Novel high-resolution pressure-sensor arrays allow treating pressure readings as standard images. Computer vision algorithms and methods such as convolutional neural networks (CNN) can be used to identify contact objects. In this paper, a high-resolution tactile sensor has been attached to a robotic end-effector to identify contacted objects. Two CNN-based approaches have been employed to classify pressure images. These methods include a transfer learning approach using a pre-trained CNN on an RGB-images dataset and a custom-made CNN (TactNet) trained from scratch with tactile information. The transfer learning approach can be carried out by retraining the classification layers of the network or replacing these layers with an SVM. Overall, 11 configurations based on these methods have been tested: eight transfer learning-based, and three TactNet-based. Moreover, a study of the performance of the methods and a comparative discussion with the current state-of-the-art on tactile object recognition is presented.

The use of tactile perception can help first response robotic teams in disaster scenarios, where visibility conditions are often reduced due to the presence of dust, mud, or smoke, distinguishing human limbs from other objects with similar shapes. Here, the integration of the tactile sensor in adaptive grippers is evaluated, measuring the performance of an object recognition task based on deep convolutional neural networks (DCNNs) using a flexible sensor mounted in adaptive grippers. A total of 15 classes with 50 tactile images each were trained, including human body parts and common environment objects, in semi-rigid and flexible adaptive grippers based on the fin ray effect. The classifier was compared against the rigid configuration and a support vector machine classifier (SVM). Finally, a two-level output network has been proposed to provide both object-type recognition and human/non-human classification. Sensors in adaptive grippers have a higher number of non-null tactels (up to 37% more), with a lower mean of pressure values (up to 72% less) than when using a rigid sensor, with a softer grip, which is needed in physical human–robot interaction (pHRI). A semi-rigid implementation with 95.13% object recognition rate was chosen, even though the human/non-human classification had better results (98.78%) with a rigid sensor.

This paper proposes a hybrid learning and optimization framework for mobile manipulators for complex and physically interactive tasks. The framework exploits an admittance-type physical interface to obtain intuitive and simplified human demonstrations and Gaussian Mixture Model (GMM)/Gaussian Mixture Regression (GMR) to encode and generate the learned task requirements in terms of position, velocity, and force profiles. Next, using the desired trajectories and force profiles generated by GMM/GMR, the impedance parameters of a Cartesian impedance controller are optimized online through a Quadratic Program augmented with an energy tank to ensure the passivity of the controlled system. Two experiments are conducted to validate the framework, comparing our method with two approaches with constant stiffness (high and low). The results showed that the proposed method outperforms the other two cases in terms of trajectory tracking and generated interaction forces, even in the presence of disturbances such as unexpected end-effector collisions.

In this paper, a novel method of active tactile perception based on 3D neural networks and a high-resolution tactile sensor installed on a robot gripper is presented. A haptic exploratory procedure based on robotic palpation is performed to get pressure images at different grasping forces that provide information not only about the external shape of the object, but also about its internal features. The gripper consists of two underactuated fingers with a tactile sensor array in the thumb. A new representation of tactile information as 3D tactile tensors is described. During a squeeze-and-release process, the pressure images read from the tactile sensor are concatenated forming a tensor that contains information about the variation of pressure matrices along with the grasping forces. These tensors are used to feed a 3D Convolutional Neural Network (3D CNN) called 3D TactNet, which is able to classify the grasped object through active interaction. Results show that 3D CNN performs better, and provide better recognition rates with a lower number of training data.

Collaborative robots (cobots) open up new avenues in the fields of industrial robotics and physical Human-Robot Interaction (pHRI) as they are suitable to work in close approximation and in collaboration with humans. The integration and control of variable stiffness elements allow inherently safe interaction. Apart from notable work on Variable Stiffness Actuators, the concept of Variable-Stiffness-Link (VSL) manipulators promises safety improvements in cases of unintentional physical collisions. However, position control of these type of robotic manipulators is challenging for critical task-oriented motions (e.g., pick and place). Hence, the study of open-loop position control for VSL robots is crucial to achieve high levels of safety, accuracy and hardware cost-efficiency in pHRI applications. In this paper, we propose a hybrid, learning based kinematic modelling approach to improve the performance of traditional open-loop position controllers for a modular, collaborative VSL robot. We show that our approach improves the performance of traditional open-loop position controllers for robots with VSL and compensates for position errors, in particular, for lower stiffness values inside the links: Using our upgraded and modular robot, two experiments have been carried out to evaluate the behaviour of the robot during taskoriented motions. Results show that traditional model-based kinematics are not able to accurately control the position of the end-effector: the position error increases with higher loads and lower pressures inside the VSLs. On the other hand, we demonstrate that, using our approach, the VSL robot can outperform the position control compared to a robotic manipulator with 3D printed rigid links.