Machine Movement Prediction for Collision Avoidance

Abstract

Construction sites are among the most dangerous work environments. Worldwide, thousands of workers are injured or killed while working with or around machinery. For Volvo Construction Equipment, safety of the workers is one of the core values. The company strives for the vision of "Zero accidents with Volvo Group Products " through high quality and innovative products that reduce the frequency of accidents as well as their consequences. The purpose of this thesis is to provide a solution towards this vision by developing a collision avoidance system for construction machinery. This is achieved by implementing a model based deterministic threat assessment approach in which the movement predictions of the machine is calculated and evaluated to determine the risk of a collision. The important aspect of generalization has also been considered, in which a new combined machine model has been developed which can be utilized to represent the kinematics of three different types of construction machinery. The results obtained from the combined machine model are compared with true machine models. Kalman filters and their extensions used for state/parameter estimation are investigated. A new method is formulated for predicting the states using Extended Kalman filter which has been proved to be better performing than the usual prediction methods. For collision detection, an algorithm based on the separating axis theorem has been developed. The developed system is investigated and validated using real-world data. The final result obtained from the thesis was an accurate threat assessment system performing for both linear and nonlinear trajectories by utilizing only GPS signals as input and also producing real time collision detection measures.