Time-To-Contact Forecasting by Modeling the Apparent Size of Obstacles

The growing demand for rapid response to tasks that robotic systems carried out, makes us increasingly think about robust and fast approaches. This paper presents a novel approach based on modeling the task known as Time-to-contact, in order to forecast the time it would take for a robot to collide with a detected object. For this we use an statistical approach based on time series. In addition, the location of potential obstacles is performed by passive sensors (monocular vision) in order to save energy and money than other more sophisticated sensors. Experimental results from the application of the method on real scenarios from other publications are reported.