IROS 2024 Papers — Page 9
IEEE/RSJ International Conference on Intelligent Robots and Systems · 1581 papers
LCP-Fusion: A Neural Implicit SLAM with Enhanced Local Constraints and Computable Prior
Jiahui Wang, Yufeng Yue
Robotic IntelligenceSimultaneous Localization and MappingImage
🎯 What it does: Proposes LCP-Fusion, a neural implicit SLAM system that integrates sparse voxel octrees, feature grids, and SDF priors to achieve scalable and robust dense SLAM.
LDIP: Real-time on-road object detection with depth estimation from a single image
Chengpeng Xu, Ruolin Wang
Object DetectionDepth EstimationAutonomous DrivingImage
🎯 What it does: Proposed and implemented a model called LDIP for real-time road object detection from a single image with absolute depth information.
LDP: A Local Diffusion Planner for Efficient Robot Navigation and Collision Avoidance
Wenhao Yu, Yanyong Zhang
Robotic IntelligenceDiffusion model
🎯 What it does: Proposes a Local Diffusion Planner (LDP) based on conditional diffusion models, achieving efficient robot navigation and collision avoidance through multi-agent multi-preference data generation and lightweight global observation integration.
Leader-Follower Cooperative Manipulation Under Spatio-Temporal Constraints
Mayank Sewlia, Dimos V. Dimarogonas
Robotic Intelligence
🎯 What it does: Developed a collaborative manipulation control algorithm under a leader-follower framework for mobile manipulators, avoiding the use of object dynamics information, where only the leader robot knows the task, and the follower is responsible for lifting and maintaining posture;
Learned Regions of Attraction for Safe Motion Primitive Transitions
Wyatt Ubellacker, Aaron D. Ames
Robotic Intelligence
🎯 What it does: Proposes a neural network-based method that approximates the basin of attraction of a dynamical system using labeled data obtained from offline sampling and simulation, and adjusts the model online according to actual system behavior to achieve safe switching of motion primitives in quadruped robots
Learned Sensor Fusion For Robust Human Activity Recognition in Challenging Environments
Max Conway, Christopher M. Reardon
ClassificationRecognitionMultimodality
🎯 What it does: Propose a framework for robust human action recognition through learned sensor fusion, automatically learning weighted combinations of different modalities to enable the classifier to fully utilize the strengths of each sensor.
Learned Slip-Detection-Severity Framework using Tactile Deformation Field Feedback for Robotic Manipulation
Neel Jawale, Xu Chen
Robotic Intelligence
🎯 What it does: The study proposes a framework based on tactile deformation field feedback for slip detection and severity assessment, which can simultaneously identify slip events and quantify their severity.
Learning a Pre-Grasp Manipulation Policy to Effectively Retrieve a Target in Dense Clutter
Marios Kiatos, Zoe Doulgeri
Domain AdaptationRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposes a strategy for pre-grasping target objects in dense cluttered environments through push actions to enhance grasping success rates.
Learning a Shape-Conditioned Agent for Purely Tactile In-Hand Manipulation of Various Objects
Johannes Pitz, Berthold Bauml
Representation LearningRobotic IntelligenceReinforcement Learning
🎯 What it does: Train a shape-conditioned agent to reorient various objects using a multi-fingered hand through tactile feedback alone.
Learning Agile Locomotion on Risky Terrains
Chong Zhang, Marco Hutter
Robotic IntelligenceReinforcement Learning
🎯 What it does: Train a general policy to achieve agile walking on sparse stepping stones and transfer it to more challenging hazardous terrains via dynamic walking through fine-tuning expert policies; define the task as a navigation task for rapid speed adaptation and propose an exploration strategy to overcome sparse rewards, achieving a forward speed of ≥2.5 m/s validated on simulation and real ANYmal-D robots.
Learning autonomous driving from aerial imagery
Varun Murali, Daniela Rus
Pose EstimationDepth EstimationAutonomous DrivingNeural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: Utilizing neural radiance fields (NeRF) to synthesize ground vehicle perspectives from aerial images, training end-to-end driving policies, and deploying imitation learning models in a self-built mini city environment to verify their localization and navigation capabilities in the real world.
Learning Bimanual Manipulation Policies for Bathing Bed-bound People
Yijun Gu, Y. Demiris
Robotic IntelligenceReinforcement Learning
🎯 What it does: Developed a dual-arm robot bathroom care system that learns to lift target limbs within safe force limits and use cleaning tools for bathing.
Learning Concept-Based Causal Transition and Symbolic Reasoning for Visual Planning
Yilue Qian, Lifeng Fan
Explainability and InterpretabilityRepresentation LearningVision-Language-Action ModelImage
🎯 What it does: Proposed an interpretable and transferable visual planning framework, including a novel replacement-based concept learner (SCL) that abstracts visual input into separable concept representations, symbolic abstraction and reasoning modules, and a visual causal transfer model (ViCT) that maps visual causal transfers to semantically similar real-world actions.
Learning Coordinated Maneuver in Adversarial Environments
Zechen Hu, Xuan Wang
OptimizationRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposes the problem of minimizing total cost for multi-robot coordinated movement in environments with randomly positioned adversaries, and achieves coordinated behavior using reinforcement learning methods
Learning Deep Dynamical Systems using Stable Neural ODEs
Andreas Sochopoulos, S. Vijayakumar
Representation LearningImage
🎯 What it does: Propose a class of provably stable latent dynamical systems that support multiple attractors and eliminate the need for state derivative information by leveraging the training methods of Neural ODEs.
Learning Dynamic Tasks on a Large-scale Soft Robot in a Handful of Trials
Sicelukwanda Zwane, M. Deisenroth
Robotic Intelligence
🎯 What it does: Proposed a data-efficient method based on Bayesian optimization for learning control policies for dynamic tasks on large soft robots.
Learning dynamics models for velocity estimation in autonomous racing
Jan Wegrzynowski, Krzysztof Walas
Autonomous DrivingOptimizationTime Series
🎯 What it does: Proposes a speed estimation method based on UKF and a learning dynamics model, combined with online friction coefficient estimation to achieve zero-shot adaptation.
Learning from Demonstration Framework for Multi-Robot Systems Using Interaction Keypoints and Soft Actor-Critic Methods
Vishnunandan L. N. Venkatesh, Byung-Cheol Min
Robotic IntelligenceReinforcement Learning
🎯 What it does: Proposes a learning demonstration framework for multi-robot systems that captures interaction keypoints between robots and robot-object interactions through visual demonstrations, and executes tasks via reinforcement learning.
Learning from Spatio-temporal Correlation for Semi-Supervised LiDAR Semantic Segmentation
Seungho Lee, Hyunjung Shim
SegmentationAutonomous DrivingPoint Cloud
🎯 What it does: Proposes a semi-supervised LiDAR semantic segmentation method leveraging spatiotemporal correlation, primarily generating high-quality pseudo-labels through spatial consistency of adjacent scans, and employing a dual-branch structure to alleviate label imbalance issues.
Learning Generalizable Manipulation Policy with Adapter-Based Parameter Fine-Tuning
Kai Lu, Andrew Markham
Domain AdaptationRobotic IntelligenceReinforcement Learning
🎯 What it does: Using adapters to implement reinforcement learning-based virtual hand (disembodied hand) learning of abstract manipulation skills, and transferring these skills to multiple robots with different kinematic constraints without retraining the entire model.
Learning Generalizable Tool-use Skills through Trajectory Generation
Carl Qi, David Held
Robotic IntelligencePoint Cloud
🎯 What it does: Learning how to manipulate deformable objects using unseen tools by generating sequences of tool usage trajectories to guide robot actions.
Learning High-level Semantic-Relational Concepts for SLAM
J. A. Millan-Romera, J. L. Sánchez-López
Graph Neural NetworkSimultaneous Localization and MappingGraph
🎯 What it does: Automatically learn high-level semantic relationship concepts (such as rooms and walls) from low-level factor graphs using graph neural networks, and integrate them into the SLAM S-Graphs+ algorithm.
Learning Human-to-Humanoid Real-Time Whole-Body Teleoperation
Tairan He, Guanya Shi
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningImage
🎯 What it does: Proposed and implemented the Human to Humanoid (H2O) framework, which achieves real-time whole-body teleoperation of full-scale humanoid robots using only RGB cameras through reinforcement learning.
Learning incipient slip with GelSight sensors: Attention Classification with Video Vision Transformers
Amit Parag, E. Misimi
Robotic IntelligenceTransformerVideo
🎯 What it does: Trained using video vision transformers combined with GelSight tactile sensors on dynamic data from a Panda robot gripper grasping, lifting, and shaking 30 everyday objects, achieving zero-shot prediction on 10 unseen objects.
Learning Long-Horizon Predictions for Quadrotor Dynamics
P. Rao, Giuseppe Loianno
Robotic IntelligenceTime SeriesSequential
🎯 What it does: Study and learn long-term prediction of quadrotor dynamics, analyzing the impact of different architectures, historical data, and multi-step loss on prediction accuracy.
Learning Multi-Reference Frame Skills from Demonstration with Task-Parameterized Gaussian Processes
Mariano Ramirez Montero, C. D. Santina
Robotic Intelligence
🎯 What it does: Under a multi-reference framework, local skills are learned using Gaussian processes and overlaid through a global framework, with self-supervised learning of correlations between each reference framework at every moment from demonstrations.
Learning Safe Locomotion for Quadrupedal Robots by Derived-Action Optimization
Deye Zhu, Yong Liu
OptimizationRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposed a Derived-Action Optimization framework that guides quadrupedal robot walking decisions by introducing high-level actions (e.g., foot placement) into the reward function, achieving rapid convergence and safer locomotion;
Learning Sampling Distribution and Safety Filter for Autonomous Driving with VQ-VAE and Differentiable Optimization
S. Idoko, A. K. Singh
Autonomous DrivingOptimizationAuto Encoder
🎯 What it does: This paper proposes a self-driving trajectory sampling and correction framework that combines VQ-VAE with a differentiable safety filter, aiming to learn a multimodal sampling distribution and achieve collision avoidance on trajectories.
Learning Social Cost Functions for Human-Aware Path Planning
Andrea Eirale, Marcello Chiaberge
OptimizationRobotic Intelligence
🎯 What it does: Identify common social scenarios and modify the cost function of traditional path planners to enable robots to follow various social norms while maintaining navigation robustness.
Learning Symbolic and Subsymbolic Temporal Task Constraints from Bimanual Human Demonstrations
Christian R. G. Dreher, Tamim Asfour
Robotic Intelligence
🎯 What it does: Learn and synthesize symbolic and sub-symbolic temporal constraint models, achieving dual-arm robot task planning and execution based on multiple human demonstrations with both hands.
Learning Temporally Composable Task Segmentations with Language
Divyanshu Raj, N. Gopalan
RetrievalDomain AdaptationRobotic IntelligenceVision-Language-Action ModelVideoText
🎯 What it does: Identifying subtasks in robot trajectories guided by language instructions
Learning the Inverse Kinematics of Magnetic Continuum Robot for Teleoperated Navigation
Pingyu Xiang, Haojian Lu
Robotic Intelligence
🎯 What it does: Propose end-to-end learning of inverse kinematics for magnetic continuum robots, achieving precise control by binding robot configurations to external magnet positions.
Learning to Estimate the Pose of a Peer Robot in a Camera Image by Predicting the States of its LEDs
Nicholas Carlotti, Alessandro Giusti
Pose EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Train a fully convolutional network using pre-training with LED states to estimate relative 6D pose
Learning to Imitate Spatial Organization in Multi-robot Systems
Ayomide O. Agunloye, Mohammad Divband Soorati
Robotic IntelligenceGenerative Adversarial Network
🎯 What it does: Reconstruct the spatial organization behavior of multi-robot systems using prior demonstration data without relying on group controller information, and achieve behavior reconstruction through multi-agent generative adversarial imitation learning (MA-GAIL).
Learning to Recover from Plan Execution Errors during Robot Manipulation: A Neuro-symbolic Approach
Namasivayam Kalithasan, Rohan Paul
Robotic IntelligenceGraph Neural NetworkGraph
🎯 What it does: Propose a neuro-symbolic method combining learning and symbolic search for automatically discovering and recovering from robot execution errors, based on a dense scene graph state representation without requiring labeled error data.
Learning to Walk and Fly with Adversarial Motion Priors
Giuseppe Lerario, Davide Scaramuzza
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a robot multi-modal locomotion method that can automatically and smoothly transition between walking and flying modes.
Learning Variable Compliance Control From a Few Demonstrations for Bimanual Robot with Haptic Feedback Teleoperation System
Tatsuya Kamijo, Masashi Hamaya
Robotic IntelligenceTransformer
🎯 What it does: Demonstrating variable compliance control for rigid dual-arm robots using a VR hand controller-based haptic remote control and the Comp-ACT method, enabling improved flexibility and safety through learning from a small number of demonstrations, suitable for complex contact-rich operational tasks.
Learning Visual Quadrupedal Loco-Manipulation from Demonstrations
Zhengmao He, Huazhe Xu
Robotic IntelligenceReinforcement Learning
🎯 What it does: Achieve real-world manipulation tasks using the legs of a quadruped robot, combining a low-level reinforcement learning (RL) controller with a high-level behavior cloning (BC) planner for cooperative control.
Learning When to Stop: Efficient Active Tactile Perception with Deep Reinforcement Learning
Christoph Niemann, R. Haschke
ClassificationRobotic IntelligenceReinforcement Learning
🎯 What it does: Extended the Haptic Attention Model (HAM) to enable it to actively decide when to stop collecting tactile samples and perform classification.
Learning-based Adaptive Admittance Controller for Efficient and Safe pHRI in Contact-rich Manufacturing Tasks
Pouya P. Niaz, C. Basdogan
Safty and PrivacyRobotic IntelligenceTime Series
🎯 What it does: Proposed a learning-based adaptive auxiliary controller combined with a two-layer human intention recognition mechanism for contact-based physical human-robot interaction tasks in small-batch manufacturing, aiming to reduce manpower, shorten task completion time, and improve accuracy and stability.
Learning-based Adaptive Control of Quadruped Robots for Active Stabilization on Moving Platforms
Minsung Yoon, Sung-Eui Yoon
Robotic IntelligenceReinforcement Learning
🎯 What it does: An adaptive balance control system called LAS-MP is proposed to provide quadruped robots on six-degree-of-freedom mobile platforms with adaptive balance control, including self-balancing strategies and a system state estimator, along with the introduction of platform trajectory generation and scheduling methods.
Learning-based Hierarchical Control: Emulating the Central Nervous System for Bio-Inspired Legged Robot Locomotion
Ge Sun, Guillaume Sartoretti
Robotic Intelligence
🎯 What it does: Designed and implemented a two-layer neural network hierarchical control framework, with the spinal policy layer responsible for rhythm generation and the descending modulation policy layer for下行调制, to mimic hierarchical motion control in animals and achieve precise locomotion on uneven terrains (e.g., steps, high obstacles, gaps); simultaneously investigated and validated the impact of perception-motor delay on the control system and its robustness.
Learning-informed Long-Horizon Navigation under Uncertainty for Vehicles with Dynamics
Abhish Khanal, Gregory J. Stein
Autonomous DrivingWorld Model
🎯 What it does: Proposes a learning-enhanced long-range navigation method that combines sample-based motion planning with learning-based high-level planning, capable of considering vehicle dynamics and planning feasible paths in uncertain large-scale environments.
Learning-on-the-Drive: Self-supervised Adaptive Long-range Perception for High-speed Offroad Driving
Eric Chen, S. Scherer
Domain AdaptationAutonomous DrivingContrastive LearningImagePoint Cloud
🎯 What it does: Proposes the ALTER framework, which combines self-supervised visual models with close-range LiDAR data to achieve adaptive long-range perception for high-speed off-road driving.
LEEPS: Learning End-to-End Legged Perceptive Parkour Skills on Challenging Terrains
Tangyu Qian, Zhen Kan
Robotic IntelligenceReinforcement LearningImage
🎯 What it does: Developed an end-to-end leg-perception parkour skill learning framework (LEEPS) to train quadruped robots to master parkour skills in complex environments.
LeGo-Drive: Language-enhanced Goal-oriented Closed-Loop End-to-End Autonomous Driving
Pranjal Paul, K. Krishna
Autonomous DrivingOptimizationVision Language Model
🎯 What it does: Introduce the LeGo-Drive system, which uses VLM to predict only the target position and feeds it into a differentiable trajectory optimizer. Safety and kinematic constraints are achieved through end-to-end training, ultimately achieving an 81% success rate.
Leveraging Computation of Expectation Models for Commonsense Affordance Estimation on 3D Scene Graphs
M. A. Saucedo, G. Nikolakopoulos
Representation LearningGraph Neural NetworkGraph
🎯 What it does: Studies the concept of affordance for common objects, leveraging contextual relationships from sparse information in 3D scene graphs to reason about the inherent utility of objects during task execution; proposes the Correlation Information (CECI) model, which employs a graph convolutional network (GCN) to learn probability distributions, thereby extracting commonsense affordance for semantic class members; experimentally validates the method in real indoor environments, demonstrating performance comparable to human commonsense.
Leveraging GNSS and Onboard Visual Data from Consumer Vehicles for Robust Road Network Estimation
Bal'azs Opra, C. Stachniss
SegmentationAutonomous DrivingConvolutional Neural NetworkImageMultimodalityTime Series
🎯 What it does: Automatically construct road centerline maps using GNSS trajectories from consumer-grade vehicles and basic visual data
Leveraging Neural Radiance Field in Descriptor Synthesis for Keypoints Scene Coordinate Regression
Huy-Hoang Bui, Joo-Ho Lee
Pose EstimationGraph Neural NetworkNeural Radiance Field
🎯 What it does: This paper proposes a pipeline that utilizes NeRF to generate keypoint descriptors in order to improve the localization accuracy of keypoint scene coordinate regression (KSCR).
Leveraging Simulation-Based Model Preconditions for Fast Action Parameter Optimization with Multiple Models
M. Seker, Oliver Kroemer
OptimizationRobotic IntelligenceWorld ModelImage
🎯 What it does: This paper proposes a framework that utilizes multiple prediction models (analytical models, learning models, simulation models) and rapidly and efficiently optimizes robot operation parameters through a Model Deviation Estimator (MDE).
Leveraging Symmetry in RL-based Legged Locomotion Control
Zhi Su, K. Sreenath
Robotic IntelligenceReinforcement Learning
🎯 What it does: Improving model-free reinforcement learning for legged robot control by leveraging symmetry constraints, employing two methods: strict equivariant network architecture and data augmentation, applied to motion-manipulation and bipedal gait tasks.
LF-3PM: a LiDAR-based Framework for Perception-aware Planning with Perturbation-induced Metric
Kaixin Chai, Fei Gao
Autonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposed the LiDAR-based perception-aware planning framework LF-3PM, which improves the localization performance of autonomous robots in sparse environments by evaluating the impact of LiDAR observations on localization accuracy and stability, and accelerating motion planning using a static observation loss map
LF2SLAM: Learning-based Features For visual SLAM
Marco Legittimo, G. Costante
Pose EstimationConvolutional Neural NetworkSimultaneous Localization and Mapping
🎯 What it does: Propose a hybrid visual odometry method LF2SLAM that integrates deep neural network feature extractors with traditional visual odometry pipelines.
LGMCTS: Language-Guided Monte-Carlo Tree Search for Executable Semantic Object Rearrangement
Haonan Chang, Abdeslam Boularias
OptimizationRobotic IntelligenceLarge Language ModelTextBenchmark
🎯 What it does: Propose the LGMCTS framework, combining language guidance with geometric information sampling to generate executable semantic object rearrangement action plans.
LiDAR-based 4D Occupancy Completion and Forecasting
Xinhao Liu, Chen Feng
Autonomous DrivingPoint CloudBenchmark
🎯 What it does: Proposed the LiDAR perception task OCF, achieving a unified framework for sparse-to-dense, partial-to-complete, and 3D-to-4D prediction, and evaluated baselines based on the OCFBench dataset
LiDAR-based HD Map Localization using Semantic Generalized ICP with Road Marking Detection
Yansong Gong, Dan Zhang
Autonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Developed a LiDAR-based online localization system that integrates real-time road marking detection with HD map registration;
LiDAR-camera Online Calibration by Representing Local Feature and Global Spatial Context
Se-Min Moon, Sung-Eui Yoon
Autonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: Proposes an online calibration framework that utilizes a Transformer network to learn key interactions between camera and LiDAR sensors, achieving effective calibration through corresponding point information from both sensors.
LiDAR-Visual-Inertial Tightly-coupled Odometry with Adaptive Learnable Fusion Weights
Vsevolod Hulchuk, J. Faigl
Pose EstimationAutonomous DrivingSimultaneous Localization and MappingImageMultimodalityPoint Cloud
🎯 What it does: Propose a feature-based fusion learning method that combines planar LiDAR features with near and far camera features, and adaptively adjusts feature weights online based on environmental ambiguity to support autonomous mobile robot localization in new environments.
Lightweight Fisheye Object Detection Network with Transformer-based Feature Enhancement for Autonomous Driving
Hu Cao, Alois Knoll
Object DetectionAutonomous DrivingConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed a lightweight fisheye object detection network and introduced a Transformer-based feature enhancement module to improve detection performance.
Lightweight Language-driven Grasp Detection using Conditional Consistency Model
Nghia Nguyen, Anh Nguyen
Object DetectionRobotic IntelligencePrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposed a language-driven grasp detection method based on a lightweight diffusion model, and utilized a consistency model during the inference phase to reduce denoising steps for fast inference.
LiOn-XA: Unsupervised Domain Adaptation via LiDAR-Only Cross-Modal Adversarial Training
Thomas Kreutz, Alejandro Sánchez Guinea
SegmentationDomain AdaptationGenerative Adversarial NetworkImagePoint Cloud
🎯 What it does: Proposes LiOn-XA, an unsupervised domain adaptation method combining LiDAR-Only Cross-Modal learning with adversarial training, for 3D LiDAR point cloud semantic segmentation, aiming to bridge domain gaps caused by environmental and sensor setup changes.
LLaKey: Follow My Basic Action Instructions to Your Next Key State
Zheyi Zhao, Xilong Sun
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelText
🎯 What it does: LLaKey achieves fine-grained control over 3D object manipulation by decomposing skill instructions into more detailed action instructions based on key states, and utilizes a pre-trained large-scale model to fine-tune an action instruction scheduler, which is then executed by downstream action models.
LLM3: Large Language Model-based Task and Motion Planning with Motion Failure Reasoning
Shu Wang, Hangxin Liu
Robotic IntelligenceTransformerLarge Language ModelPrompt Engineering
🎯 What it does: Developed a task and motion planning framework called LLM3 based on large language models (LLMs), which generates symbolic action sequences and selects continuous action parameters using pre-trained LLMs, and iteratively improves them through motion planning feedback.
Local Linearity is All You Need (in Data-Driven Teleoperation)
Michael Przystupa, Samuele Tosatto
Robotic IntelligenceSupervised Fine-TuningBenchmark
🎯 What it does: Studied action mapping models in data-driven teleoperation, compared local linear and nonlinear models, and proposed two nonlinear parity encoding methods not constrained by local linear assumptions, verifying their effectiveness in simulations and real-world cases
Local Path Planning among Pushable Objects based on Reinforcement Learning
Linghong Yao, D. Kanoulas
Robotic IntelligenceReinforcement Learning
🎯 What it does: Propose a reinforcement learning-based method for robot local path planning among movable obstacles;
Loco-Manipulation with Nonimpulsive Contact-Implicit Planning in a Slithering Robot
Adarsh Salagame, Alireza Ramezani
OptimizationRobotic Intelligence
🎯 What it does: This paper proposes a localization-manipulation optimization method for the snake robot COBRA based on non-impulsive implicit contact path planning, aiming to achieve the integration of terrain locomotion and object manipulation, i.e., loco-manipulation.
LocoMan: Advancing Versatile Quadrupedal Dexterity with Lightweight Loco-Manipulators
Changyi Lin, Ding Zhao
Robotic Intelligence
🎯 What it does: Developed the LocoMan quadruped robot equipped with a lightweight modular 3-DoF loco-manipulator on the front legs, and proposed a unified whole-body control framework to achieve precise 6D end-effector positioning.
Long-Term Map-Maintenance in Changing Environments using Ray-Bundle-Impact-Factor Estimation
Matthias Breitfuss, Christoph Gruber
Autonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposes a technology for achieving accurate and robust map maintenance through real-time integration of 3D LiDAR scans.
Look Before You Leap: Socially Acceptable High-Speed Ground Robot Navigation in Crowded Hallways
Lakshay Sharma, Jonathan P. How
Robotic Intelligence
🎯 What it does: Propose a planner capable of performing peek-and-pass operations to address robot freezing issues in crowded corridors
Look Gauss, No Pose: Novel View Synthesis using Gaussian Splatting without Accurate Pose Initialization
Christian Schmidt, Bastian Leibe
GenerationData SynthesisPose EstimationOptimizationGaussian SplattingImage
🎯 What it does: By optimizing the extrinsic camera parameters in the 3D Gaussian Splatting framework, achieving fast 3D reconstruction and novel view synthesis using photometric residuals without requiring accurate camera pose initialization.
Loss Distillation via Gradient Matching for Point Cloud Completion with Weighted Chamfer Distance
Fangzhou Lin, Ziming Zhang
RestorationOptimizationKnowledge DistillationPoint CloudBenchmark
🎯 What it does: Proposes a method for loss distillation through gradient matching, searching for a weighted Chamfer distance (CD) loss that does not require parameter tuning, and introduces a two-layer optimization training framework based on this loss;
Low-Cost Air Hockey Robot Using a Five-Bar Linkage Mechanism Driven by Position-Control Servomotors
Mirai Shinjo, Kazutoshi Tanaka
Robotic Intelligence
🎯 What it does: Designed and implemented a low-cost air hockey robot using a five-bar linkage mechanism and position control servos to accurately return the puck.
Low-Cost Urban Localization with Magnetometer and LoRa Technology
Derek Benham, Joshua G. Mangelson
Autonomous Driving
🎯 What it does: Developed a system for low-cost urban positioning that relies solely on LoRa receivers and magnetometers.
LTL-D*: Incrementally Optimal Replanning for Feasible and Infeasible Tasks in Linear Temporal Logic Specifications
Jiming Ren, Ye Zhao
Optimization
🎯 What it does: Propose an incremental replanning algorithm called LTL-D* to satisfy Linear Temporal Logic (LTL) tasks in dynamic environments, capable of minimizing task specification violations through replanning when unexpected environmental changes cause task failure;
M3-GMN: A Multi-environment, Multi-LiDAR, Multi-task dataset for Grid Map based Navigation
Guanglei Xie, Zhenping Sun
Autonomous DrivingPoint CloudBenchmark
🎯 What it does: Proposed a multi-environment, multi-LiDAR, multi-task dataset to enhance grid-based autonomous driving navigation capabilities
MADE: Malicious Agent Detection for Robust Multi-Agent Collaborative Perception
Yangheng Zhao, Siheng Chen
Anomaly DetectionAutonomous DrivingImagePoint Cloud
🎯 What it does: Proposed a reactive defense method called MADE for detecting and removing potential malicious agents in multi-agent collaborative perception.
Magnetic Field Aided Vehicle Localization with Acceleration Correction
Mrunmayee Deshpande, J. Ramos
Autonomous DrivingSimultaneous Localization and MappingPhysics Related
🎯 What it does: Achieve vehicle localization and accelerometer bias correction by utilizing environmental magnetic fields and employing global mathematical function mapping and Euclidean distance matching techniques.
Magnetic tactile sensor with load tolerance and flexibility using frame structures for estimating triaxial contact force distribution of humanoid
Takuma Hiraoka, Masayuki Inaba
Robotic Intelligence
🎯 What it does: Proposed a load-tolerant tactile sensor that separates the force-receiving part from the measurement part using magnetic force, and protects the measurement components within the robot's framework; simultaneously, it learns the relationship between changes in physical measurements caused by elastic body deformation and contact force distribution to overcome the difficulties of modeling thick elastic bodies.
MAkEable: Memory-centered and Affordance-based Task Execution Framework for Transferable Mobile Manipulation Skills
Christoph Pohl, Tamim Asfour
Robotic Intelligence
🎯 What it does: Propose the MAkEable framework, integrating affordance-based task descriptions into the memory-centric cognitive architecture of the ARMAR series of robots, enabling cross-task, cross-environment, and cross-robot capability and knowledge transfer, with real experimental validation conducted across multiple robots, tasks, and environments.
Making the Flow Glow – Robot Perception under Severe Lighting Conditions using Normalizing Flow Gradients
Simon Kristoffersson Lind, Volker Krüger
Object DetectionOptimizationRobotic IntelligenceFlow-based Model
🎯 What it does: Using regularized flow gradients to optimize camera parameters, thereby enhancing the target detection performance of robots under extreme illumination
Malicious Path Manipulations via Exploitation of Representation Vulnerabilities of Vision-Language Navigation Systems
Chashi Mahiul Islam, Piyush Kumar
Adversarial AttackVision-Language-Action ModelImage
🎯 What it does: Using gradient-based optimization methods to make imperceptible modifications to images in visual language navigation (VLN) systems to attack robot navigation paths, and proposing an algorithm that utilizes noise-added sensitivity to detect malicious modifications.
ManiFoundation Model for General-Purpose Robotic Manipulation of Contact Synthesis with Arbitrary Objects and Robots
Zhixuan Xu, Lin Shao
Robotic IntelligencePoint Cloud
🎯 What it does: Propose a universal robot manipulation foundation model based on contact synthesis, which can output object contact points and corresponding contact forces or subsequent movements for various manipulation tasks by leveraging point clouds of objects and robot arms, physical properties, target motion, and contact area masks.
MANIP: A Modular Architecture for Integrating Interactive Perception for Robot Manipulation
Justin Yu (University of California Berkeley), Ken Goldberg (University of California Berkeley)
Robotic Intelligence
🎯 What it does: Proposed a modular system architecture called MANIP for the design and development of robot operating systems, systematically combining learned sub-policies with traditional algorithm primitives (such as inverse kinematics, Kalman filtering, RANSAC anomaly detection, PID control, etc.).
ManipVQA: Injecting Robotic Affordance and Physically Grounded Information into Multi-Modal Large Language Models
Siyuan Huang, Hao Dong
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: Developed the ManipVQA framework, which injects robotic grasping capabilities and physical knowledge into multimodal large language models through the VQA format, achieving tool detection, graspability recognition, and physical concept understanding.
Manta Ray-Inspired Soft Robotic Swimmer for High-speed and Multi-modal Swimming
Zefeng Xu, Yitong Zhou
Robotic Intelligence
🎯 What it does: Designed a flexible soft robotic swimmer based on manta ray fins
Map-Aware Human Pose Prediction for Robot Follow-Ahead
Qingyuan Jiang, Volkan Isler
Pose EstimationAutonomous DrivingRobotic Intelligence
🎯 What it does: For robot following tasks, predicting complete 3D human trajectories in complex environments
Map-based Modular Approach for Zero-shot Embodied Question Answering
Koya Sakamoto, M. Kawanabe
Robotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelSimultaneous Localization and MappingText
🎯 What it does: Propose a map-based modular approach for zero-shot embodied question answering (EQA), enabling real robots to explore unknown environments, construct maps, and answer natural language queries.
MAP-NBV: Multi-agent Prediction-guided Next-Best-View Planning for Active 3D Object Reconstruction
Harnaik Dhami, Pratap Tokekar
OptimizationRobotic IntelligenceAgentic AIMesh
🎯 What it does: Proposes a Multi-Agent Prediction Guidance-based Next Best View Planning (MAP-NBV) for Active 3D Object Reconstruction
MapLocNet: Coarse-to-Fine Feature Registration for Visual Re-Localization in Navigation Maps
Hang Wu, Tong Qin
Autonomous DrivingTransformerSimultaneous Localization and MappingImage
🎯 What it does: Proposes MapLocNet, a Transformer-based visual relocalization method that employs a coarse-to-fine neural feature registration to align navigation map features with visual bird's-eye view features, achieving localization without HD maps.
Markerless Aerial-Terrestrial Co-Registration of Forest Point Clouds using a Deformable Pose Graph
Benoit Casseau, Maurice F. Fallon
OptimizationSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Provides an automated pipeline to fuse data from airborne laser scanning and ground-based laser scanning into a unified forest point cloud reconstruction.
MARPF: Multi-Agent and Multi-Rack Path Finding
Hiroya Makino, Seigo Ito
Optimization
🎯 What it does: The study plans AGV paths in a channel-less environment to transport target racks, achieving collision-free movement by repositioning obstacle racks.
MARVIS: Motion & Geometry Aware Real and Virtual Image Segmentation
Jiayi Wu (University Of Maryland), Y. Aloimonos
SegmentationData SynthesisDomain AdaptationImage
🎯 What it does: Proposes a segmentation method for real and virtual image regions, utilizing synthetic images, domain-invariant information, motion entropy kernel, and epipolar geometric consistency to address segmentation under the influence of underwater dynamic disturbances;
Masked Mutual Guidance Transformer Tracking
Baojie Fan, Caiyu Zhang
Object TrackingTransformer
🎯 What it does: Propose a target feature learning method utilizing an encoding-decoding architecture and mask mutual guidance tracking (MMG). After joint visual feature extraction from the template and search region, self-decoding and mutual guidance decoding are performed separately to reconstruct the original image, promoting mutual understanding between the two. During the inference phase, the decoder is offloaded to achieve a simple and efficient tracker.
MaskingDepth: Masked Consistency Regularization for Semi-Supervised Monocular Depth Estimation
Jongbeom Baek, Seungryong Kim
Depth EstimationAutonomous DrivingImage
🎯 What it does: Proposes the MaskingDepth framework, which performs semi-supervised monocular depth estimation by leveraging depth consistency between strongly and weakly augmented images, thereby reducing reliance on large-scale real depth annotations.
Mastering Scene Rearrangement with Expert-Assisted Curriculum Learning and Adaptive Trade-Off Tree-Search
Zan Wang, Wei Liang
Robotic Intelligence
🎯 What it does: Propose a fine-grained action space definition, construct a large-scale indoor scene rearrangement dataset, and develop the PLATO framework, integrating expert-assisted curriculum learning with adaptive trade-off tree search to achieve efficient training and inference.
MCGMapper: Light-Weight Incremental Structure from Motion and Visual Localization with Planar Markers and Camera Groups
Yusen Xie, Jun Ma
OptimizationSimultaneous Localization and MappingImage
🎯 What it does: Propose a lightweight incremental framework that utilizes planar markers and known extrinsic parameters of multi-camera systems for optical structure from motion (SfM) and visual localization, capable of reconstructing marker maps in low-texture indoor environments and industrial scenarios.
MDHA: Multi-Scale Deformable Transformer with Hybrid Anchors for Multi-View 3D Object Detection
Michelle Adeline, Vishnu Monn Baskaran
Object DetectionAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Propose a sparse query framework MDHA, which constructs adaptive 3D proposals by combining hybrid anchors and depth prediction, and introduces Anchor Encoder and Circular Deformable Attention to improve efficiency.
MEMROC: Multi-Eye to Mobile RObot Calibration
Davide Allegro, S. Ghidoni
OptimizationRobotic IntelligenceSimultaneous Localization and MappingImageTime Series
🎯 What it does: Proposed a calibration method called MEMROC based on motion for multi-camera and mobile robot reference frameworks.
MERSYS: A Collaborative Estimation and Dense Mapping System for Multi-Agent Generic SLAM
Qianhua Lai, Jianxiao Zou
Robotic IntelligenceSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: Designed and implemented a multi-agent collaborative SLAM system called MERSYS, which fuses Lidar-Inertial Odometry (LIO) and Visual-Inertial Odometry (VIO) to generate global dense 3D point cloud maps and provide a bidirectional data transmission strategy with low communication costs.
Meta SAC-Lag: Towards Deployable Safe Reinforcement Learning via MetaGradient-based Hyperparameter Tuning
Homayoun Honari, Homayoun Najjaran
Hyperparameter SearchRobotic IntelligenceReinforcement Learning
🎯 What it does: Proposed the Meta SAC-Lag algorithm for safe reinforcement learning, which automatically adjusts safety-related hyperparameters through meta gradient optimization and validated its performance in simulated environments and real robotic arm experiments.