Abstract-Recent decades have seen significant progress in the field of artificial hands. Most of the surveys, which try to capture the latest developments in this field, focused on actuation and control systems of these devices. In this work, our goal is to provide a comprehensive survey of the sensors for artificial hands. In order to present the evolution of the field, we cover five year periods starting at the turn of the millennium. At each period, we present the robot hands with a focus on their sensor systems dividing them into categories such as prosthetics, research devices and industrial end-effectors. We also cover the sensors developed for robot hand usage in each era. Finally, the period between 2010 and 2015 introduces the reader the state-ofthe-art and also hints to the future directions in the sensor development for artificial hands.

This paper presents a grasping database collected from multiple human subjects for activities of daily living in unstructured environments. The main strength of this database is the use of three different sensing modalities: color images from a head-mounted action camera, distance data from a depth sensor on the dominant arm and upper body kinematic data acquired from an inertial motion capture suit. 3826 grasps were identified in the data collected during 9-hours of experiments. The grasps were grouped according to a hierarchical taxonomy into 35 different grasp types. The database contains information related to each grasp and associated sensor data acquired from the three sensor modalities. We also provide our data annotation software written in Matlab as an open-source tool. The size of the database is 172 GB. We believe this database can be used as a stepping stone to develop big data and machine learning techniques for grasping and manipulation with potential applications in rehabilitation robotics and intelligent automation.

No abstract

We consider the task of detecting sentences that express causality, as a step towards mining causal relations from texts. To bypass the scarcity of causal instances in relation extraction datasets, we exploit transfer learning, namely ELMO and BERT, using a bidirectional GRU with self-attention (BIGRUATT) as a baseline. We experiment with both generic public relation extraction datasets and a new biomedical causal sentence detection dataset, a subset of which we make publicly available. We find that transfer learning helps only in very small datasets. With larger datasets, BIGRU-ATT reaches a performance plateau, then larger datasets and transfer learning do not help.

The performance of a mobile robot can be improved by utilizing different locomotion modes in various terrain conditions. This creates the necessity of having a supervisory controller capable of recognizing different terrain types and changing the locomotion mode of the robot accordingly. This work focuses on the locomotion strategy selection problem for a hybrid legged wheeled mobile robot. Supervisory control of the robot is accomplished by the terrain recognizer, which classifies depth images obtained from a commercial time of flight depth sensor and selects different locomotion mode subcontrollers based on the recognized terrain type. For the terrain recognizer, a database is generated consisting of five terrain classes (Uneven, Level Ground, Stair Up, Stair Down, and Nontraversable). Depth images are enhanced using confidence map based filtering. The accuracy of the terrain classification using Support Vector Machine classifier for the testing database in five-class terrain recognition problem is 97%. Real-world experiments assess the locomotion abilities of the quadruped and the capability of the terrain recognizer in real-time settings. The results of these experiments show depth images processed in real time using machine learning algorithms can be used for the supervisory control of hybrid robots with legged and wheeled locomotion capabilities.