Context-Aware Action with a Small Mobile Robot

Abstract

Simultaneous advances in mobile GPU computing and real-time object recognition now enable machines to make decisions and take actions based on the detection of objects of interest in the environment. An implementation of a mobile robot system that combines autonomous exploration and mapping capabilities with a real-time object recognition method based on a deep neural network running on a mobile GPU, is described. The system is able to detect objects of interest and then take real-time actions to interact with the objects, in this case, by moving to acquire inspection-style images of the object, from multiple angles. The robot system is small, self-contained and runs on battery power. The system shows the potential for the development of robotic systems with context awareness, permitting advanced autonomy.