Proximity Prediction of Mobile Objects to Prevent Contact-Driven Accidents in Co-Robotic Construction

Abstract

Robotic solutions have garnered increased attention from the construction industry as an effective means of improving construction safety and productivity. However, in deploying such robots to real fields many safety concerns have remained untackled, particularly contact-driven accidents that can be potentially escalated by mobile robots. To address this issue, the authors developed a fully automated framework that enables predicting the proximity between mobile objects, leveraging a camera-mounted unmanned aerial vehicle (UAV), computer vision, and deep neural networks, and then conducted a field test to evaluate its validity. In the test, the framework showed a promising result: It achieved an average proximity error of 0.95 m in predicting 5.28 s future proximity between a worker and a truck. The major contribution of this study is in predicting the risk of impending collision in advance, thereby making proactive interventions possible. Computationally, the predictive functionality based on computer vision and deep neural network including convolutional neural network and generative adversarial network would allow robots to examine alternative multiple paths beforehand and enable providing advance alerts to workers. These proactive interventions would effectively reduce the chances of impending collisions between mobile robots and construction workers.