Pasar al contenido principal

ROBOTS MOVILES AUTONOMOS PARA RECOGIDA SELECTIVA DE RESIDUOS DOMESTICOS EN EXTERIORES

PID2021-122685OB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia
Subprograma Subprograma Estatal de Generación de Conocimiento
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2021
Plan Estatal 2021-2023 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSIDAD DE ALICANTE

Resultados relacionados

Resultados totales (Incluyendo duplicados): 19
Encontrada(s) 1 página(s)

RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

ViKi-HyCo: A Hybrid-Control Approach for Complex Car-Like Maneuvers

  • Velasco, Edison P.
  • Muñoz-Bañón, Miguel Á.
  • Candelas-Herías, Francisco A.
  • Puente Méndez, Santiago T.
  • Torres, Fernando
  • Candelas-Herías, Francisco A.|||/items/d4ce6549-c436-439d-ab35-c08c9c4398c2
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
While Visual Servoing is deeply studied to perform simple maneuvers, the literature does not commonly address complex cases where the target is far out of the camera’s field of view (FOV) during the maneuver. For this reason, in this paper, we present ViKi-HyCo (Visual Servoing and Kinematic Hybrid-Controller). This approach generates the necessary maneuvers for the complex positioning of a non-holonomic mobile robot in outdoor environments. In this method, we use LiDAR-camera fusion to estimate objects bounding boxes using image and metrics modalities. With the multi-modality nature of our representation, we can automatically obtain a target for a visual servoing controller. At the same time, we also have a metric target, which allows us to hybridize with a kinematic controller. Given this hybridization, we can perform complex maneuvers even when the target is far away from the camera’s FOV. The proposed approach does not require an object-tracking algorithm and can be applied to any robotic positioning task where its kinematic model is known. ViKi-HyCo has an error of 0.0428±0.0467 m in the X-axis and 0.0515±0.0323 m in the Y-axis at the end of a complete positioning task.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Robust Single Object Tracking and Following by Fusion Strategy

  • Olivas, Alejandro
  • Muñoz-Bañón, Miguel Á.
  • Velasco, Edison P.
  • Torres, Fernando
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Olivas, Alejandro|||/items/7077df23-8e63-4f0e-beed-27079ef54ba1
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
Single Object Tracking methods are yet not robust enough because they may lose the target due to occlusions or changes in the target’s appearance, and it is difficult to detect automatically when they fail. To deal with these problems, we design a novel method to improve object tracking by fusing complementary types of trackers, taking advantage of each other’s strengths, with an Extended Kalman Filter to combine them in a probabilistic way. The environment perception is performed with a 3D LiDAR sensor, so we can track the object in the point cloud and also in the front-view image constructed from the point cloud. We use our tracker-fusion method in a mobile robot to follow pedestrians, also considering the dynamic obstacles in the environment to avoid them. We show that our method allows the robot to follow the target accurately during long experimental sessions where the trackers independently fail, demonstrating the robustness of our tracker-fusion strategy.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Trajectory Generation Using Dual-Robot Haptic Interface for Reinforcement Learning from Demonstration

  • Frau Alfaro, Daniel|||/items/63e115be-278e-4066-97ad-066b1700d144
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
In learning robotics, techniques such as Learning from Demonstrations (LfD) and Reinforcement Learning (RL) have become widely popular among developers. However, this approximations can result in inefficient strategies when it comes to train more than one agent interacting in the same space with several objects and unknown obstacles. To solve this problematic, Reinforcement Learning from Demonstration (RLfD) allows the agent to learn and evaluate its performance from a set of demonstrations provided by a human expert while generalising from them using RL training. In dual-robot applications this approach is suitable for training agents that perform collaborative tasks. For this reason, a dual-robot haptic interface has been designed in order to produce dual manipulation trajectories to feed a RLfD agent. Haptics allows to perform high quality demonstrations following an impedance control approach. Trajectories obtained will be used as positive demonstrations so the training environment can generate automatic ones. As a result, this dual-robot haptic interface will provide a few trajectory demonstrations on dual manipulation in order to train agents using RL strategies. The aim of this research is to generate trajectories with this dual-robot haptic interface to train one or more agents following RLfD paradigms. Results show that trajectories performed with this interface present less error and deviation than others performed with a non-haptic interface, increasing the quality of the training data.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Sim-to-real transfer by hybrid Gaussian Splatting and geometric reconstruction for autonomous driving

  • Olivas, Alejandro
  • Muñoz-Bañón, Miguel Á.
  • Torres, Fernando
  • Olivas, Alejandro|||/items/7077df23-8e63-4f0e-beed-27079ef54ba1
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
Learning complex navigation behaviors in simulation and successfully transferring them to the real world remains a significant challenge due to the sim-to-real gap. To address this issue, we propose a hybrid reconstruction approach that combines Light Detection and Ranging (LiDAR) data and monocular images to create visually and structurally realistic simulated environments. Specifically, we extract an accurate mesh from LiDAR measurements and integrate it with geometry-consistent Three-Dimensional Gaussian Splatting (3DGS) to render photorealistic images. Additionally, we present an end-to-end neural network designed to generate control commands for a mobile robot, and it is used to assess the effectiveness of the hybrid reconstructions in narrowing the sim-to-real gap. In the proposed model, we introduce a novel auxiliary branch and define loss functions to improve the network’s ability to learn safe and robust actions. The method, which is based on imitation learning, is evaluated by deploying it on a real robot to navigate over a total distance of 3.5 kilometers, and experimental results show that models trained purely in simulation achieve performance comparable to those trained on real-world data.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

End-To-End Imitation Learning for Socially Aware Navigation using Memory-based Trajectory and Action Planning

  • Olivas, Alejandro
  • Muñoz-Bañón, Miguel Á.
  • Torres, Fernando
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Olivas, Alejandro|||/items/7077df23-8e63-4f0e-beed-27079ef54ba1
The increasing deployment of service robots and autonomous vehicles operating in human populated environments underscores the importance of planning trajectories to ensure pedestrian comfort while optimizing robot travel times. End-to-end models are promising for shortcutting the complex problem of pedestrian detection, trajectory prediction, and robot movement planning by using a single network. Imitation Learning (IL) has been widely adopted in self-driving because it enables implicit understandingof complex environments and expert driving behaviour, whereas socially aware navigate on has usually relied on Reinforcement Learning (RL) approaches that learn from abstracted environment representations, prioritizing information relevant to human interaction. In this research, we present Memory-based Trajectory and Action Planning (MTAP), an IL approach extended for socially aware navigation to perform intelligent pedestrian avoidance while maintaining a deeper understanding of the environment and the expert driving behaviour. Our method predicts the trajectory and the control action commands directly from the information of the sensors and the global target, employing a recurrent layer to anticipate pedestrian movements based on past frames. The network is trained and evaluated in the CARLA simulator, where we introduce a benchmark for assessing pedestrian avoidance efficacy, presenting also new social navigation metrics to measure both trajectory efficiency and pedestrian comfort during avoidance maneuvers.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Predicción de deslizamiento mediante la segmentación de imágenes táctiles, Slippage Prediction from Segmentation of Tactile Images

  • Castaño Amorós, Julio
  • Gil, Pablo
  • Castaño Amorós, Julio|||/items/773d394b-045e-42fe-8326-a8cbd3d8d5b7
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
El uso de sensores táctiles está comenzando a ser una práctica común en tareas complejas de manipulación robótica. Este tipo de sensores proporcionan información extra sobre las propiedades físicas de objetos que están siendo agarrados y/o manipulados. En este trabajo, se ha implementado un sistema capaz de medir el deslizamiento rotacional que pueden sufrir objetos durante su manipulación. Nuestra propuesta emplea sensores táctiles ópticos DIGIT, a partir de los cuáles se capturan imágenes de contacto que luego se procesan e interpretan. En concreto, nuestro método hace uso de un modelo neuronal para la detección de la región de contacto. Y posteriormente, mediante extracción de características visuales de la región detectada, se estima el ángulo causado por movimientos de deslizamiento. Nuestro método estima correctamente la región de contacto obteniendo un 95% y 91% usando las métricas Dice e IoU. Y es capaz de obtener un error medio máximo de 3º en agarres de objetos nunca vistos anteriormente.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Learning Dexterous Object Handover

  • Frau Alfaro, Daniel
  • Castaño Amorós, Julio
  • Puente Méndez, Santiago T.
  • Gil, Pablo
  • Calandra, Roberto
  • Frau Alfaro, Daniel|||/items/63e115be-278e-4066-97ad-066b1700d144
  • Castaño Amorós, Julio|||/items/773d394b-045e-42fe-8326-a8cbd3d8d5b7
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
Object handover is an important skill that we use daily when interacting with other humans. To deploy robots in collaborative setting, like houses, being able to receive and handing over objects safely and efficiently becomes a crucial skill. In this work, we demonstrate the use of Reinforcement Learning (RL) for dexterous object handover between two multi-finger hands. Key to this task is the use of a novel reward function based on dual quaternions to minimize the rotation distance, which outperforms other rotation representations such as Euler and rotation matrices. The robustness of the trained policy is experimentally evaluated by testing w.r.t. objects that are not included in the training distribution, and perturbations during the handover process. The results demonstrate that the trained policy successfully perform this task, achieving a total success rate of 94% in the best-case scenario after 100 experiments, thereby showing the robustness of our policy with novel objects. In addition, the best-case performance of the trained policy decreases by only 13.8% when the other robot moves during the handover, proving that our policy is also robust to this type of perturbation, which is common in realworld object handovers. Code and videos can be found here.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

LiLO: lightweight and low-bias LiDAR odometry method based on spherical range image filtering

  • Velasco, Edison P.
  • Muñoz-Bañón, Miguel Á.
  • Candelas-Herías, Francisco A.
  • Puente Méndez, Santiago T.
  • Torres, Fernando
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Candelas-Herías, Francisco A.|||/items/d4ce6549-c436-439d-ab35-c08c9c4398c2
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
Existing low-bias LiDAR odometry methods are often computationally demanding, thus complicating their integration into more complex autonomous navigation systems such as path planning or global localization. To address this problem, we present a lightweight LiDAR odometry method that converts unorganized point cloud data into a spherical range image (SRI) and filters out surface, edge, and ground features in the image plane. This significantly reduces computation time and the required features for odometry estimation in LOAM-based algorithms. As our odometry estimation method is designed to serve as a stepping stone in mobile robotics systems, it does not rely on global maps or loop closure algorithms, which further reduces computational costs. Experimental results demonstrate that even when we achieve a time-efficient strategy, we still achieve state-of-the-art performance, generating a translation and rotation error of 0.86% and 0.0036∘/m on the KITTI dataset with an average runtime of 78 ms. In addition, we tested the method with our data, obtaining an average closed-loop error of 0.8 m and a runtime of 27 ms over eight loops covering 3.5 km. The code is available at: https://github.com/AUROVA-LAB/applications/tree/main/app_LiLO.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Diseño de recompensas mediante LLMs para tareas de manipulación robótica, Reward design via LLMs for robot manipulation tasks

  • Fernández-Herrero, Moisés
  • Puente Méndez, Santiago T.
  • Páez Ubieta, Ignacio de Loyola
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
El diseño manual de funciones de recompensa para el Aprendizaje por Refuerzo (RL) en robótica es complejo y propenso a errores. Este trabajo investiga la automatización de dicho proceso mediante Grandes Modelos de Lenguaje (LLMs), ampliando la metodología Eureka. Se evaluó la capacidad de diversos LLMs del mercado, más allá de GPT-4 y GPT-3.5 estudiados en el trabajo original, para generar recompensas en tres tareas de manipulación robótica con las manos Shadow y Allegro en el simulador Isaac Gym. Los resultados muestran que los LLMs, especialmente modelos recientes y aquellos con razonamiento Chain-of-Thought, pueden superar las recompensas diseñadas por expertos humanos en el 100 % de las tareas evaluadas, logrando éxito en entornos de alta complejidad como pen spinning con la Allegro Hand. Modelos como O1 y algunas variantes de Claude destacan significativamente. El estudio confirma el gran potencial de los LLMs para optimizar el diseño de recompensas en RL aplicado a la realización de tareas complejas en robótica.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Geo-Localization Based on Dynamically Weighted Factor-Graph

  • Muñoz-Bañón, Miguel Á.
  • Olivas, Alejandro
  • Velasco, Edison P.
  • Candelas-Herías, Francisco A.
  • Torres, Fernando
  • Candelas-Herías, Francisco A.|||/items/d4ce6549-c436-439d-ab35-c08c9c4398c2
  • Muñoz-Bañón, Miguel Á.|||/items/aaa2466d-a60d-4d03-9607-36d34ecacc9a
  • Olivas, Alejandro|||/items/7077df23-8e63-4f0e-beed-27079ef54ba1
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
Feature-based geo-localization relies on associating features extracted from aerial imagery with those detected by the vehicle's sensors. This requires that the type of landmarks must be observable from both sources. This lack of variety of feature types generates poor representations that lead to outliers and deviations produced by ambiguities and lack of detections, respectively. To mitigate these drawbacks, in this letter, we present a dynamically weighted factor graph model for the vehicle's trajectory estimation. The weight adjustment in this implementation depends on information quantification in the detections performed using a LiDAR sensor. Also, a prior (GNSS-based) error estimation is included in the model. Then, when the representation becomes ambiguous or sparse, the weights are dynamically adjusted to rely on the corrected prior trajectory, mitigating outliers and deviations in this way. We compare our method against state-of-the-art geo-localization ones in a challenging and ambiguous environment, where we also cause detection losses. We demonstrate mitigation of the mentioned drawbacks where the other methods fail.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Force-based touch approach for volume estimation, Aproximación de tacto basada en fuerza para estimación de volumen

  • Castaño Amorós, Julio
  • Trebuchon, Killian
  • Gil, Pablo
  • Mezouar, Youcef
  • Castaño Amorós, Julio|||/items/773d394b-045e-42fe-8326-a8cbd3d8d5b7
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
Optimal robotic grasping cannot be limited to the estimation of object grasping pose using vision-based methods. It is necessary to use tactile sensors to learn the physical properties of the objects that are to be grasped. In this work, we integrated two Contactile force-based tactile sensors with a 2F-140 ROBOTIQ gripper and a UR5 robot to estimate the volume of a waterf illed container using Multilayer Perceptron (MLP) neural networks. During experimentation, we trained and evaluated different MLPs varying the input forces (Fx, Fy, Fz) in a task of discrete-volume regression in a range of between 0ml and 300ml. The preliminary proposed approach is compared with an algebraic method based on the diagram of the equilibrium of forces, proving that our results are more precise, obtaining a R2 value of 8% higher in the worst-case scenario, and of 30% in the best.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

LiCAR: pseudo-RGB LiDAR image for CAR segmentation

  • Páez Ubieta, Ignacio de Loyola
  • Velasco, Edison P.
  • Puente Méndez, Santiago T.
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
With the advancement of computing resources, an increasing number of Neural Networks (NNs) are appearing for image detection and segmentation appear. However, these methods usually accept as in put a RGB 2D image. On the other side, Light Detection And Ranging (LiDAR) sensors with many layers provide images that are similar to those obtained from a traditional low resolution RGB camera. Following this principle, a new dataset for segmenting cars in pseudo-RGB images has been generated. This dataset combines the information given by the LiDAR sensor into a Spherical Range Image (SRI), concretely the reflectivity, near infrared and signal intensity 2D images. These images are then fed into instance segmentation NNs. These NNs segment the cars that appear in these images, having as result a Bounding Box (BB) and mask precision of 88% and 81.5% respectively with You Only Look Once (YOLO)-v8 large. By using this segmentation NN, some trackers have been applied so as to follow each car segmented instance along a video feed, having great performance in real world experiments.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Visual Servoing NMPC Applied to UAVs for Photovoltaic Array Inspection

  • Velasco, Edison P.
  • Recalde, Luis F.
  • Guevara, Bryan S.
  • Varela-Aldás, José
  • Candelas-Herías, Francisco A.
  • Puente Méndez, Santiago T.
  • Gandolfo, Daniel C.
  • Candelas-Herías, Francisco A.|||/items/d4ce6549-c436-439d-ab35-c08c9c4398c2
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
The photovoltaic (PV) industry is seeing a significant shift toward large-scale solar plants, where traditional inspection methods have proven to be time-consuming and costly. Currently, the predominant approach to PV inspection using unmanned aerial vehicles (UAVs) is based on the capture and detailed analysis of aerial images (photogrammetry). However, the photogrammetry approach presents limitations, such as an increased amount of useless data and potential issues related to image resolution that negatively impact the detection process during high-altitude flights. In this work, we develop a visual servoing control system with dynamic compensation using nonlinear model predictive control (NMPC) applied to a UAV. This system is capable of accurately tracking the middle of the underlying PV array at various frontal velocities and height constraints, ensuring the acquisition of detailed images during low-altitude flights. The visual servoing controller is based on extracting features using RGB-D images and employing a Kalman filter to estimate the edges of the PV arrays. Furthermore, this work demonstrates the proposal in both simulated and real-world environments using the commercial aerial vehicle (DJI Matrice 100), with the purpose of showcasing the results of the architecture.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Robotic Approach Trajectory Using Reinforcement Learning with Dual Quaternions

  • Frau Alfaro, Daniel
  • Puente Méndez, Santiago T.
  • Páez Ubieta, Ignacio de Loyola
  • Velasco, Edison P.
  • Frau Alfaro, Daniel|||/items/63e115be-278e-4066-97ad-066b1700d144
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
Manipulation tasks in robotics usually involve two phases: an approach to the object and the grasp itself. The first action allows the robot to reach a certain pose in space that is likely to allow the object to be manipulated. Reinforcement Learning (RL) techniques allow a policy to be learned through experience given a set of states and actions, so this is a powerful tool for developing controllers for specific tasks such as positioning the robot in a particular point in space. However, when manipulating an object, orientation is as relevant as position. For this reason, a method of RL for positioning the robot's end effector in a suitable position and orientation for manipulation in simulation is presented. This approach models the problem of computing the distance for the reward function using dual quaternions parameterisation, an element that can represent the pose and attitude of a rigid body in Euclidean space in a compact way without having to apply any constraints.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

DualQuat-LOAM: LiDAR odometry and mapping parameterized on dual quaternions

  • Velasco, Edison P.
  • Recalde, Luis
  • Li, Guanrui
  • Candelas-Herías, Francisco A.
  • Puente Méndez, Santiago T.
  • Torres, Fernando
  • Candelas-Herías, Francisco A.|||/items/d4ce6549-c436-439d-ab35-c08c9c4398c2
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Torres, Fernando|||/items/17ced7cb-5c1d-429d-b922-315f91fddc33
  • Velasco, Edison P.|||/items/66a95307-8df4-43a8-ac52-24c774c581ab
This paper reports on a novel method for LiDAR odometry estimation, which completely parameterizes the system with dual quaternions. To accomplish this, the features derived from the point cloud, including edges, surfaces, and Stable Triangle Descriptor (STD), along with the optimization problem, are expressed in the dual quaternion set. This approach enables the direct combination of translation and orientation errors via dual quaternion operations, greatly enhancing pose estimation, as demonstrated in comparative experiments against other state-of-the-art methods. Our approach reduced drift error compared to other LiDAR-only-odometry methods, especially in scenarios with sharp curves and aggressive movements with large angular displacement. DualQuat-LOAM is benchmarked against several public datasets. In the KITTI dataset it has a translation and rotation error of 0.79 and 0.0039°/m, with an average run time of 53 ms.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

QDGset: A Large Scale Grasping Dataset Generated with Quality-Diversity

  • Huber, Johann
  • Hélénon, François
  • Kappel, Mathilde
  • Páez Ubieta, Ignacio de Loyola
  • Puente Méndez, Santiago T.
  • Gil, Pablo
  • Amar, Faïz Ben
  • Doncieux, Stéphane
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
Recent advances in AI have led to significant results in robotic learning, but skills like grasping remain partially solved. Many recent works exploit synthetic grasping datasets to learn to grasp unknown objects. However, those datasets were generated using simple grasp sampling methods using priors. Recently, Quality-Diversity (QD) algorithms have been proven to make grasp sampling significantly more efficient. In this work, we extend QDG-6DoF, a QD framework for generating object-centric grasps, to scale up the production of synthetic grasping datasets. We propose a data augmentation method that combines the transformation of object meshes with transfer learning from previous grasping repertoires. The conducted experiments show that this approach reduces the number of required evaluations per discovered robust grasp by up to 20%. We used this approach to generate QDGset, a dataset of 6DoF grasp poses that contains about 3.5 and 4.5 times more grasps and objects, respectively, than the previous state-of-the-art. Our method allows anyone to easily generate data, eventually contributing to a large-scale collaborative dataset of synthetic grasps.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Measuring Object Rotation via Visuo-Tactile Segmentation of Grasping Region

  • Castaño Amorós, Julio
  • Gil, Pablo
  • Castaño Amorós, Julio|||/items/773d394b-045e-42fe-8326-a8cbd3d8d5b7
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
When carrying out robotic manipulation tasks, objects occasionally fall as a result of the rotation caused by slippage. This can be prevented by obtaining tactile information that provides better knowledge on the physical properties of the grasping. In this paper, we estimate the rotation angle of a grasped object when slippage occurs. We implement a system made up of a neural network with which to segment the contact region and an algorithm with which to estimate the rotated angle of that region. This method is applied to DIGIT tactile sensors. Our system has additionally been trained and tested with our publicly available dataset which is, to the best of our knowledge, the first dataset related to tactile segmentation from non-synthetic images to appear in the literature, and with which we have attained results of 95% and 90% as regards Dice and IoU metrics in the worst scenario. Moreover, we have obtained a maximum error of ≈ 3° when testing with objects not previously seen by our system in 45 different lifts. This, therefore, proved that our approach is able to detect the slippage movement, thus providing a possible reaction that will prevent the object from falling.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Transferability of labels between multilens cameras

  • Páez Ubieta, Ignacio de Loyola
  • Frau Alfaro, Daniel
  • Páez Ubieta, Ignacio de Loyola|||/items/f0f7ccea-9e51-49c7-80ca-6367f1003936
  • Frau Alfaro, Daniel|||/items/63e115be-278e-4066-97ad-066b1700d144
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
In this work, a new method for automatically extending Bounding Box (BB) and mask labels across different channels on multilens cameras is presented. For that purpose, the proposed method combines the well known phase correlation method with a refinement process. During the first step, images are aligned by localizing the peak of intensity obtained in the spatial domain after performing the cross correlation process in the frequency domain. The second step consists of obtaining the best possible transformation by using an iterative process maximising the IoU (Intersection over Union) metric. Results show that, by using this method, labels could be transferred across different lens on a camera with an accuracy over 90% in most cases and just by using 65 ms in the whole process. Once the transformations are obtained, artificial RGB images are generated, for labeling them so as to transfer this information into each of the other lens. This work will allow users to use this type of cameras in more fields rather than satellite or medical imagery, giving the chance of labeling even invisible objects in the visible spectrum.
RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Learning Dexterous Object Handover

  • Frau Alfaro, Daniel
  • Castaño Amorós, Julio
  • Puente Méndez, Santiago T.
  • Gil, Pablo
  • Calandra, Roberto
  • Frau Alfaro, Daniel|||/items/63e115be-278e-4066-97ad-066b1700d144
  • Castaño Amorós, Julio|||/items/773d394b-045e-42fe-8326-a8cbd3d8d5b7
  • Puente Méndez, Santiago T.|||/items/b02d19c1-e93f-45cd-9367-7c228d585248
  • Gil, Pablo|||/items/bb6cdca6-1fdd-44c0-a64a-6e328dbb74e9
Object handover is an important skill that we use daily when interacting with other humans. To deploy robots in collaborative setting, like houses, being able to receive and handing over objects safely and efficiently becomes a crucial skill. In this work, we demonstrate the use of Reinforcement Learning (RL) for dexterous object handover between two multi-finger hands. Key to this task is the use of a novel reward function based on dual quaternions to minimize the rotation distance, which outperforms other rotation representations such as Euler and rotation matrices. The robustness of the trained policy is experimentally evaluated by testing w.r.t. objects that are not included in the training distribution, and perturbations during the handover process. The results demonstrate that the trained policy successfully perform this task, achieving a total success rate of 94% in the best-case scenario after 100 experiments, thereby showing the robustness of our policy with novel objects. In addition, the best-case performance of the trained policy decreases by only 13.8% when the other robot moves during the handover, proving that our policy is also robust to this type of perturbation, which is common in real-world object handovers.