After decades of theoretical study in physiology and neurology communities, the paradigm of muscle synergies is now being explored in rehabilitation robotics as a strategy to control mechanical artifacts with many degrees-of-freedom (DoF) in a simple yet effective and human-like way. In particular, muscle synergies during grasping and in graded-force tasks are of great interest for the control of dexterous hand prostheses. To this end, we have designed and tested a novel device to accurately and simultaneously measure fingertip forces. The device, called FFLS (Finger-Force Linear Sensor), measures the forces applied by the human fingertips in both directions (flexion and extension of index, middle, ring and little finger plus thumb rotation and abduction/adduction). It is suited for several different hand sizes, enforces high accuracy in the measurement and its signal is guaranteed to be linear in a high range of forces (100N in both directions for each finger). It outputs six analog voltages (±10V), suited for processing with a DAQ card.

In previous work we showed that some human Voluntary Muscle Contractions (VMCs) of high interest to the prosthetics community, namely finger flexions/extensions and thumb rotation, can be effectively predicted using muscle activation signals coming from surface electromyography (sEMG). In this paper we study the effectiveness of various subsampling strategies to limit the size of the training data set, with the aim of extending the approach to an online VMC-prediction system whose main application will be force-controlled hand prostheses. We performed an experiment in which 10 ablebodied participants flexed and extended their fingers according to a visual stimulus, while muscle activations and VMCs (represented as synergistic fingertip forces) were gathered using sEMG electrodes and a custom-built measurement device. A Support Vector Machine (SVM) was trained on a fixed-sized subset of the collected data, obtained using seven different subsampling strategies. The SVM was then tested on subsequent new data. Our experimental results show that two subsampling strategies attain a prediction error as low as 6% to 12%, which is comparable to the error values obtained in our previous work when the entire data set was used and processed offline.

As the desire to see robots ubiquitous in society grows, so does the need for providing the robots with the means of building awareness of any humans with which it may be sharing the environment. This paper presents a real-world suitable system which enables robots to robustly perceive the presence of people acoustically. The proposed binaural system first identifies voiced signal by means of a novel approach to Voice Activity Detection that exploits the spectral signature and characteristics of speech without reliance on a priori knowledge. Bearing estimates for each speaker are then made using a multitrack particle filter with a belief update function comprised of a Cross-correlation bearing estimate and an estimate of the speaker's fundamental frequency. Results, from an evaluation of each of the major system components and a system evaluation in which the robot successfully built human-centric situational awareness of the three humans with which it shared an office lunch-room containing typical background noises, are presented and discussed.

Action recognition in surveillance systems has to work 24/7 under all kinds of weather and lighting conditions. Towards this end, most action recognition systems only work in the visible spectrum which limits their general usage to daytime applications. In this work Hough forests are applied to the longwave infrared spectrum which can capture humans both in the dark and in daylight. Further, Integral Channel Features which have shown promising results in the spatial domain are applied to the spatio-temporal domain and are incorporated into the Hough forest approach. This approach is evaluated on a new outdoor dataset containing different violent and non-violent actions recorded in the visible and infrared spectrum. It is further shown that for the visible spectrum the proposed approach achieves state-of-the-art results on the KTH and i3DPost dataset.

Appearance-based action recognition can be considered as a natural extension of appearance-based object detection from the spatial to the spatio-temporal domain. Although this step seems natural, most action recognition approaches are evaluated in isolation. Towards this end the contribution of this paper is twofold. First, a view-independent approach to action recognition is proposed and second the sensitivity w.r.t. a combination of person detection and action recognition is evaluated. Action recognition is performed in a hierarchical manner: First, the relative camera orientation in the scene is estimated and second, the action is determined using view-dependent Hough forests. The proposed approach is evaluated on the multi-view i3DPost dataset [1] and its performance is compared to single-step approaches using Hough forests. The results suggest that the recognition rate increases, when using the proposed hierarchical method compared to single-step approaches. Further, the performance rates of hierarchical Hough forests on ground truth data are compared to the results of hierarchical Hough forests in combination with a person detector.