Tomasz Mańkowski

In this article, the design of a five-fingered anthropomorphic gripper is presented specifically designed for the manipulation of elastic objects. The manipulator features a hybrid design, being equipped with three fully actuated fingers for precise manipulation, and two underactuated, tendon-driven digits for secure power grasping. For ease of reproducibility, the design uses as many off-the-shelf and 3D-printed components as possible. The on-board controller circuit and firmware are also presented. The design includes resistive position and angle sensors in each joint, resulting in full joint observability. The controller has a position-based controller integrated, along with USB communication protocol, enabling gripper state reporting and direct motor control from a PC. A high-level driver operating as a Robot Operating System node is also provided. All drives and circuitry of the PUT-Hand are integrated within the hand itself. The sensory system of the hand includes tri-axial optical force sensors placed on fully actuated fingers’ fingertips for reaction force measurement. A set of experiments is provided to present the motion and perception capabilities of the gripper. All design files and source codes are available online under CC BY-NC 4.0 and MIT licenses.

In this paper, we present a putEMG dataset intended for the evaluation of hand gesture recognition methods based on sEMG signal. The dataset was acquired for 44 able-bodied subjects and include 8 gestures (3 full hand gestures, 4 pinches and idle). It consists of uninterrupted recordings of 24 sEMG channels from the subject’s forearm, RGB video stream and depth camera images used for hand motion tracking. Moreover, exemplary processing scripts are also published. The putEMG dataset is available under a Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0). The dataset was validated regarding sEMG amplitudes and gesture recognition performance. The classification was performed using state-of-the-art classifiers and feature sets. An accuracy of 90% was achieved for SVM classifier utilising RMS feature and for LDA classifier using Hudgin’s and Du’s feature sets. Analysis of performance for particular gestures showed that LDA/Du combination has significantly higher accuracy for full hand gestures, while SVM/RMS performs better for pinch gestures. The presented dataset can be used as a benchmark for various classification methods, the evaluation of electrode localisation concepts, or the development of classification methods invariant to user-specific features or electrode displacement.

In this article, a method for kinematic configuration estimation of a structure similar to a human finger, is presented. The method is based on the EKF and a model reflecting kinematic constraints of a finger-like structure (2-DOF metacarpophalangeal joint, and one 1-DOF proximal interphalangeal rotational joint), using 3 low cost IMUs. During tests, the IMUs were attached to a 3D-printed setup equipped with encoders. The setup was fixed to the flange of a UR3 robotic arm. System accuracy and robustness was tested during various kinds of movement, including fast and slow movements in all joints (maximum angular rate of 550 deg/s), and simultaneous base rotation (up to 170 deg/s) and linear acceleration (up to 0.25 g). The obtained RMSE of estimated joint angle was 3.5 deg for all joints and all types of performed movements. In case of joint rotations around an axis parallel to gravity vector, RMSE of 4 deg was achieved.

A forearm band consisting of 7 EMG sensors was developed. The band is dedicated to serve as a human-machine interface and its applicability as a gesture-based interface was presented. The gesture recognition was performed by ANN with softmax output function. The classifier uses output entropy function to discriminate between known command gestures and unknown gestures. 15 features of low computational complexity were selected. The preliminary test was performed for 4 healthy volunteers. Continuous time series were used, including unknown gestures and transitions. Special attention was paid to analyze system properties of detecting and rejecting unknown gestures (not commands movements), even if they were not included in training set. Obtained preliminary results show the sensitivity for command gestures (including transition phases) of 97% and fall out of 1.9%. In case of the system trained to reject unknown gestures the sensitivity and fall out were 82% and 9%, respectively. The calculated unknown gestures rejection rate was at 96%. It was shown that in most cases the interface is robust to false positive detection of command gestures while performing other gestures, even if the same muscles groups were recruited for movement.