based on robot Signed Distance Field (SDF), seamlessly integrated with a self-collision detection module; decomposed SDF via forward kinematics, used multiple groups of extremely lightweight networks to parallelly approximate SDF, and introduced support vector machines to unify SDF with self-collision detection for collision distance representation; optimized collision-free trajectories while maintaining the differentiable properties of the framework, and developed a reactive motion controller for real-time dynamic obstacle avoidance based on this framework.

Efficient Coordination and Synchronization of Multi-Robot Systems Under Recurring Linear Temporal Logic

D. Peron, Dimos V. Dimarogonas

Robotic Intelligence

🎯 What it does: Propose a bottom-up approach that combines offline plan synthesis with online coordination to achieve efficient planning for linear temporal logic (LTL) tasks in multi-robot systems. The method dynamically adjusts plans through real-time communication and introduces a synchronization mechanism to address action delays. The software is implemented in Python and ROS2, with experiments on 9 physical robots and simulations with up to 90 robots demonstrating its adaptability and scalability.

Efficient Cross-Boundary Grasping in Stacked Clutter with Single-Visual Mapping Multi-Step

Yudong Luo, Yantao Shen

Domain AdaptationRobotic IntelligenceReinforcement LearningPoint Cloud

🎯 What it does: Propose a grasping strategy based on single visual mapping multi-step (SVMMS), and design a multi-functional integrated Deep Q-learning network model to extract visual features, output hierarchical relationships of stacked objects, and quantify the relationship between action logic and RGB-D changes. Utilize time series and prioritized experience replay to achieve global action sequence optimization, while combining domain-randomized sim2real methods to address size differences between simulated and real objects.

Efficient Gradient-Based Inference for Manipulation Planning in Contact Factor Graphs

Jeongmin Lee, Dongjun Lee

Computational EfficiencyRobotic IntelligenceGraph Neural NetworkGraph

🎯 What it does: Developed an operational planning framework based on contact factor graphs (CFG), utilizing graph models to reason about contact and dynamic constraints, generating feasible solutions.

Efficient Imitation Without Demonstrations via Value-Penalized Auxiliary Control from Examples

Trevor Ablett, Jonathan Kelly

Reinforcement Learning

🎯 What it does: Proposed an algorithm called VPACE, which significantly improves exploration efficiency and maintains the boundedness of value estimates by incorporating simple auxiliary task examples and value penalties into example-based control.

Efficient Non-Myopic Layered Bayesian Optimization for Large-Scale Bathymetric Informative Path Planning

A. Kiessling, J. Folkesson

OptimizationRobotic Intelligence

🎯 What it does: Proposed a dual-layer Bayesian optimization-based non-greedy online IP method for path planning in large-scale bathymetric mapping.

Efficient Online Learning of Contact Force Models for Connector Insertion

K. Tracy, Yuval Tassa

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: This paper proposes a quasi-static contact force model learning method for connector insertion tasks involving rich contact and hard friction. It constructs a linear mapping using feature vectors containing configuration and control information, and solves the global optimal mapping in real-time without matrix inversion using a new linear model learning algorithm.

Efficient Optimization of a Permanent Magnet Array for a Stable 2D Trap

Ann-Sophia Müller, T. Qiu

OptimizationRobotic IntelligenceBiomedical DataPhysics Related

🎯 What it does: Proposed a two-dimensional stable magnetic force capture using a permanent magnet array for wireless manipulation of biomedical microrobots.

Efficient Path Planning in Complex Environments with Trust Region Continuous Belief Tree Search

Andre Nuñez, Robert Fitch

Autonomous DrivingOptimization

🎯 What it does: Proposes a trust region-based adaptive control duration continuous belief tree search algorithm (TR-CBTS) for path planning in environments with complex constraints and objective functions.

Efficient Scale-Uniform 3D Visual Coverage Algorithm for UAV Based on Elastic Photogrammetric Constraints

J. Zong, Hongpeng Wang

Autonomous DrivingOptimizationImage

🎯 What it does: Proposed a UAV 3D visual coverage algorithm compatible with existing generic visual algorithms, which maintains uniform image scale for ground targets. The algorithm generates waypoints using photogrammetric constraints and introduces an Elastic Photogrammetric Constraint (EPC) to address valley aggregation issues caused by mountainous terrain.

Efficient Second-Order Cone Programming for the Close Enough Traveling Salesman Problem

Geordan Gutow, H. Choset

OptimizationComputational Efficiency

🎯 What it does: Improve the second-order cone programming (SOCP) solution method for CETSP by reducing variables and constraints, reusing computation and memory, and adopting warm-start strategies to enhance solving efficiency.

Efficient Submap-based Autonomous MAV Exploration using Visual-Inertial SLAM Configurable for LiDARs or Depth Cameras

Sotiris Papatheodorou, Stefan Leutenegger

Computational EfficiencyRobotic IntelligenceSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Propose a UAV autonomous exploration framework based on local submaps, which utilizes loop closure correction to maintain global consistency, efficiently constructs global frontiers through local submap frontiers, employs a sample-based next best view planner, and supports LiDAR or depth cameras.

Efficient Trajectory Generation Based on Traversable Planes in 3D Complex Architectural Spaces

Mengke Zhang, Yanjun Cao

OptimizationRobotic IntelligencePoint Cloud

🎯 What it does: Proposed a planner based on efficient trajectory generation using traversable planes for autonomous navigation of ground robots in complex multi-level building spaces.

Efficiently Generating Expressive Quadruped Behaviors via Language-Guided Preference Learning

Jaden Clark, Dorsa Sadigh

Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language Model

🎯 What it does: Propose a preference learning method guided by language, which uses a pre-trained LLM to generate initial behavior samples and refines them through human preference feedback, to achieve efficient generation of expressive quadruped robot behaviors.

Ego-$A^{\mathbf{3}}$: Adaptive Fusion-Based Disentangled Transformer for Egocentric Action Anticipation

Minhyuk Kim, S. Yoo

RecognitionTransformerVideo

🎯 What it does: Proposed the Ego-A³ model, which improves action prediction from the wearable camera perspective using an adaptive fusion and separation Transformer architecture;

EgoMimic: Scaling Imitation Learning via Egocentric Video

Simar Kareer, Danfei Xu

Domain AdaptationRobotic IntelligenceReinforcement LearningVideo

🎯 What it does: Propose the EgoMimic framework, which leverages front-view human videos combined with 3D hand tracking data for human-like imitation learning to expand robotic manipulation capabilities; and achieves unified utilization of human and robot data through low-cost dual-arm manipulators, cross-domain data alignment techniques, and a joint training architecture.

EHC-MM: Embodied Holistic Control for Mobile Manipulation

Jiawen Wang, Bin Fang

OptimizationRobotic Intelligence

🎯 What it does: Proposes the Embodied Holistic Control for Mobile Manipulation (EHC-MM) scheme, which utilizes the sig(ω) function to convert the principles of distant mobile and close-range grasping (DMCG) into a quadratic programming (QP) problem, and combines it with Monitor-Position-Based Servoing (MPBS) to achieve target tracking, thereby enabling collaborative control of the robot's base, manipulator, and camera.

Elderly Bodily Assistance Robot (E-BAR): A Robot System for Body-Weight Support, Ambulation Assistance, and Fall Catching, Without the Use of a Harness

Roberto Bolli, H. Asada

OptimizationRobotic Intelligence

🎯 What it does: Developed a senior care robot named E-BAR, capable of human lifting, posture transformation/gait assistance, and fall capture and stabilization without any wearable devices or safety belts; the system integrates an 18-bar linkage lifting mechanism, omnidirectional nonholonomic drive chassis, and four airbags, achieving fall capture within 250 ms and navigating narrow home passages in typical domestic environments; user experience testing was conducted through interviews with caregivers and elderly users, with design parameter optimization based on functional requirements using computational models and trade-off analysis.

Elite-EvGS: Learning Event-based 3D Gaussian Splatting by Distilling Event-to-Video Priors

Zixin Zhang, Lin Wang

GenerationOptimizationKnowledge DistillationGaussian SplattingVideo

🎯 What it does: Proposed a 3D Gaussian Splatting (3DGS) framework called Elite-EvGS based on event cameras, which achieves 3D scene reconstruction from event streams by distilling prior knowledge from existing event-to-video models.

EMATO: Energy-Model-Aware Trajectory Optimization for Autonomous Driving

Zhaofeng Tian, Weisong Shi

Autonomous DrivingOptimization

🎯 What it does: Proposed an online nonlinear programming method for trajectory optimization based on energy-aware models, integrating Frenet polynomial trajectories, traffic trajectory data, and road slope predictions, with case studies and quantitative analysis conducted on passenger car and truck models.

Embedded IPC: Fast and Intersection-Free Simulation in Reduced Subspace for Robot Manipulation

Wenxin Du, Chenfanfu Jiang

Robotic Intelligence

🎯 What it does: Propose an incremental potential contact (IPC) method based on subspace representation, reducing degrees of freedom through model dimensionality reduction, while retaining collision constraints on embedded high-resolution surfaces, and using a barrier function to ensure trajectories and configurations do not intersect under any material stiffness, time step, or contact strength conditions;

Embedded Robust Model Predictive Path Integral Control Using Sensitivity Tubes and GPU Acceleration

Frederik Falk Nyboe, Antonio Franchi

OptimizationComputational Efficiency

🎯 What it does: Proposed a method to enhance the robustness of the Model Predictive Path Integral (MPPI) controller by directly considering parameter uncertainty.

Embodied Adaptive Sensing for Odor Concentration Maximization in Bio-Inspired Robotics

Jettanan Homchanthanakul, P. Manoonpong

Robotic Intelligence

🎯 What it does: Implement an active olfactory sensor on a legged robot, and adopt a bio-inspired adaptive height control system that dynamically adjusts the robot's height based on real-time gas concentration feedback to maximize odor detection.

Embodiment-agnostic Action Planning via Object-Part Scene Flow

Weiliang Tang, Chi-Wing Fu

Robotic IntelligenceReinforcement Learning from Human FeedbackOptical FlowVideo

🎯 What it does: Proposes a method for planning action trajectories for different implementations by generating 3D object-part scene flow and extracting its transformations

Enabling In-Flight Metamorphosis in Multirotors with a Center-Driven Scissor Extendable Airframe for Adaptive Navigation

Tao Yang, Yantao Shen

Robotic Intelligence

🎯 What it does: Designed and implemented a center-driven scissor-like expandable body (CDSEA) enabling quadrotors to achieve morphing during flight;

Enabling Multi-Robot Collaboration from Single-Human Guidance

Zhengran Ji, Boyuan Chen

Robotic IntelligenceReinforcement Learning from Human Feedback

🎯 What it does: Propose a method that learns multi-agent collaborative behaviors with only single-person guidance, where the human operator dynamically switches control between agents within a short time and incorporates a theory-of-mind model similar to humans.

End-to-End Underwater Multi-View Stereo for Dense Scene Reconstruction

Guidong Yang, Ben M. Chen

Convolutional Neural NetworkImageBenchmark

🎯 What it does: Proposes a physics-guided underwater multi-view image synthesis method, constructs the first large-scale underwater multi-view stereo reconstruction dataset, and designs an improved UwMVS network with enhanced geometric feature encoding.

Endpoint-Explicit Differential Dynamic Programming via Exact Resolution

Maria Parilli, Carlos Mastalli

Optimization

🎯 What it does: Proposed an exact analytical method for endpoint-explicit differential dynamic programming (DDP) to handle endpoint constraints.

Energy Efficient Planning for Repetitive Heterogeneous Tasks in Precision Agriculture

Shuangyun Xie, Dezhen Song

OptimizationAgriculture Related

🎯 What it does: This study investigates repetitive heterogeneous task planning (RHTP) in precision agriculture, proposing task space partitioning and a region-based set cover method, and modeling the problem as a mixed-integer nonlinear programming problem, with solutions implemented using a branch-and-bound solver.

Enhanced View Planning for Robotic Harvesting: Tackling Occlusions with Imitation Learning

Lun Li, H. Kasaei

Robotic IntelligenceTransformerMultimodalityAgriculture Related

🎯 What it does: Propose a view planning method based on imitation learning, which actively adjusts the camera perspective to capture unobstructed images of target crops, solving occlusion problems in agricultural harvesting.

Enhancing 3D Robotic Vision Robustness by Minimizing Adversarial Mutual Information through Curriculum Training

Nastaran Darabi, A. Trivedi

Autonomous DrivingAdversarial AttackRobotic IntelligenceGraph Neural NetworkTransformerPoint Cloud

🎯 What it does: Propose a training objective that simultaneously minimizes prediction loss and mutual information under adversarial perturbations, and gradually introduces adversarial objectives through a curriculum learning guide to enhance the robustness of 3D vision.

Enhancing 3D Scene Graphs with Real-Time Room Classification

Simon Janzon, P. J. Marrón

ClassificationComputational EfficiencyRobotic IntelligenceImageGraph

🎯 What it does: Propose a real-time 3D scene graph generation pipeline that integrates random forest-based room classification to enable real-time updates of the robot's understanding of complex large-scale environments.

Enhancing Adaptivity of Two-Fingered Object Reorientation Using Tactile-Based Online Optimization of Deconstructed Actions

Qiyin Huang, Yao Jiang

OptimizationRobotic Intelligence

🎯 What it does: Propose a method utilizing tactile-based online optimization for decomposing actions to enhance the adaptability of a two-finger gripper in constrained environments during object reorientation.

Enhancing Agricultural Environment Perception via Active Vision and Zero-Shot Learning

Michele Carlo La Greca, Matteo Matteucci

SegmentationConvolutional Neural NetworkTransformerSimultaneous Localization and MappingAgriculture Related

🎯 What it does: Built an active vision (AV) pipeline integrating zero-shot learning-based semantic segmentation models, implementing next-best-view planning and 3D occupancy map updates on ROS 2 to enhance perception and interaction capabilities of fruit-picking robots.

Enhancing AR-to-Robot Registration Accuracy: A Comparative Study of Marker Detection Algorithms and Registration Parameters

Tonia Mielke, Christian Hansen

Pose EstimationRobotic IntelligenceSimultaneous Localization and MappingImagePoint Cloud

🎯 What it does: Systematically evaluated and compared four AR marker detection algorithms (ARToolkit, Vuforia, ArUco, retroreflective tracking) in robot registration, investigated the impact of parameters such as viewing distance, angle, marker size, registration point distance, distribution, and quantity on registration error, and proposed a fine-grained method based on point cloud registration to further improve registration accuracy.

Enhancing Autonomous Navigation by Imaging Hidden Objects Using Single-Photon LiDAR

Aaron Young, Ramesh Raskar

Autonomous DrivingConvolutional Neural NetworkPoint Cloud

🎯 What it does: Utilizing single-photon LiDAR for non-line-of-sight (NLOS) perception to enable mobile robots to achieve 'around-the-corner' vision, enhancing autonomous navigation capabilities.

Enhancing Connection Strength in Freeform Modular Reconfigurable Robots Through Holey Sphere and Gripper Mechanisms

Pei-Chung Wang, T. Lam

Robotic Intelligence

🎯 What it does: Proposed rigid free-form connectors and rigid magnetic track designs, integrating multi-channel rope-driven grippers, densely distributed circular holes on the rear of the metal spherical shell, and rigid chain structures adapted to the spherical surface on SnailBot to enhance module-to-module connection and motion performance.

Enhancing Feature Tracking Reliability for Visual Navigation Using Real-Time Safety Filter

Dabin Kim, H. Kim

Autonomous DrivingOptimizationSimultaneous Localization and MappingImage

🎯 What it does: Propose a real-time safety filter to ensure sufficient feature visibility in visual navigation while minimizing deviation from the reference speed.

Enhancing Multi-Agent Systems via Reinforcement Learning with LLM-Based Planner and Graph-Based Policy

Ziqi Jia, Jianzong Wang

Meta LearningGraph Neural NetworkLarge Language ModelReinforcement LearningBenchmark

🎯 What it does: Proposed the LGC-MARL framework, which decomposes complex tasks into executable subtasks through an LLM planner and graph collaboration meta-policy, achieving efficient multi-agent collaboration.

Enhancing Navigation Efficiency of Quadruped Robots via Leveraging Personal Transportation Platforms

Minsung Yoon, Sung-Eui Yoon

Robotic IntelligenceReinforcement Learning

🎯 What it does: Developed a reinforcement learning-based active transportation platform riding method (RL-ATR), enabling quadruped robots to navigate using personal transportation platforms (e.g., Segway).

Enhancing Repeatability and Reliability of Accelerated Risk Assessment in Robot Testing

L. Capito, Bowen Weng

Robotic Intelligence

🎯 What it does: Propose a new algorithm to enhance the accelerated risk assessment framework, achieving repeatability and reliability while maintaining formality and efficiency; demonstrate through evaluations of instability risks caused by forward impacts using different control algorithms on a controlled inverted pendulum and a 7-degree-of-freedom planar bipedal robot Rabbit.

Enhancing Robotic Perception with Low-Cost Fast Active Vision Achieving Sub-Millimeter Accurate Marker-Based Pose Estimation

Dennis Knobbe, Sami Haddadin

Pose EstimationRobotic IntelligenceImageBenchmark

🎯 What it does: Designed and implemented a low-cost, easy-to-deploy active vision system that achieves sub-millimeter and sub-degree accuracy in pose estimation using ArUco markers, and proposed a new measurement and evaluation process.

Enhancing Robotic System Robustness via Lyapunov Exponent-Based Optimization

Gabriele Fadini, Stelian Coros

OptimizationRobotic Intelligence

🎯 What it does: Proposed a differentiable method using Lyapunov exponents to evaluate and optimize the stability of robotic systems.

Enhancing the Utilization of Color Information in Point Cloud Semantic Segmentation

Xinyu Guo, Min Cao

SegmentationContrastive LearningPoint Cloud

🎯 What it does: Propose the Color Point Cloud Enhancement (CPCE) method, which improves the performance of point cloud semantic segmentation through a color information enhancement module and a contrastive learning module.

Ensemble Control of a 2-DOF Parallel Link Arm in a Capsule Robot Using Oscillating External Magnetic Fields

Zihan Zhao, Shuhei Miyashita

Robotic IntelligencePhysics Related

🎯 What it does: Proposes a method for group control of a 2-degree-of-freedom parallel link manipulator integrated into an oral capsule robot using an external oscillating magnetic field.

Environment as Policy: Learning to Race in Unseen Tracks

Hongze Wang, Davide Scaramuzza

Robotic IntelligenceReinforcement Learning

🎯 What it does: Proposed an adaptive environment shaping framework that enables drones to successfully fly on unseen tracks without retraining

Environmental Map Learning with Multiple-Robots

Azin Shamshirgaran, Stefano Carpin

OptimizationRobotic Intelligence

🎯 What it does: Proposed an online distributed multi-robot sampling algorithm that combines Monte Carlo Tree Search (MCTS) with Gaussian regression for scalar field reconstruction in uncertain environments.

EnvoDat: A Large-Scale Multisensory Dataset for Robotic Spatial Awareness and Semantic Reasoning in Heterogeneous Environments

Linus Nwankwo, Elmar Rueckert

Robotic IntelligenceSimultaneous Localization and MappingMultimodalityBenchmark

🎯 What it does: Created a large multimodal dataset called EnvoDat for robotic spatial perception and semantic reasoning in heterogeneous environments.

Ephemerality Meets Lidar-Based Lifelong Mapping

Hyeonjae Gil, Ayoung Kim

Autonomous DrivingSimultaneous Localization and MappingWorld ModelPoint Cloud

🎯 What it does: We propose the ELite framework, which achieves seamless alignment of multi-session LiDAR data, dynamic object removal, and end-to-end map updating.

EPRecon: An Efficient Framework for Real-Time Panoptic 3D Reconstruction from Monocular Video

Zhen Zhou, M. Tan

SegmentationDepth EstimationComputational EfficiencyVideo

🎯 What it does: Proposed an efficient and real-time panoramic 3D reconstruction framework called EPRecon.

Equivariant Filter Design for Range-Only SLAM

Yixiao Ge, Robert E. Mahony

OptimizationRobotic IntelligenceSimultaneous Localization and MappingMultimodality

🎯 What it does: Studies on mobile robots using only range sensors and IMU for RO-SLAM, and proposes an equivariant filter (EqF) based on symmetric Lie groups