Covy: An AI-powered Robot with a Compound Vision System for Detecting Breaches in Social Distancing
This paper introduces a compound vision system that enables robots to localize people up to 15m away using a cheap camera. And, it proposes a robust navigation stack that combines Deep Reinforcement Learning (DRL) and a probabilistic localization method. To test the efficacy of these systems, we prototyped a low-cost mobile robot that we call Covy. Covy can be used for applications such as promoting social distancing during pandemics or estimating the density of a crowd. We evaluated Covy's performance through extensive sets of experiments both in simulated and realistic environments. Our results show that Covy's compound vision algorithm doubles the range of the used depth camera, and its hybrid navigation stack is more robust than a pure DRL-based one.