Radar-based Situational Awareness for Industrial Safety Applications

Collaborative robots are intended to operate in close proximity to human co-workers to improve efficiency of industrial process systems. Safeguarding humans from potential accidents caused by collisions with robots or other dangerous machinery requires situational awareness to prevent close encounters. In this paper, we present a sensing and processing platform based on lidar and radar, as well as algorithms to detect and classify target objects in the proximity of the system. Our experimental evaluation of machine learning algorithms based on hand-crafted radar features as well as convolutional neural networks applied to radar range-Doppler signatures indicates that classification into human activities (standing/walking) and robots or machinery can be performed with an accuracy of up to 96%.