Elderly Assisting Robot: Communication Through Physical Interaction

The world’s population is aging at an ever increasing pace. There is the need to develop technologies that will support the independent functioning of older people so that they are able to remain living in the community, postponing the transition to living in institutional settings for as long as possible – a movement referred to as aging in place. Assistive robots hold great promise in that respect. However, one of the main obstacles to the widespread use of robotic assistants is the lack of adequate interfaces: the elderly person should be able to communicate with the robot naturally, similarly to the way he/she would communicate with a human caregiver. Therefore, if the robot is to assist with daily activities it needs to respond to all types of user input. We hypothesize that in addition to well established modalities of communication like language, gestures and vision, physical interaction – a bi-directional exchange of forces during a direct or indirect (through an object held by the robot and the user) contact – also plays an important role and should be studied to develop an effective robotic interface. A multimodal interface will not only make communication better, by sharing the information between the modalities there is no need for highly accurate interfaces for each individual modality.

Reliability of Robots

Correct functioning of cyber-physical systems is of critical importance. This is more so in the case of safety critical systems such as in medical, automotive and many other applications. Since verification of correctness, in general, is infeasible and testing is not exhaustive, it is of critical importance to monitor such system during their operation and detect erroneous behaviors to be acted on.

Monitoring techniques are investigated, when the system state is not fully observable, it’s behavior is stochastic and the property to be monitored is specified on the system trajectory. Theoretically, two notions called monitorability and strong monitorability are difined and the necessary and sufficient conditions characterizing them are also given. General monitoring techniques for cases when systems are modeled as stochastic hybrid automata, and the properties are specified as safety or liveness automata are presented State estimation is another key step in the monitoring, since the property is defined on system behavior which is hidden and partial observable. Rao-Blackwellised Particle Filtering is employed in this hybrid state framework.

More efficient and complicated experiments are in progress.

Wireless Sensor Networks

When exploring a scalar field, for example, concentration of oil spill over Mexico Gulf, by a group of sensors, it is always a question that where to deploy them over the area of interest. In general, the deployment of the sensors satisfies our demand, such as covering the whole area of interest.

In our project, we need more sensors to cover the part where the amount of information over that part is higher than others. Thus, we need an index for every point over the area of interest to quantify information density. In our research, We found that curvature is a good choice to quantify the information density of the field. By definition, curvature of a surface describes the amount by which the surface deviates from being a flat plane. At a certain point, the higher the curvature is, the more sharply the surface bends. Accordingly, the higher information density that point has. Since curvature is a function of the field, i.e., it can not be read directly from the sensors, we estimate it from the splines which approximate the field. Moreover, all sensors are mobile ones. So, a step-by-step motion algorithm should be embedded to drive them to an optimal sensing configuration, which means within each sensor’s region of responsibility, the integral of the curvature, i.e., the amount of the information, should be the same.

It’s easy to reach optimal deployment given there is a central server that collects sensor readings and positions of all sensor nodes. Nevertheless, We want to make our task a distributed one which is more challenging. In our problem settings, we assume that there is no central server, each sensor node decides where to move based on the information only provided by its instant neighbors.

So, there are two key problems for us. The first one is how to approximate the field by splines. Once we know the field, we may compute the curvature via the field. The second one is based on local information, where each sensor should move to achieve the global optimization deployment.